AI-Native Growth: Scaling Without the Bloat

01:27:16.590 –> 01:27:18.330
Anis Bennaceur: Doing? Great. How are you, Julia?

513
01:27:18.800 –> 01:27:28.879
Julia Nimchinski: Excited to host. You anise to all of you. Watching is a co-founder, and CEO CEO of attention.com comma right.

514
01:27:29.401 –> 01:27:32.529
Anis Bennaceur: Yeah, yeah, we’re we’re growing up.

515
01:27:33.280 –> 01:27:35.749
Julia Nimchinski: Awesome. We have an all-star panel here.

516
01:27:36.100 –> 01:27:40.150
Julia Nimchinski: Welcome to the show, everyone I need.

517
01:27:40.150 –> 01:27:43.500
Anis Bennaceur: Everyone very excited to have this conversation with y’all.

518
01:27:44.320 –> 01:27:45.669
Julia Nimchinski: Cool, take it away.

519
01:27:46.870 –> 01:27:54.199
Anis Bennaceur: Sweet. So to everyone who’s listening. Welcome to the this executive run around table.

520
01:27:54.300 –> 01:28:05.229
Anis Bennaceur: So a lot of people now are speaking about AI native growth. Right? How can you grow your business? Without necessarily scaling your sales? Team? How can you

521
01:28:05.350 –> 01:28:17.039
Anis Bennaceur: have leaner teams way better at execution? And so I’m very happy here to host this conversation with an all-star team of operators.

522
01:28:18.820 –> 01:28:19.465
Anis Bennaceur: So

523
01:28:20.370 –> 01:28:32.160
Anis Bennaceur: we prepared a few questions, you know, or a few themes around around this right? So would love to ask you all as a 1st question, where do you all stand with agent deployment?

524
01:28:35.910 –> 01:28:36.970
trevorrodriguestemplar: Who’s going first? st

525
01:28:40.330 –> 01:28:42.830
trevorrodriguestemplar: Well, I mean, I could take that. So

526
01:28:43.830 –> 01:29:11.969
trevorrodriguestemplar: it really depends on the use case. Right? So I think, what? We’ve seen a lot of success so far is with, you know, customer support. And we’ve seen a profiliration over there. You know, we just had the CEO of Hubspot, you know post today morning about the success she’s seen with Hubspot customer success offering, I think. When it comes to Sdrs, you know, outbound inbound. There’s a lot. There’s a lot of buzz around about 11 AI. And with the other folks, but for the most part

527
01:29:13.540 –> 01:29:24.311
trevorrodriguestemplar: at the Enterprise level. We focus more on enterprise agents, like, you know, around AI account planning revenue operations sales assistance, etc, etc, and

528
01:29:25.264 –> 01:29:42.015
trevorrodriguestemplar: competing mostly with with the likes of copilot. And on Einstein, etc. we’ve seen we’ve seen as long as you you can deliver value and really help. You know the folks along the way bas. Then you’re gonna see usage. But

529
01:29:42.620 –> 01:29:55.689
trevorrodriguestemplar: If it’s if it’s not really easy to use. As you know, most salespeople have add. So if it’s not really easy to use an instant value, then, unfortunately, you know, it’s gonna be a it’s gonna be a long road, right? So that’s what we’re seeing. So.

530
01:29:58.280 –> 01:30:26.919
Omer Gotlieb: I can share my experience. Obviously, we eat our own dog. Food itself. Speak so. What we have is, again, an AI agent that deals with all the inbound actually, instead of me speaking with everybody, I have an AI that can present me and speak with all my website visitors. So that’s that’s for granted. And now I’m not sure about to use the word agent. But just for example, I’ve orchestrated a Multi Channel marketing campaign last week.

531
01:30:27.428 –> 01:30:36.069
Omer Gotlieb: Just me and Claude literally took me 3 or 4 h, something that 6 months ago I would need probably 3 weeks, and and 2 people on that.

532
01:30:36.658 –> 01:30:43.559
Omer Gotlieb: So some of it were agents, some of it of me corresponding with an AI, and we continue to think on.

533
01:30:43.740 –> 01:30:56.289
Omer Gotlieb: how do we not hire the next person? How can we embed AI, whether that’s agent, whether that’s, you know, corresponding with that to really improve all the process that we have, internal and external as well.

534
01:31:01.830 –> 01:31:02.300
Chang Chen: Did you want to

535
01:31:02.710 –> 01:31:15.440
Chang Chen: working with a tech startup? Like most of the company that I work with, they are already deploying? Not one but the multiple agents across coding, operation, marketing and the sales.

536
01:31:15.440 –> 01:31:40.360
Chang Chen: But so so for a while that I was thinking that AI agent is really popular. But I traveled to Las Vegas for a conference last week, and I had more exposure to more traditional industries. And then I realized that we’re still crossing the cat. And so the AI agent is popular.

537
01:31:40.360 –> 01:32:04.500
Chang Chen: not early adopters. But if you start talking to the mass majority of population that we’re still very early, that people know that agents are hard, but many enterprise leaders, that they are still prudent in terms of thinking about, to really rolling out the agent in a mass production capacity.

538
01:32:05.100 –> 01:32:28.119
Chang Chen: It’s also, I’ve heard that many companies that they’re testing definitely, they’re curious. And they’re testing. But but at the same time that still feels Cal. They still feels careful, and they still are in the face of a slowing experimentation, the rolling out. But in the Bay area for the tech startup, but that everyone’s embracing the AI agents. Efficiency.

539
01:32:30.880 –> 01:32:39.169
Anis Bennaceur: I’ll share a pretty interesting stat along the way. So a couple of weeks ago I ran a poll amongst a pretty

540
01:32:39.300 –> 01:32:59.314
Anis Bennaceur: tech oriented go to market audience. And here are the results. Actually, so the question was, how are you thinking about AI agents for your go to market? And actually, only 5% were fully implemented and seen great results, right? Which is really interesting.

541
01:32:59.880 –> 01:33:29.550
Anis Bennaceur: obviously, 30% were currently implementing. And then over 60% had have not even started right. A big part, you know, almost 10% has not even thought about it yet, which is really interesting. But there’s a still massive part of the market. And this was, bear in mind, within a tech oriented audience. So imagine what it’s like within the rest of the world.

542
01:33:30.420 –> 01:33:31.469
Anis Bennaceur: We’ll just.

543
01:33:31.470 –> 01:33:56.420
Sandy Diao: Yeah, I wanted to share a similar observation. So as a growth advisor, I work with a number of companies, both B, 2 B+B, 2 C, and I think one of the shared challenges that I see. And in all of this is, you know, I think the potential is crazy as a lot of our panelists here have expressed with use cases. But I do think the challenge for a lot of what we do in growth is real. And you know, one of the big challenges that I see is that a lot of the

544
01:33:56.420 –> 01:34:11.090
Sandy Diao: context that we have in our work is very cross functional in nature. I think Omar was sharing how he sort of set up this this clawed, multifaceted marketing campaign. But there’s various points in which there are agentic

545
01:34:11.090 –> 01:34:36.060
Sandy Diao: work, and there’s also, you know, sort of more co-pilot work where you know he’s primarily the driver here. And so you know, one of the challenges here is that from from doing growth internally in companies and helping companies with it, I actually generally see that the data is very siloed in nature. So, you know, if you talk to anyone who’s ever worked in growth or marketing, they’ll probably be able to tell you a pretty good story about the 6 or 12 months. It took them to create a single source of

546
01:34:36.060 –> 01:35:00.680
Sandy Diao: truth for even something like user acquisition data, or how it took them a really long time to even have one single dashboard that combined multiple data pipelines. And even then it was still imperfect. Right? So I do think one of the challenges of having a highly functional agent for driving growth within the context of an organization is that a lot of the work that we do is rely on so many different tools and data that lives in different systems. And a lot of these systems are data walled gardens

547
01:35:00.720 –> 01:35:27.959
Sandy Diao: like Meta ads, for example, sort of the basic performance marketing. And some data is easier to access, maybe website traffic data. But then there are campaigns that we hosted in our email marketing crms and braze and otherwise. And there’s not a really easy way to streamline all that context into a single place. So it’s really tough for the autonomous agentic systems to make decisions on our behalf, because they’re not being able to access this data in real time. So

548
01:35:28.140 –> 01:35:49.130
Sandy Diao: I’m sure that that’s a problem that can, and eventually will be solved. But you know, because of that, as of today, I do think some of the best use cases that I’m you know, more excited about immediately roll out for are going to be much more verticalized. Where there’s this data context loop that can drive autonomous decisions by an agent. The great examples that we cover here today include things like outbound sales.

