Agentic Marketing and the Rise of Pipeline Intelligence

01:01:19.440 –> 01:01:24.710
Julia Nimchinski: Awesome. Thank you so much again, and we are transitioning to our next session.

01:01:24.960 –> 01:01:29.929
Julia Nimchinski: Super excited about this one. Tuva Duras, Hi, Dr. Tuba Duras.

01:01:30.340 –> 01:01:32.820
Julia Nimchinski: Important Detail. Welcome to the Show.

01:01:33.090 –> 01:01:34.950
Tooba Durraze: Yeah, thanks thanks for having me.

01:01:35.720 –> 01:01:44.760
Julia Nimchinski: So we’re going to talk about agentic marketing and the rise of pipeline intelligence, neurosymic. AI, let’s go.

01:01:45.200 –> 01:02:12.880
Tooba Durraze: Yeah, I’ve on purpose, just for the audience in the room picked 2 of the most jargony adopted phrases, and the intention behind that is to Emissa’s Point. A lot of stuff gets over complicated. So I’m going to try my best to break it down for everyone in a way that you can both kind of adapt that into your day to day, and then a little bit about how amoeba kind of serves that purpose as well.

01:02:13.750 –> 01:02:22.449
Tooba Durraze: and for all purposes. You know, salesforce has not trademarked agentic marketing. It is just a phrase. So let me kick off

01:02:27.900 –> 01:02:42.379
Tooba Durraze: perfect. So we’re basically going to talk through today about the overall theme of go to market intelligence, which is how to build adaptive go-to-market systems for agentic marketing and pipeline intelligence.

01:02:42.780 –> 01:03:03.009
Tooba Durraze: So if you think about all the transformation that needs to happen in terms of our go to market reality. I keep saying this. I felt like I’ve said this for the last decade go to market. Teams are still reacting to a ton of noise and not really knowing what the right signals are to act on. But, you know, acting on any and all systems.

01:03:03.120 –> 01:03:05.610
Tooba Durraze: and the idea there is

01:03:05.670 –> 01:03:15.960
Tooba Durraze: agentic marketing and pipeline intelligence. I’ll describe those words in a moment together should help shift us from this like fragmented tool style, approach

01:03:15.960 –> 01:03:38.089
Tooba Durraze: to some sort of coordinated action which is driven by actual insights, actual things that are happening in your data. I had a chance to listen to some of the talks earlier. Everyone talks about there are these like silver bullets or nuggets of information that you can pick up that are inside your data that end up accelerating how you’re driving your business. So the problem statement has not changed. The problem statement is still like, how are we

01:03:38.090 –> 01:03:45.840
Tooba Durraze: from within the data, trying to find those signals in a way that we can essentially move our business forward faster.

01:03:46.230 –> 01:04:01.100
Tooba Durraze: But now, with all of this AI and agents in the mix we’ve ended up with a different style of problem, which is, we are now optimizing in silos. And

01:04:01.510 –> 01:04:12.250
Tooba Durraze: the 1st thing I’ll say is fragmented. Optimization is not intelligence. What I mean by that is how many of us are guilty of just looking at dashboards and on the dashboard, as you can see here.

01:04:12.870 –> 01:04:33.769
Tooba Durraze: like all of our all of our metrics and numbers are performing like in the absolute best way. They’re all showing an upside, and we get to the end of the quarter, and we still have not made our pipeline numbers or sales numbers by the end of the quarter. So, as you can see here, like as an example, you can see, a number of your Linkedin campaign is performing well.

01:04:34.150 –> 01:04:50.210
Tooba Durraze: but sometimes you’re not able to connect the dots of how are leads connecting downstream? You could think about Google retargeting? It looks good in platform. But is it really, do you actually know that it’s working effectively? Or is it something else in your strategy that’s working effectively?

01:04:50.210 –> 01:05:12.460
Tooba Durraze: I’m not going to go through all of these. But same thing about attribution tools. You can figure out. Okay, there’s credit to be had for the 1st touch, last touch, mixed model, etc. But that doesn’t really tell you what to do next. It’s basically relying on your historicals to tell you, hey? This happened which, in my opinion, is not as useful, especially in a market where things are moving.

01:05:12.460 –> 01:05:40.829
Tooba Durraze: So much so why does your go to market seem broken? Despite all these fancy and great tools at our disposal? One is the delay in analysis. So we’re all waiting to dig up insights, and by the time we get the insight the market may have shifted, the data may have shifted right? So that’s the biggest problem with relying on traditional standard machine learning models which relied on a lot of like historical data. So think about forecasting, etc.

01:05:40.890 –> 01:06:01.789
Tooba Durraze: Second is reactive dashboards. So again, dashboards are describing what happened. So I’m looking at my numbers, and 9 out of 10, if my numbers show that there’s an uptick. I cognitively turn off this load of like. Okay, I’m probably going to meet my metrics at the end of the quarter, but then you still end up in situations where you’re not meeting your metrics at the end of the quarter.