549
01:35:49.130 –> 01:35:58.750
Sandy Diao: probably a great one where there’s a lot of the automation logic, co-pilot nature of work that can become automatic. And we can have clear decision making based on that data context loop.

550
01:35:58.750 –> 01:36:25.959
Sandy Diao: And you know, I think an area that’s going to be tough. For example, for an agent to come in and really kind of take over would be for high performing creative work. Where, again, that data is based on creative performance that’s usually housed in a separate data system. Or this kind of data walled garden. And that data context that doesn’t quite exist yet. So you know, that’s kind of how I’m seeing some of the challenges on us to reflect some of the lack of adoption that we’re seeing with agents, especially in the growth and marketing work that we do.

551
01:36:29.480 –> 01:36:39.420
Mike Haylon: It’s really fascinating, Sandy. I think that’s really consistent with our findings as well. I mean you offered some some stats in these, but you know, we’re seeing that more than 2 thirds of

552
01:36:39.570 –> 01:37:05.299
Mike Haylon: you know. AI transformations are failing. At the you know, right at near the outset. I think, Sandy, you you call on, you know, a few reasons as to why that’s the case. I also think there’s this challenge of you know the the market overhyping what agents are really capable of, and the cost of value that and all the Cios that I’m speaking with across the world are not, you know, seeing that would deliver anything near what the hype suggests

553
01:37:05.430 –> 01:37:16.549
Mike Haylon: that it can, and that, you know, agents are still really highly dependent on kind of a step by step, instructions for how to go, execute a task or perform a thing, and to deliver

554
01:37:16.720 –> 01:37:19.965
Mike Haylon: something that’s not going to be a hallucination back to you.

555
01:37:20.470 –> 01:37:45.320
Mike Haylon: And so I think users are both finding this stuff difficult. The the agents that are easy to adopt are producing confusing results. And so when you, as an It leader, you know, go to deploy this to your organization, and that’s that’s the for users. 1st experience, you know. Change. Transformation is all, always the hardest thing, regardless of whether it’s AI or not, and that will be the case

556
01:37:45.650 –> 01:38:09.706
Mike Haylon: until the dawn of time. And when you this is the 1st experience of the user. All the more reason you’re gonna make it difficult. And so, you know, I think a lot of what we’ve been putting our time and attention to is, you know. How can you, Sandy? I think, in the example that you gave, you know, give it the right context, the right instruction at the right time and enable, you know, users people to be able to take advantage of that.

557
01:38:10.120 –> 01:38:39.759
Mike Haylon: And I think there’s this really interesting diagram I could share with the audience as a follow up to this. But about sort of like, where the Y-axis is reliability and consistency. And then the x-axis is, you know, generalizability and autonomy. And I think what I’m referring to is on the further down you go on the x-axis, the more the greater likelihood of hallucination you’re going to see. The further you go up onto the Y-axis, the more consistency and results. But that really does require. You have, you know, step by step.

558
01:38:39.890 –> 01:38:59.689
Mike Haylon: in our case, you know, rule builder, where there’s structure, you can point to a task or a project. You can get it only pointed only to the relevant information it needs. You can give it clear instructions through the Llm. That’s embedded within that workflow, and you can integrate, you know, human in the loop to execute for certain tasks. You know where the Llm. Is not capable.

559
01:39:00.350 –> 01:39:11.799
Omer Gotlieb: I wanna double down this because, you know, it sounds like those things are not talking and are very difficult, and and they all might be. But I think there’s a huge difference between

560
01:39:12.060 –> 01:39:28.839
Omer Gotlieb: just stand out Llms that could be hallucinating because they’re not fine tuning to what you’re actually doing and specific product and solution that will train exactly to have a lot of guardrails and to minimize the number of hallucination. Now is that going to be 100%. No.

561
01:39:28.990 –> 01:39:30.159
Omer Gotlieb: I don’t think so.

562
01:39:30.200 –> 01:39:52.670
Omer Gotlieb: But is it going to be much better than human force working on this? I think it is. And again, the example I’m using. We got probably 2,000 leads in in the last weekend. We’re not equipped to deal with those number of leads. If I did not have an AI that actually qualifies them and recommends what the team needs to do. I would lose many of them now, did the AI hallucinate.

563
01:39:52.670 –> 01:40:14.740
Omer Gotlieb: maybe, maybe but 90% of them is actually pretty good. So I was able to make a lot of progress in a very short time. Because of that, it wasn’t autonomous. I did not let the AI, you know, respond instead of me. And those kinds of things. I don’t think it’s there yet, but it saves us so much time. And and that’s what I think companies should look at.

564
01:40:17.550 –> 01:40:30.509
Randy Wootton: Mike, I just have one clarifying question. You opened by saying, 2 thirds of AI deployments are failing. What are those deployments of just kind of traditional AI augmentation or agentic. AI

565
01:40:30.640 –> 01:40:32.080
Randy Wootton: offerings.

566
01:40:33.170 –> 01:40:51.389
Mike Haylon: We’re good clarifying question. I think the this is from our think tank that operates with you know, independently of Asana, but is funded by Asana, and then, in conjunction with university professors across the world. Including locally here, you know, at Stanford.

567
01:40:51.520 –> 01:41:15.139
Mike Haylon: And we’re speaking to AI transformation exclusively, not specific deployments of that, the type of AI I’m more pointing to. And I think, Omar, actually what you’re saying resonates with me. And I think we’re actually saying the same thing. Dany was referring to cross functional, you know, deployments and and dependencies cross functionally where you don’t have access to the data.

568
01:41:15.140 –> 01:41:41.950
Mike Haylon: which is what I’m speaking to as well. If you can, in your case, point to a very specific thing, you’re asking it to do much greater likelihood of reliability. There. It’s when you have these sort of cross functional dependencies where data is siloed or you can’t give it the appropriate context or instruction, or integrate a human to do those things where it doesn’t have that ability. That’s where the the failure that I’m speaking to.

569
01:41:43.470 –> 01:42:04.770
Randy Wootton: Yeah, I think the thing you know, we’ve been drawing a distinction between augmented intelligence versus artificial intelligence. And I think with Omar, you know you’re interacting. And you’re building it versus having to go off on its own. I think that part of the challenge with adoption is people, even in that earlier stage of just using the Llms don’t feel confident because of the hallucinations and so releasing control

570
01:42:04.860 –> 01:42:24.049
Randy Wootton: to artificial intelligence because of the experience of current. You know Gen. Search hallucinations makes us uncertain, and makes it less reliable, I do think, to your point, Omar, though, for where it’s pretty repeatable workflow, and you can control the action. So lead calls

571
01:42:24.550 –> 01:42:43.660
Randy Wootton: right, feels like an agent can do that. And you’re not going to suffer from a hallucination, whereas I’m building a strategy, and I need to figure out my new financial forecasting and go out. And, you know, do a predictive model on that, and have an agent that is acting as your outsourced Fp and a that has to ingest data from your general ledger. And all these other things like.

572
01:42:43.660 –> 01:42:54.629
Randy Wootton: maybe it’s just the that’s not a repeatable workflow that’s applying a level of intelligence that I think perhaps a lot of people today still are kind of anxious about. I I can’t trust the black box.

573
01:42:54.630 –> 01:43:17.290
Randy Wootton: Yeah, but for a workflow, absolutely. I I tee it up. And it’s kind of like I got a robot right. I got a robot that’s gonna do the thing that I can put judgment and parameters around and metrics. And and to your point, you know, you can discount leads, and you put a feedback loop back on it. So I think there may be a distinction on those 2 level, augmented versus artificial and level of sophistication of the task being asked.

574
01:43:17.290 –> 01:43:36.900
Omer Gotlieb: I completely agree. And I love those categorization. But I challenge our audience to think about something, because eventually, what all of us are talking about. There’s still a trust issue. It’s trust right? Some people trust it more. We all remember those days where we didn’t trust our data to be in the cloud. Right?

575
01:43:37.090 –> 01:43:41.470
Omer Gotlieb: Are we going to put our financial data, our customer data? It is in the cloud.

576
01:43:41.710 –> 01:44:11.389
Omer Gotlieb: this, these trust issues are going to be solved. Now we can argue whether they’re going to be solved in 3 months or 3 years, and I have no idea but all of us need to plan for a future which I think would be sooner than later, that those things would be solved. So I wouldn’t say I’m not doing anything. I would say exactly what you do, categorize it into the right stages and the right, I would say, operations and projects and start moving quickly. Don’t wait until those trust issues are going to be solved, because then you’re going to be late.

577
01:44:11.770 –> 01:44:16.520
Anis Bennaceur: Yeah. Oh, I was, gonna say, yeah, go ahead. Go ahead.

578
01:44:16.790 –> 01:44:28.649
Anis Bennaceur: Yeah. There’s a really interesting use case here that we spun up in internally, our company, which was basically customers around customer support. And so

579
01:44:28.800 –> 01:44:35.590
Anis Bennaceur: what we did was we we spun up an Mcp server that would look up into our logs.

580
01:44:35.973 –> 01:45:03.220
Anis Bennaceur: Whenever a a customer had a question around these, and so kind of the mental framework that we had was, we cannot, just, you know, put this in the hands of customers immediately. So we made it an internal tool, and whenever a customer would ask a question around. You know their their usage, or or something that is happening. In a specific moment someone in our team would ask that Mcp

581
01:45:03.300 –> 01:45:12.440
Anis Bennaceur: about the the question, and they would get actually an answer. And so surprisingly, it was accurate.

582
01:45:12.550 –> 01:45:27.509
Anis Bennaceur: 90% of the time, right? Very rarely did it hallucinate the despite. You know whatever we we were expecting. And I think with the with this this speed of innovation you know

583
01:45:28.260 –> 01:45:40.480
Anis Bennaceur: of, not only the the Lms, but also all the different protocols that are coming out. I do think that we’re gonna get to a point where we can automate a lot

584
01:45:40.550 –> 01:46:03.474
Anis Bennaceur: faster than we think. And specifically, with the Mcp protocol, we can connect a lot of different tools and get access to data faster than ever. It still kind of breaks every now and then. But it it is surprisingly good. And that kind of raises the question, what what frameworks do you guys have in place today?

585
01:46:04.160 –> 01:46:11.799
Anis Bennaceur: you know where to have humans, even where to have AI agents. How? How are you guys thinking about the human in the loop? Here.

586
01:46:16.800 –> 01:46:19.620
trevorrodriguestemplar: Well, well, think about it right? I mean, the largest model is, I mean.

587
01:46:19.720 –> 01:46:31.511
trevorrodriguestemplar: So we had. We had Randy talk about, you know, differentiating between autonomous versus semi-autonomous, or, you know more of a co-pilot type. We we had. Omar was pretty passionate about his use case with

588
01:46:31.900 –> 01:46:53.843
trevorrodriguestemplar: you know, with qualifying leads and and getting leads in there right? But at the end of the day think about it. The largest models we have today are what? 20 to 30 million parameters you’re talking 3 Byte per, you know. 3 Byte, you know, right per word so, or tokens. So you’re talking 10 to the power of 10, and then 14 zeros right? Approximately if you think about it.

589
01:46:54.500 –> 01:47:12.689
trevorrodriguestemplar: A 4 year old child, in the 1st 4 years of his life takes in about 2 MB per second through the optic nerves, and if you take 1,600 h, which is about 4 years, and you multiply it out surprisingly, you get 10 to the power of 14 again. So in terms of Agi, and

590
01:47:12.750 –> 01:47:32.170
trevorrodriguestemplar: you know the ability to go beyond what I term as a syllogism which is essentially what most large language models are, completing sentences for other people. I don’t think, I think, thinking of large language models by themselves, holistically solving problems or being used as the fulcrum for agents is, you know, is

591
01:47:32.440 –> 01:47:55.540
trevorrodriguestemplar: you’re just not gonna solve the problem. So I think it’s more a question of the ability to marry up Gen. AI along with context, which context is always king in this case, and then bring in large quantitative models from a stochastic and deterministic, and then deciding, is it, you know? Is it Rpa? And are you going to give a fixed set of steps, or are you? Gonna do you know multi-level orchestration?

592
01:47:55.690 –> 01:47:58.530
trevorrodriguestemplar: But the hue in the loop is important, because

593
01:47:58.980 –> 01:48:07.649
trevorrodriguestemplar: if you’re not, if you’re not testing the outputs of your models and not honing them. You’re destined to failure in the long term. Right? I could run the same experiment

594
01:48:07.860 –> 01:48:21.530
trevorrodriguestemplar: a million times, and the 1 millionth and one time it’s probably going to give me a you know, a different answer. And that’s the beauty of what we do. But that’s also, you know, the challenge we face. Right? So from that perspective, I mean.

595
01:48:21.660 –> 01:48:44.329
trevorrodriguestemplar: I you know, I think I think it’s important to clearly define. You know what your output’s going to be. You know what your metrics are, what you’re hoping to really achieve. And but in terms of context, you gotta have a flywheel where you know, you both have domain expertise as well as you’ve got. You know the tools you’re going to interact with. But if you’re going to use Mcp, then you know what what systems you’re going to call, etc, etc. Right so.

596
01:48:44.670 –> 01:48:51.240
trevorrodriguestemplar: But I think I think I’d be expecting Jenny I to suddenly be this White Knight, I mean, it’s not gonna be that right? So

597
01:48:51.650 –> 01:48:53.640
trevorrodriguestemplar: I mean, it’s syllogism.

598
01:48:55.710 –> 01:49:10.280
Omer Gotlieb: I want to give another point of view, and and of course I agree to whatever said, but you know helps me and help the people I speak about decide when you feel comfortable, to actually put AI in front and really have to have the human.

599
01:49:10.850 –> 01:49:13.790
Omer Gotlieb: The best example I give. You know how many times you

600
01:49:14.228 –> 01:49:19.979
Omer Gotlieb: you know, you needed to buy a product. And eventually you ended up speaking with an Sdl

601
01:49:20.110 –> 01:49:44.209
Omer Gotlieb: now obviously ours, but in reality you probably know about their solution and your problem more than they do. You live that world right? I sold to our, you know, head of customer success previously into tango. Initially, they knew more than than I. So I feel more comfortable putting in AI that will replace that function. AI will do a better job than a junior sales person, that

602
01:49:44.390 –> 01:49:49.190
Omer Gotlieb: where I’m not going to put AI, for example, is representing me.

603
01:49:49.520 –> 01:50:12.650
Omer Gotlieb: I have ais that actually, you know, flooding leads to me people linking things like that. I would never have the AI respond with my name, because I have my reputation. I am, you know, I need to take a very careful look at that. And I think we’re years away from that. By the way, Trevor, and exactly what you said about all those kinds of things. So for me, I’m always looking about. Okay, what does it replace?

604
01:50:13.400 –> 01:50:16.770
Omer Gotlieb: If it is a place, a function that is not.

605
01:50:17.340 –> 01:50:31.639
Omer Gotlieb: you know, very educated about what we do. Yes, I feel very comfortable in that. If it’s if it’s supposed to replace something very, very complicated, then then no. Then AI could actually save me time, but not replace it. Usually that’s the framework

606
01:50:31.700 –> 01:50:47.499
Omer Gotlieb: I’m I’m looking at. And and I think Sandy mentioned it, whether it’s co-pilot or agent. That’s exactly those kinds of things I would do an agent all the way when it replaces, let’s say, a low function and a co-pilot when it’s replaces a very complex function.

607
01:50:48.020 –> 01:50:48.360
trevorrodriguestemplar: Yeah.

608
01:50:48.920 –> 01:50:57.299
trevorrodriguestemplar: I mean, I I think at the end of the day. If you just look at Enterprise software right? The one thing. And and Randy’s probably got a good experience, for this is as well as Mike, you know.

609
01:50:57.711 –> 01:51:05.590
trevorrodriguestemplar: See how I had to add Micah. He was with me at Nyc a couple of weeks back. But with enterprise software, you’re limited to seats.

610
01:51:05.770 –> 01:51:28.950
trevorrodriguestemplar: But I I think what most people miss is that with. You know, with with Jenny and with agents. What you’re really going after is those tasks that most companies probably wouldn’t want to hire a human being for right, like looking at a bunch of contracts or writing a couple of lines of code or basically, you know, the road tasks, like qualifying leads as you as you as you alluded, Omar, right? And

611
01:51:29.010 –> 01:51:38.420
trevorrodriguestemplar: if you look at just the Us. We’re talking about anywhere, depending on what data source you look at. You’re looking about 5 trillion dollars in fte spend right? Or salaries, whatever you want to call it.

612
01:51:38.750 –> 01:51:44.220
trevorrodriguestemplar: If if people believe, if organizations believe that you can give them just 5%.

613
01:51:44.676 –> 01:52:09.490
trevorrodriguestemplar: You know that Jenny is worth 5% in terms of what they’re willing to pay. You’re looking at anywhere from 200 million to about 400 billion. So 200 billion to 400 billion dollars in Tamil market cap, and that’s what’s there. So you know, you ignore AI and the ability to use AI at least in enterprise software at your own peril is where I’m going. Because at 400 billion.

614
01:52:09.540 –> 01:52:24.340
trevorrodriguestemplar: I mean, you’re like 4 times the existing market cap, the existing time. That most enterprise software companies are right because you’re not limited anymore by people and by headcount. It’s more about, you know, atomic units that you’re buying, and if demand is elastic, then supply is elastic, right

615
01:52:24.610 –> 01:52:26.250
trevorrodriguestemplar: not to get through esoteric weather.

616
01:52:26.250 –> 01:52:34.269
Randy Wootton: Yeah, you’ve got the best stats, man. I gotta track you down and follow. You got these great points. So I’ve been in

617
01:52:34.270 –> 01:52:58.430
Randy Wootton: horizontal workflow automation for 20 something years right? And one of the things about designing software in that construct is, I think, about the you try to operate, define a process, you try to automate it and then ultimately try to eliminate it. And what I’ve been thinking about with the gentic AI is, there’s this other category beyond that. And I’m calling it elevate like, there’s this whole new world in terms of how to think about a specific process. So I think your point, Trevor, and yours, Omar

618
01:52:58.430 –> 01:53:01.640
Randy Wootton: is. If you’re just at the level of trying to automate

619
01:53:01.860 –> 01:53:27.669
Randy Wootton: workflows, and the human is in the loop like you’re behind the curve. You can already have agents do that for you. I think the the move is. It’s it’s up the level of abstraction, Omar, that you’re pointing to. It’s like when you need human judgment. So the reason you don’t want someone to replace you, Omar, is because you’ve had all this experience, is it informs your judgment. And how do you reply? And it’s hard to train the 4 year old

620
01:53:27.670 –> 01:53:43.939
Randy Wootton: using your data, Trevor, right? Like when they get more sophisticated. And they’re 18 now, maybe you do trust them because they have judgment in a way that’s different. So I think of the human in this condition and what I’m so you know, interested and excited about is, I think, workflow automation tools are absolutely obsolete.

621
01:53:44.270 –> 01:54:11.480
Randy Wootton: absolutely obsolete, and the humans now become orchestrators. They become orchestrators like on top of the loop. They’re not in the loop. They’re over the loop and they’re thinking about things they’re watching. They’re tuning. They’re escalating. So for you. Get that one out of a thousand models that comes out wrong. You’ve got some sort of dashboard, but someone with judgment who’s able to assess the output and understand if it’s an aberration, or is it an improvement

622
01:54:11.710 –> 01:54:26.619
Randy Wootton: based on what happened? My big fear is, Omar, you know you and I look like we’re kind of on the other side in terms of we’ve been around for a while is, how do you take someone out of college and teach them judgment to be an orchestrator of agents

623
01:54:26.890 –> 01:54:38.469
Randy Wootton: like what we’ve used. Workflow automation for in the past is a breeding ground. You take that support agent. You have them do a couple of calls. You give them training and then over 10 or 15 years, it becomes sophisticated. Cust customer support

624
01:54:38.470 –> 01:55:01.200
Randy Wootton: director. Now, you’re gonna miss this whole middle, like mentorship of where people are developing judgment in whatever activity or function they’re executing sales leader, success leader, marketing leader. If we’re reliant on a generation of of agents that are doing that for us. So I you know the thing I’m sort of struggling with more broadly, and I’m glad I’m at the end of my career is kind of like.

625
01:55:01.200 –> 01:55:07.220
Randy Wootton: what’s my son going to do coming out of college in a world where you don’t have the opportunity to be mentored, and learn.

626
01:55:07.830 –> 01:55:23.080
trevorrodriguestemplar: Your son’s gonna join Honeywell or your son’s gonna join Lenovo, or your son’s gonna join Cisco or your son’s gonna go to HP. Or your son’s gonna go to Intel. And what he’s gonna end up doing is getting a digital twin, and that digital twin is gonna be someone

627
01:55:23.160 –> 01:55:46.240
trevorrodriguestemplar: who represents the best of the IP that they have. Remember what I said. Content, you know, context is king right? But domain expertise is king right, and you alluded to it earlier, you know, about vertical knowledge, right? But the problem, you know, if you talk to Vimal Kapoor at Honeywell, or you talk to even Jensen Wang at, you know, at Nvidia, or you talk to. You know you talk to Intel or any of the guys, right? But the problem is.

628
01:55:46.240 –> 01:56:10.989
trevorrodriguestemplar: they don’t have that IP readily available in terms of skill set. So what they really want is to capture all that IP that they’ve had the last 1520, 30 years and then share that information in the form of digital twins. So when your son joins them, he gets the expertise and the ability to ask questions out of the gate. On what would Randy do in this scenario? What would Omar do in this scenario, or what would Julia do in this scenario, and he gets the benefit of all the years of experience.

629
01:56:11.220 –> 01:56:35.479
trevorrodriguestemplar: The thing is as long as you’re focusing on text, you’re fine when you bring in video. And you start talking about predicting where the pixels are going to land like I have a pencil. And I say, is it going to drop here or here. That’s where we’re nowhere over there but replicating people using, you know, going and looking all their videos, their text and getting opinions. I mean, for example, a year ago, I mean, I always tell people I’m jealous of Alexander the Great

630
01:56:35.480 –> 01:57:01.820
trevorrodriguestemplar: because he had Aristotle teaching him for 14 years, and unfortunately I’m a Tad, young right, I mean, I mean, but I’m a huge fan of Aristotle, and I’ve read it in Greek right? But what we did the 1st thing we did was, and partly because of me was, we created a digital twin of Aristotle on rhetoric, because I think rhetoric’s been lost right in our education. Right? So it’s not that difficult as long as it’s around text and finding the next best sentence because language is pretty predictable. Now, nonverb non adverb, right.

631
01:57:01.870 –> 01:57:22.090
trevorrodriguestemplar: The problem comes in. When you look at large quantitative models, you’re talking about stochastic with deterministic. And you’re trying to predict what the probability is of a certain event happening which you were alluding to earlier. Randy, that’s where Gen. AI is is a nice way to synthesize data and talk to your data. But it’s not going to do that for you right. At least, I haven’t figured it out yet. So maybe you guys have.

632
01:57:24.380 –> 01:57:45.990
Sandy Diao: Yeah, Randy, I also wanted to double down on your point. There really liked your heuristic of you know whether or not we’re applying human judgment to, you know, kind of going back to the question here, where do we determine how to use humans versus agents? I also wanted to share a framework that I personally use when I’m working with b 2 b companies to determine this question, both for my own work as well as the work that I’m advising the teams on, and it’s.

633
01:57:45.990 –> 01:58:09.579
Sandy Diao: you know, kind of at a high level. Simply speaking, you know, we use agents. I use agents for work that requires low creative complexity. And we generally lean on humans for work that requires high creative complexity. Just to kind of give an example here, you know, if a task is repetitive as Omar and Randy and many of you have pointed to, and it follows a clear pattern, and it doesn’t necessarily require novel thinking. Generally speaking, it’s kind of a good candidate

634
01:58:09.580 –> 01:58:18.530
Sandy Diao: for us to consider whether or not it’s a good fit for agents to run it, and here are some of those use cases or tasks include things like pulling reports, summarizing support tickets, enriching leads

635
01:58:18.530 –> 01:58:48.209
Sandy Diao: so on and so forth. But if the task does have high creative complexity, or it requires some level of emotional intelligence or original creativity. A lot of the work that I do, for example, includes things like crafting a go to market narrative, or spinning up a new set of creatives that reflects new brand positioning or handling, perhaps a tricky client escalation, or even setting strategy for the next half of the year. That’s where I really find that human plus augmented by AI is going to be a lot more powerful. So

636
01:58:48.410 –> 01:59:04.719
Sandy Diao: in that situation myself or the team would end up using an AI tool or copilot that’s primarily orchestrated and or operated by the human. That just gives us more control on that creative output, because we know that the outcome or the quality has so much more variance between what is good and what is bad.

637
01:59:04.720 –> 01:59:29.069
Sandy Diao: And you know I do think one of the ways to overcome this overall agent adoption barrier that we’ve been talking about here is to really try to figure out how we can apply this framework and figure out which parts of the task that we need to complete have that low versus high creative complexity. I’ll give one use case that I recently utilized AI agents for actually hiring right like hiring actually overall seems like something with really high

638
01:59:29.070 –> 01:59:40.739
Sandy Diao: create complexity. However, when you actually break down the different components of it, there are parts of that workflow that can actually be very useful to apply AI agents to, for example, screening resumes right

639
01:59:40.740 –> 02:00:05.549
Sandy Diao: use a combination of AI prompts as well as tools to help identify and pinpoint who we should be moving forward in the screening process. And then in that process, recently, when I made a hire, I also use an AI voice interview agent called mirror work to autocomplete the screenings and summarize the results for me. So that basically in my inbox I see exactly who I should have a human connection with and complete final round interviews. For, because during that process, as Omar was pointing out to.

640
02:00:05.550 –> 02:00:18.930
Sandy Diao: it’s kind of representing me. It’s representing, you know, my brand. It’s representing kind of that sales conversation, too, like you want to work with me. Not just you know. The the AI screener here, so you know, completed the final round interviews, hired someone within less than a week.

641
02:00:18.930 –> 02:00:40.429
Sandy Diao: and like that’s something that I never imagined would have been able to happen without those agentic powers. So you know, one of the frameworks that I use, you know, very simply, perhaps oversimplified. We’re really thinking about, you know, what is the level of creative complexity involved in the type of work. And how can agents and or a combination of human plus you know, co-pilots and or AI augmented tools help me complete this task.

642
02:00:40.810 –> 02:00:45.750
Omer Gotlieb: Send me. Let me push this a bit. Push the envelope a bit on this specific use case because I love it.

643
02:00:45.820 –> 02:01:10.970
Omer Gotlieb: And that’s example of a business processes that AI can enhance and augment and help. And you spoke about one part. But I’m going to ask you 2 questions. I don’t know what the answer is. But one. Do you think that the way you’re actually interviewing people? Is it good enough? And 2 are your decisions of hiring the right people. Are they good enough? And what I’m saying is, if we apply AI to this.

644
02:01:10.970 –> 02:01:33.679
Omer Gotlieb: AI could actually find patterns, it could tell you. You know what your interview was amazing, but you probably should have asked those kinds of questions. Now, is that going to be right or wrong? I don’t know. But that’s a good brainstorming, I would say, colleague, to actually see if you can improve the way you interview the same way. You know you could tell AI. I’ve interviewed those 5 people. These are interviews, and I’ve decided to pick this one

645
02:01:33.810 –> 02:01:37.819
Omer Gotlieb: and that these are the reasons AI could actually reflect and say, you know what

646
02:01:37.900 –> 02:01:50.779
Omer Gotlieb: this is, what I think about it again, it could be completely wrong, but in my experience it will help you gain more insights and thus improve the process as well. So it’s not just automating. It’s also about

647
02:01:50.840 –> 02:02:11.670
Omer Gotlieb: really, I would say, a brainstorming PAL for me, and I’m trying to combine it in as many aspects as we can. Sometimes the results are ridiculous, but you know, we’re experienced enough to say, that’s fine. That’s that’s not relevant. But sometimes oh, I didn’t think about those things, and that’s what really amazes me, and I think it will improve and improve.

648
02:02:12.080 –> 02:02:36.860
Sandy Diao: I love it. In fact, I think you’re spot on. In fact, I haven’t applied the AI to my portion of the interview right? And so I think one of the things that’s missing there is actually firmly believe that the screener, the AI, you know, interview agent, does a better job than me, and identifying the signals that a candidate is going to need to qualify for the round. You know, for example, I put in about 8 criteria, you know, different skills, like communication, collaboration.

649
02:02:37.060 –> 02:03:00.099
Sandy Diao: etc, etc. And I believe that in my conversations I’m actively looking for all 8 of those signals. But I think the reality is that my mind share what I’m actually processing is probably a maximum of one at a time, and realistically in a 45 min, even 1 h interview conversation. I’m probably not getting as much clarity on the signals as possible, so I think that. And then, also making sure that the way that I

650
02:03:00.100 –> 02:03:17.549
Sandy Diao: run those interviews I’m actually looking for the exact same signals. I think that I was originally screening, for I don’t have a feedback loop there, and that’s a great example of how you know I can further use, you know. Agentic powers to, you know, provide feedback to the interviewer as well to say, did the interviewer actually hold up their part in completing this process? So I love that push.

651
02:03:18.940 –> 02:03:42.890
Chang Chen: Yeah, we one of the company that we are working with, that they started to use their agents to to to do the interview to, to at least do the to do the 1st round of interviews, but one- one of the things that we, because we all we also review all the recordings and the different to review all the transcript as well. And the one scenario, we actually observe that one- one of the candidate is also using a agent. So

652
02:03:42.890 –> 02:03:53.120
Chang Chen: so we are in the situation that we have AI agent, the candidates that’s actually talking to AI Interviewer candidate, which is a very interesting scenario. So I’m really curious to see how this goes.

653
02:03:53.770 –> 02:03:58.103
trevorrodriguestemplar: Yeah, I’m surprised they didn’t go into their own language if one AI agent was talking to the other. But

654
02:03:58.430 –> 02:03:59.990
trevorrodriguestemplar: we’ve seen that happen so.

655
02:03:59.990 –> 02:04:03.239
Chang Chen: Yeah, just are you talking to doing the ones exactly?

656
02:04:03.240 –> 02:04:04.110
trevorrodriguestemplar: Okay.

657
02:04:05.390 –> 02:04:12.180
trevorrodriguestemplar: But you know, we’re all talking about Asians and and Gen. AI and AI, like it’s, you know, it’s it’s new. And

658
02:04:12.633 –> 02:04:25.070
trevorrodriguestemplar: I mean machine learning deep learning. Mlp. Has been around for 9, 10 years. Maybe it was a black box, deep learning. You couldn’t really figure out what the answer was, but the answer looked great. But tell me, any one of you right, I mean, would you trust

659
02:04:25.360 –> 02:04:33.860
trevorrodriguestemplar: if you were running a sales cycle and you had to bet your life on a number. Would you trust what your rep or Rv. Told you, or would you

660
02:04:33.970 –> 02:04:41.510
trevorrodriguestemplar: would you trust? What what a machine learning or deep learning model would tell you, or using Nlp subtext of conversation. Stuff like that.

661
02:04:41.860 –> 02:04:46.129
trevorrodriguestemplar: Would you trust the sales rep. And the Rvp. Or would you trust AI going back and forth.

662
02:04:46.130 –> 02:04:46.980
Anis Bennaceur: To hear.

663
02:04:47.120 –> 02:04:57.179
Anis Bennaceur: Yeah, I think I think we’re getting more and more in a world where people actually trust the AI right? Because it captures all the data points, and it will be a lot more objective, especially if you

664
02:04:57.585 –> 02:05:18.409
Anis Bennaceur: set up the the right guardrails around your own framework or process, right? One of the things I was gonna add is in a really good way to think about it, I think, is what is the cost of failure right? If the cost of failure is minimal, then you can go ahead and put in an AI agent. And so even, you know, to to

665
02:05:19.170 –> 02:05:34.769
Anis Bennaceur: to to do forecasting, for instance, I think, hey, if the every single model is probably going to be off by a few percentage points. So I think there you can be very secure about having an AI forecast.

666
02:05:35.330 –> 02:05:45.340
Anis Bennaceur: But if you’re actually having a self-driving car, you know, the cost of failure is gigantic, right? So that’s a complete, different way to approach things. So

667
02:05:46.170 –> 02:05:48.439
Anis Bennaceur: it’s definitely kind of like one of the things. Yeah.

668
02:05:48.440 –> 02:05:49.090
Randy Wootton: But I think.

669
02:05:49.090 –> 02:05:49.720
Mike Haylon: We actually.

670
02:05:49.720 –> 02:05:59.239
Randy Wootton: Point the the point you’re making, which interesting because we were using at Max here where I was last, we we were one of the 1st companies use gong forecasting.

671
02:05:59.450 –> 02:06:20.189
Randy Wootton: And but this falls under that augmented intelligence paradigm, and it was awesome. It was awesome. It ingested all this data historically it. You know, it measured all the interactions you’re having. It had all this predictability. It helped, you understand where things we had our sales methodology in it, but it still required the sales manager and the sales leader to check it.

672
02:06:20.440 –> 02:06:22.480
Randy Wootton: and so to your point. Anise like.

673
02:06:22.480 –> 02:06:22.980
Randy Wootton: Thank you, Russell.

674
02:06:22.980 –> 02:06:35.010
Randy Wootton: running my business on a agent, a selling agent that was gonna go out and sell and have its embedded forecast and tell me what was going to hit. I mean the cost of failure is really high in that context I’d be

675
02:06:35.010 –> 02:06:55.839
Randy Wootton: anxious. But the power of the forecasting versus what we used to I mean, I’ve been in sales forever, I mean, looks like Trevor. You’re at Salesforce when we when I was there. It just, you know, this monster process of going through like augmented intelligence with AI is gonna make us all better. But I just I I don’t know on a gentic like. When do you turn it over to have

676
02:06:56.120 –> 02:07:01.422
Randy Wootton: up agent, leader sales leader. That’s gonna run your your go to market.

677
02:07:01.830 –> 02:07:15.835
Mike Haylon: I think these are, we actually built this this very example. So you know, I stepped into this new role I had. You know, Ed Trevor. He gave me 7 days to get not just a waterfall, but a bottoms up forecast

678
02:07:16.140 –> 02:07:16.580
trevorrodriguestemplar: There!

679
02:07:16.580 –> 02:07:38.095
Mike Haylon: With too many deals to be able to count and get insight on. So we used the product, you know, for the go to market, of which I lead to establish a workflow sprawling data, and I think it probably produced some stuff of similar to what you came to experience, Randy from gong. But it was essentially taking all transcripts and email back and forth, and notes and account plans

680
02:07:38.520 –> 02:08:05.916
Mike Haylon: from within. Asana and I gave it a prompt. That was my indication of what we use today to call a deal in or out, and then I would have I would cross reference that against my own interpretation of this, just to determine, you know what was the quality of what was being called there fast forward to the end of the quarter. I think you know what I was getting was really valuable insights, and it’s extremely short amount of time as to where these deals stood without having to bother the sales people for it.

681
02:08:06.310 –> 02:08:29.490
Mike Haylon: But it turned out I had a relatively cynical take, you know, upfront as to the likelihood that these or the prompt that I was using that. So the likelihood these deals were gonna come in and we we achieved better results than were immediately forecasted. Obviously, that’s a great general outcome. But I think still, today, as far as we can see, we require both and and I think this is why, you know you mentioned

682
02:08:29.490 –> 02:08:56.345
Mike Haylon: with the elimination of workflows, too. I I think it’s more just the elevation of the how those workflows, what they enable us to execute. You know your your son, or you know the next generation, I think, has an opportunity, and I think this is what Sandy was hitting on, too, is, you know, to to step in and and determine. Where is it that you know AI can still be really powerful? The agents can be deployed to perform these tasks. How would that continue into evolve over time? And where do humans still need to?

683
02:08:56.620 –> 02:09:20.269
Mike Haylon: you know, step in and execute certain functions that are higher reasoning. And you know, collaboration has been a mess for a long time. As we get more and more into this world, and we are automating a lot of the task. I think we’ll start to face new problems about. You know those things that have made it most difficult for us to transform for us to adopt these new ways of thinking and to collaborate and to have data. You mentioned the Mcp servers earlier. I agree there’s like.

684
02:09:20.270 –> 02:09:33.669
Mike Haylon: you know, we should be spinning these up. Every tech company should be spinning these up, and would approve a lot of our ability to get access to the data that we need in order to perform the function, the actions that we’re asking for these to to take already today. But I think workflows.

685
02:09:33.680 –> 02:10:01.089
Mike Haylon: And this is sort of like what Reid Hoffman’s book about the new AI world starts to get at is that you know we have these headlines, and we’ve seen it with the Industrial Revolution all the way through to today that robots are going to take over. They’re going to eliminate everything. I really see it still as an elevation that humans will need to evolve those that take initiative and start to learn. You know the things that will be required of them, as our work gets elevated, will benefit those that don’t, I think.

686
02:10:01.140 –> 02:10:05.800
Mike Haylon: may may ultimately fear that their their jobs will, you know, become redundant.

687
02:10:06.890 –> 02:10:19.409
Chang Chen: Yeah. We also started to observe that the agent not only to help us to take on some of the repetitive tasks is also helps some of our entry level employees, some of the younger generation, to accelerate the learning like

688
02:10:20.070 –> 02:10:45.019
Chang Chen: they are they so so so now that we actually have a fewer teammates, but that they are more moderators of the agents, but also we use agent to help us to read faster, and then to learn more knowledge to summarize all the summarize, all the summarize, all the experience that they gained. So so we do observe that many of our younger generation, that you know our team, that they have gained more than 10 x. Or more experience in the

689
02:10:45.020 –> 02:10:52.359
Chang Chen: last whole year, just by moderating and observing that the multiple agents, that kind of reporting to them.

690
02:10:52.360 –> 02:11:03.430
Chang Chen: So so I don’t think it is going to replace us, but it will definitely help us to all evolve that help us to accelerating and help us to become a better ourselves.

02:11:08.260 –> 02:11:31.900
Anis Bennaceur: Jumping in here since, you know. We we talked about sales specifically right? How are you guys thinking or envisioning a world with sales, teams with fewer Aes and Sdrs hitting higher revenue targets right? There’s a ton to be done here. How are you guys thinking about it. How? How are you thinking about transitioning to this world.

692
02:11:34.730 –> 02:11:36.359
trevorrodriguestemplar: I think it’s more about, you know.

693
02:11:36.840 –> 02:11:39.399
trevorrodriguestemplar: And again, I think it’s more about

694
02:11:40.270 –> 02:11:45.699
trevorrodriguestemplar: being AI, whereas it’s not so much about replacing human beings as much as being able to leverage AI to do your job better.

695
02:11:45.740 –> 02:12:13.500
trevorrodriguestemplar: So I think you know, with the information and with the tool sets and you were going. You were talking. Mike and Randy were talking about gong, and then you talked about. But even over there, right if you’re looking purely at calls, emails, etc, etc. You’re still going to incur. You know the issues we have in terms of hallucinations or bias, what we call bias, which is a big deal in models, because you’re looking at text. But if you bring in large quantitative models and you bring in what we call Lqms.

696
02:12:13.500 –> 02:12:37.059
trevorrodriguestemplar: And you start doing regression analysis. Or you start doing Markov. Then all of a sudden, you tend to account for that bias. And you can get 98% results, etc, etc. But I think at the end of the day I mean, I run a billion dollar business at salesforce, and it was brute force. I made everyone do Qbrs every week. They probably hated my guts. I told everyone it was 3 X. Their quota

697
02:12:37.060 –> 02:13:04.490
trevorrodriguestemplar: was the standard. It wasn’t 1 x Ed remembers that fondly. So that’s how you perform right? But today, you know, when we work with Lenovo, we work with Cisco, some of the large companies in the world, and we regularly give them 90% top down and about 90% bottoms up calling the forecast without bias and at billions of dollars. And the beauty there is. It’s more about making everyone get out there in a superpowers. Right?

698
02:13:04.490 –> 02:13:27.349
trevorrodriguestemplar: Your inner Steve Jobs or your inner. Whoever’s your hero right, whether it’s Elon Musk, or whoever you know your hero is but essentially giving them superpowers because you’re giving them information they never had before at their fingertips. And it’s humanly not possible. Our Corpus is not large enough for us to be able to get the types of AI insights you can get from data today.

699
02:13:27.400 –> 02:13:42.760
trevorrodriguestemplar: Talk to it on the fly and get responses in real time. That help you put more money in your pocket versus versus you focusing on doing all that research, you know, account research. And, you know, calls, etc, etc. Right? So I think, where that’s concerned, you’re gonna see greater productivity.

700
02:13:42.890 –> 02:14:09.549
trevorrodriguestemplar: And it’s not so much replacing people. I just think the per capita, the Gdp is going to keep going up, and we’re going to see better knowledge workers, and instead of spending time on road tasks like, I don’t want to bring, you know iphones back to the Us. And and build them right. I still want to be able to. I still want to be able to do the stuff that’s cool, like building AI models and doing knowledge work right? And that’s what’s going to happen to sales people right? Instead of the road tasks. You’re going to see them more productive. That’s what I look.

701
02:14:09.550 –> 02:14:27.719
Mike Haylon: Totally agree. I think that’s a great, really salient point, like the reality. We’re talking about a world where, not we. The panel, I think, generally agrees this isn’t about replacement, but about augmentation of knowledge workers. I think we you hear a lot of the talk elsewhere that that it’s about. You know, the potential for replacement. And

702
02:14:27.740 –> 02:14:42.069
Mike Haylon: you know, the reality is 67% of the 60 to 70% of the work they’re still doing today is extremely administrative knowledge workers generally and salespeople specifically, you know, in the example that I gave. Now, I can empower a salesperson rather than having to take. And input all these notes

703
02:14:42.120 –> 02:15:01.519
Mike Haylon: maybe spend more time looking at that 13 min snippet of how the customer engaged with us, you know, on that particular topic, what was their language, their mannerisms? What did they say or not say? Who did speak up and didn’t? How can you engage them in a way that could help reframe. You know their thinking. We’re not putting nearly enough attention towards these things now, that’s

704
02:15:01.520 –> 02:15:15.599
Mike Haylon: you know, the forecast. Yes, ultimately, is it a question of who’s going to be better forecaster. I do think it’s a collaborative effort, because now I can deploy the salesperson to go out and better engage those people on a personal level rather than spending their time, you know, behind their laptop inputting all these notes

705
02:15:15.600 –> 02:15:18.499
Mike Haylon: that takes so much of that their time today.

706
02:15:18.860 –> 02:15:19.210
trevorrodriguestemplar: Yeah.

707
02:15:19.210 –> 02:15:39.040
Randy Wootton: I think a couple of things are true, you know, in the team sales teams I’ve worked with is they gotta make money. So the quota gets set so that a certain percentage of the sellers make money. And and my rule of thumb is 80% of your sellers need to be making 80% of the quota. If they’re doing that, then you can start to lever up like another truth in sales. This quota always goes up

708
02:15:39.160 –> 02:15:59.019
Randy Wootton: one of the metrics we’ve used in the past is like, it’s a 4 x multiple. You pay someone $200,000. You put a quota at. You know, $800,000. I think with augmentation you could increase that. And maybe it’s a 6 x multiple on top of what they’re making because you’re augmenting them, and they’re able to do more. But at the end of the day they’re still able to hit their number.

709
02:15:59.210 –> 02:16:23.810
Randy Wootton: 3rd truth of sales is, you’re going to have people who take advantage of this stuff, and they’re gonna make a crap load of money. And then you’re gonna have the the people on the tail end that want to do it the way they did it. And they’re gonna get fired. And in sales in particular, I think you have a higher attrition rate than broadly across the organization. Let’s say your sweet spot for attrition. If you’re running a company is under 15%, 10 to 15% right turnover every year. Your sales team is maybe north of that 15 to 20%.

710
02:16:23.810 –> 02:16:38.769
Randy Wootton: And I think what you’ll find is the people that are out in front that are making the most money that are taking. The bigger quota. Hitting President’s club are going to be the ones adopting these tools and augmentation. They’re gonna figure it out right. They’re the CEO of their own territories. And then the rest of the people gonna say, I want to do what they’re doing.

711
02:16:39.139 –> 02:17:03.499
Randy Wootton: I want to make money. And I want to be effective and efficient. And I’m okay with taking more quota, because I know I can. I can deliver, and they get the better accounts. It’s like the classic sales growth strategy. You start off in early stage, you know, hard transactional sales, and you move up to enterprise and you do relationship management. But you got bigger quotas as you go up. So I do think sales is such a great function because you can look at it, and you can measure success. You can measure.

712
02:17:03.500 –> 02:17:23.580
Randy Wootton: And they’re always the best sellers, the most creative. You got. Add. They’re trying everything. My seller, my top seller. Last year Maxio made 2 x more than everyone else that guy was in the office every single day he was doing Linkedin posts. He was using AI to augment everything he was doing. And the other guys that weren’t hitting quota. I was like, why aren’t you doing what they’re doing. Well, I want to play golf.

713
02:17:23.580 –> 02:17:37.180
Randy Wootton: you know, or I wanted to kick off on Friday at 4 and drink a beer. So I think there is an absolute difference in orientation, and there’s a system that will allow it to expand, to drive more efficiency by giving more quota. But the Aes are still gonna make their money.

714
02:17:38.299 –> 02:17:42.994
Omer Gotlieb: I wanna be a bit provocative and suggest that we might be looking at

715
02:17:43.939 –> 02:17:51.129
Omer Gotlieb: at the problem differently, or should look at differently, because up till now the conversation was inward. How will? How will the sales team gonna look like?

716
02:17:51.179 –> 02:18:15.769
Omer Gotlieb: And what is gonna be? And I actually think that the entire sales team or entire sales motion might change, because we need to look at the buying experience and think of yourself. The 1st time I think I’ve mentioned that last time I’ve met you, Randy. Think of the 1st time you’ve encountered or you chatted with Chat Gpt! I was amazed like, Oh, that is amazing. And you know, we we heard some use cases today that all of us are amazed. From what AI can actually do.

717
02:18:16.149 –> 02:18:21.409
Omer Gotlieb: you actually think that your customers are going to continue to agree

718
02:18:21.599 –> 02:18:29.439
Omer Gotlieb: to buy in the same way that you’re selling today that they’re going to start with forms on the website. And then Sdrs, and then a salesperson doesn’t know, and then

719
02:18:29.439 –> 02:18:52.149
Omer Gotlieb: everything has to change. I don’t know what’s going to change, but everything has to change. So for me. The focus is really 1st trying to vision what’s going to be in the future, how a buyer is going to buy and then actually decide how is my sales team going to look like? I think sales is an amazing and crucial functions. We have to have sales. We have to have people there. Nobody’s going to buy a million dollar solution without a person.

720
02:18:52.259 –> 02:19:11.169
Omer Gotlieb: But what is the soul of this person? How will we motivated? How will we measure that I don’t know but it’s going to start with how the buyer wants to buy, and I am 100% sure that this is going to be different than how buyers are buying today because of that experience, because everybody is exposed to those things.

721
02:19:11.420 –> 02:19:22.649
Mike Haylon: I think that’s right. And you know you you mentioned earlier, how silly would it be not to just adopt this because you’re gonna miss out. And I think the opportunity here is actually.

722
02:19:23.168 –> 02:19:36.079
Mike Haylon: maybe put slightly differently, but I think is striking. The same chord is as a as a vendor. How can I? What are the ways that I can attach myself to enabling their AI transformation?

723
02:19:36.389 –> 02:19:57.410
Mike Haylon: And if if their association then with my brand becomes that I enabled their ability to go through this transformation much more efficiently, rapidly, you know, with ease than they otherwise would have. There is the opportunity. And so I think that’s sort of what you’re saying is, I understand how buyers are going to buy, you know, map to that likelihood, and I think

724
02:19:57.410 –> 02:20:26.819
Mike Haylon: for us at least, that’s acknowledging that you know, most of these are failing today. It’s a big learning curve, and people don’t want to change. And so how can we lean into that reality and the reality of what is available via AI today and start to carve out that path, make it feel more approachable, more accessible, to solve the problem that they need to and give them sort of a journey map for how to do that. And then the association with our brand, or that the relevant brand becomes.

725
02:20:26.980 –> 02:20:35.269
Mike Haylon: wow! Look, you know, look what they’ve enabled us to do, my department to do my company at large to achieve as a result.

726
02:20:36.050 –> 02:21:03.040
Chang Chen: Yeah, discussion on like, so so whenever we are thinking about to buy a solution, that we will also face the question like, should we buy, or should we distribute internally? But now, with all the AI developing tool, that building internally is actually getting a lot cheaper. So we’re also talking to several enterprise buyers that they are, they are proactively thinking about that. They’re buying decision, how they, a actually allocate their budget

727
02:21:03.040 –> 02:21:20.089
Chang Chen: in terms of buy versus actually built internally. So so that that whole buying experience, and whether or not they’re actually going to buy a solution. I think that that’s also something that I’m very curious to alter. The trends.

728
02:21:21.700 –> 02:21:45.100
Sandy Diao: Yeah, I wanted to add a bit more to what Omar and Mike were saying around the changes in how people are businesses are buying, and how people are changing, how they buy more broadly. One of the things I observe in working with b 2 b companies, that more and more b 2 b companies are excited about building. We call what we call product, like growth, right? And it’s kind of this behavior in which we’re a lot more excited, or we expect to be able to try before we buy.

729
02:21:45.100 –> 02:22:03.670
Sandy Diao: And that kind of changes, how traditional sales led or enterprise led models are thinking about offering their products and services to their customers. And in particular, one of the areas in which I’m really excited and bullish on how agentic AI can transform a product like growth models is specifically with onboarding.

730
02:22:03.670 –> 02:22:18.400
Sandy Diao: I do think that onboarding is actually one of the most important and biggest services that companies and products will use to qualify and build intent for users. You know. And if you look at most speedb products out there, you know a lot of them that are product led. They’re stuck in these product. Led flows

731
02:22:18.400 –> 02:22:35.610
Sandy Diao: right? They have these sort of like click through guides like next step. Here are the 1st 3 things to do. They have these tool tips. They have these onboarding questionnaires, and then you have an entire product or marketing team figuring out what are the 5 questions we ask, and then figure out what are the 6 multiple Max 6. Multiple choice. You know, categories that we can offer people.

732
02:22:35.610 –> 02:22:50.290
Sandy Diao: Whereas nowadays. This is something I think you know, a lot of b 2 b companies can learn from consumer apps. Actually, a lot of apps like companion apps or habit trackers, or even language learning apps that onboarding is very much driven by voice agents, or these chat like experiences where you open the app.

733
02:22:50.560 –> 02:23:11.979
Sandy Diao: somebody guides you through this very branded experience. You get all the information that you ever wanted for onboarding. And even more than that, and you get far more context around what the user’s goals are. And I think that more b 2 b companies can actually benefit from using these these transcripts, these conversations to extract even more intent. And I think that creates this pipeline for product like growth in a way that

734
02:23:11.980 –> 02:23:23.519
Sandy Diao: was not necessarily possible before right? Because when people start talking. You know, people aren’t very concise. I’m not concise, and we just start to give away more information about our needs and what we want than we expect to. So I think.

735
02:23:23.980 –> 02:23:40.459
Sandy Diao: as a result of that, these experiences that b 2 b products can provide are going to be far more personalized. Those sales calls can be more personalized, whether the sales call is conducted by an AI agent and or a human, a human being, and there are so many tools out there, like voice flow, for example, that let you build these off the shelf.

736
02:23:40.470 –> 02:23:57.609
Sandy Diao: Chat voice AI agents for experiences like this. So you know, really from a growth perspective. I’m really excited and bullish about agents radically improving the quality of personalization which you know in turn improves user activation expansion, potential customer, lifetime value, etc.

737
02:23:58.770 –> 02:24:02.520
Randy Wootton: Yeah, I just offer one other acronym agent led growth.

738
02:24:03.693 –> 02:24:30.649
Randy Wootton: I think, to Omar’s Point. And you’re spot on Omar. You’re taking us from an augment augmented experience to an artificial experience based on the revolution in the buyer’s journey. Buyers using agents companies are gonna have to figure out how to respond to that with agents so agent, agent interaction, and then, Sandy, to your point, the way that you get up and running the the way we think of that paradigm is gonna radically shift as well. So I think it’s beyond product led. I think there’s a whole new

739
02:24:31.268 –> 02:24:40.850
Randy Wootton: you know, wave that we’re going to be anticipating along marketing sales and service across the whole. Go to market. It’s not just revolutionizing sales. It’s the whole. Go to market.

740
02:24:47.790 –> 02:24:49.302
Anis Bennaceur: I’ll jump in here.

741
02:24:49.730 –> 02:25:03.260
Anis Bennaceur: yeah, that that is very spot on right. I we we’ve literally been rethinking about how to redo our entire onboarding. And and it’s not just through voice a voice led motion. There’s even more than that, right

742
02:25:03.280 –> 02:25:27.050
Anis Bennaceur: as a product you can already read. Let’s say, you know, through Mcp again bringing it up. But you can definitely read through different products that you integrate with. How? Let’s say your Crm, right? How is your data structure in the Crm, what are you typically filling? What are you? And then automatically infer from that

743
02:25:27.280 –> 02:25:28.937
Anis Bennaceur: how you can

744
02:25:29.610 –> 02:25:57.400
Anis Bennaceur: customize the onboarding of your own product based on these integrations. Right? And so, instead of having a user even go in and manually customize their own settings, you can actually infer a lot of these settings leveraging AI. And so a very, very exciting application of AI agents is through onboarding. I think we’re Juliet. Tell me if I’m wrong, but I think we’re getting up on time right.

745
02:25:57.650 –> 02:26:08.610
Julia Nimchinski: Yep, we have a minute and thank you so much and nice, and everyone. How can we make it a little bit more? I don’t know commercial, and just do a quick run of

746
02:26:09.816 –> 02:26:11.529
Julia Nimchinski: shameless, selfless.

747
02:26:12.080 –> 02:26:15.799
Julia Nimchinski: Is it an engentic promotion? Anise? Let’s start with you.

748
02:26:16.030 –> 02:26:17.399
Julia Nimchinski: What’s up with attention.

749
02:26:17.960 –> 02:26:32.330
Anis Bennaceur: Thank you. Yeah. attention.com. Where? Where your system of AI agents for sales, we automate a lot of work out of your customer conversations and interactions going from sales reps all the way up to leadership.

750
02:26:33.480 –> 02:26:34.270
Julia Nimchinski: Trevor.

751
02:26:35.130 –> 02:26:42.049
trevorrodriguestemplar: So what we do? We automate your entire soup to nuts process right from lead to renewals.

752
02:26:42.601 –> 02:26:49.009
trevorrodriguestemplar: Leveraging agents and agent. AI, of course, but we we strongly believe you need the platform and need to own the data

753
02:26:49.220 –> 02:26:54.260
trevorrodriguestemplar: in order to prevent the kind of you know hallucinations that you guys were referencing, etc. So

754
02:26:54.893 –> 02:27:11.379
trevorrodriguestemplar: that’s what that’s what we do. And you know, we strongly believe that over the next couple of years enterprise software, and especially Crm, as we know it today. Is under going to go a massive transformation and disruption. Hopefully, we’ll be leading the way over there. So.

755
02:27:12.480 –> 02:27:13.220
Julia Nimchinski: Cheng.

756
02:27:14.123 –> 02:27:26.850
Chang Chen: I lead the and so so I lead a hockey state growth agency, and we help companies to attend the actual growth. And we we specialize in AI power the growth for AI companies.

757
02:27:28.920 –> 02:27:32.590
Randy Wootton: I’m not sure about this one around the startup guy. That’s what I’ve been talking

758
02:27:32.910 –> 02:27:56.970
Randy Wootton: right. So I’ve been. I just left my last company. Maxio, I’ve started off, started an advisory service, CEO X. Dot I/O. And I’m really focused on operationalizing agentic. And what that means. So this whole conversation. I’ve been advising a couple of early stage companies, one of which I might run with the other ones. And yeah, just playing around. So always interested in talking to people about the the impact of agentic.

759
02:27:58.410 –> 02:27:59.290
Julia Nimchinski: Sandy.

760
02:28:00.440 –> 02:28:19.339
Sandy Diao: Hey, everyone! I’m Sandy. I’m pulling some data around retention for AI products. If you’re a company or a founder that has an AI product. And you’re struggling with retention problems, you know, in exchange for learning more about your business. I’d love to share with you some of my learnings, and perhaps some some tips and strategies on how you can improve retention for yours.

761
02:28:20.440 –> 02:28:21.150
Julia Nimchinski: Michael.

762
02:28:22.800 –> 02:28:31.296
Mike Haylon: Thanks so much for having me really enjoyed the discussion. I lead our go to market for our recently launched product, AI studio, which is a no code

763
02:28:31.670 –> 02:28:34.830
Mike Haylon: workflow builder to empower teams, to

764
02:28:34.870 –> 02:29:02.800
Mike Haylon: work cross functionally, establish really clean and easy workflow leveraging AI to solve a lot of the cross functional problems that we described exist today. And that’s, of course, on top of our collaborative work management platform that allows you to connect individual tasks all the way up through to your mission, critical goals and organization and everything in between, you can access it for free [email protected].

765
02:29:03.950 –> 02:29:06.620
Julia Nimchinski: So, and last, but not least, bummer.

766
02:29:07.230 –> 02:29:15.699
Omer Gotlieb: Thank you. It was great speaking with you guys, sales peak AI is actually redefining the b 2 b buying experience. And our 1st product deals with inbound.

767
02:29:15.820 –> 02:29:45.709
Omer Gotlieb: So imagine everybody that gets to your website. Big, small, qualified, unqualified, gets to have a conversation with an AI as smart as the founders can answer any question that they have, but also is a sales expert. So calculates their intent runs. A discovery process, runs a qualification process, guides them to a goal and then provide you insights back about what they’re asking and what they’re not. So all of a sudden, you know what’s going on in your website. It’s like the founder is at the gate of the website. That’s the experience we’re trying to create.

768
02:29:47.230 –> 02:29:56.490
Julia Nimchinski: Thank you so much for participating again, everyone. Please follow all of our leaders, and with that we are transitioning to our next