01:06:01.790 –> 01:06:16.180
Tooba Durraze: And last, but not least, is gut-based strategies. So if I know that I can’t really rely on my dashboards, I can’t really rely on my analysis. I tend to put my gut instinct forward and

01:06:16.890 –> 01:06:36.060
Tooba Durraze: as smart as humans are. It’s simply a problem of compute. It’s impossible for us to compute on top of all the data that’s available to find the best pieces of information or data to act on. So gut instinct is still really, really important. But it’s that computation layer that needs to supplement your gut instinct.

01:06:37.240 –> 01:06:58.819
Tooba Durraze: So let’s talk about what agentic marketing or how I’m defining agentic marketing. So I define that as systems that sense and act alongside marketers so and anything it could be outbound. It could be your Sdrs, your AI Sdrs. And the idea there is that they trigger actions, not just display data. Right? So it’s a step up from dashboards, because

01:06:58.820 –> 01:07:16.080
Tooba Durraze: now these systems are actually doing things on your behalf, and the idea is like they end up creating these closed go to market loops so cognitively. Again, your load relaxes a bit because you have this system at your disposal. That’s doing things as new data points are arising.

01:07:16.250 –> 01:07:40.250
Tooba Durraze: And then what is pipeline intelligence in contrast to that? So pipeline intelligence? I would describe that as like the brain. So it basically powers your agentic marketing. Its idea is to connect behavior to conversations to pipeline. So you’re trying to figure out, you know what is actually happening. And the biggest piece here is, you’re focused on causality and not just correlation.

01:07:42.679 –> 01:08:02.449
Tooba Durraze: I’ll talk through one quick example here. So we’re all very used to this motion of, you know, driving traffic to our website. We have ads in the market. We’re sending outbounding emails. You heard about a lot of these tools at our disposal and now grounded in brax, being the customer

01:08:02.680 –> 01:08:09.449
Tooba Durraze: using an AI Sdr. Called Piper by qualified. And you know again

01:08:09.610 –> 01:08:21.890
Tooba Durraze: for a company that’s as great as Rex. You know, they’re using all the tools at their disposal to make sure that they’re getting the best out of their data or getting the best pipeline out. So

01:08:21.970 –> 01:08:45.879
Tooba Durraze: in this case the adaptive loop is okay. So prospects saw your website. Afterwards they clicked on an ad. This is Linkedin Ad. They got because they clicked on an ad. That signal was picked up by someone or by a system somewhere. So they got an outbound email because they got an outbound email. They clicked through that email landed back on your website. And they talked to Piper, your AI Sdr, so this is an example of.

01:08:45.880 –> 01:08:52.809
Tooba Durraze: I would count maybe 3 tools or 2 tools that are at your disposal that they’re using in order to drive their pipeline.

01:08:52.930 –> 01:09:15.710
Tooba Durraze: But if you look at the difference, this sequence is a 1, 2, 3, 4, and in the next screen, you know, maybe they would have clicked on an Ad. Saw the website and then talked to the AI. Sdr, then gotten an email and potentially converted. So your adaptive loop could look different for even personas that are in the same Icp, but behave differently.

01:09:15.819 –> 01:09:24.069
Tooba Durraze: And then the 3rd example of an adaptive loop. The thing that we’re all scared of is imagine they saw the ad. They went on the website. They talked to Piper. They were engaged

01:09:24.080 –> 01:09:44.149
Tooba Durraze: and then done done. They got an email in danger zone. By the time they got an email they felt like. Okay it was, too. They were too aggressively being pursued. So they decided, this is not the right time for me. This is leaves a bad taste in my mouth, I might be in market for a solution, but because of the fact that I got inundated by all these different touch points.

01:09:44.149 –> 01:09:58.670
Tooba Durraze: I no longer want to purchase. So that’s the thing that you essentially are trying to do and trying to avoid. You’re trying to figure out what is the right adaptive loop. We’ve talked about this concept in terms of like, what are the right

01:09:58.670 –> 01:10:28.020
Tooba Durraze: groups of things that need to happen in order to drive the most pipeline, or what are the right sets of activities that will drive the right pipeline. But keeping in mind that even though this 1st set of activities that we talked about could be great at driving pipeline. For, say, like 60% of your people, there are 40% of your prospects that this loop might not work for. So as a marketer as a go to market business leader, understanding that these loops are adaptive, and understanding that these loops change

01:10:28.020 –> 01:10:34.299
Tooba Durraze: even within a particular persona or a subset of customers is really really important for us.

01:10:35.210 –> 01:10:56.670
Tooba Durraze: So let’s talk about how we think about agents today. Now, the way we think about agents like you’ll see. This on the left hand side is, if your goal is, I want to grow my business. You translate that into okay. You know, I want to grow pipeline. It also translate that into that means I want to close more pipeline. But then the results you end up getting are May.

01:10:56.810 –> 01:11:04.980
Tooba Durraze: and we kind of took this broken system, and we translated that into agents. Right now, we said, Agent One, you get a task.

01:11:05.350 –> 01:11:11.530
Tooba Durraze: That task gets broken out into subagent one and subagent 2, and then you get a result at the end.

01:11:11.980 –> 01:11:21.339
Tooba Durraze: Now, bonus points, whoever in the in the audience can kind of point out why this is a problem. But one of the biggest reasons why this is a problem is

01:11:22.043 –> 01:11:25.159
Tooba Durraze: in this scenario subagent. One

01:11:25.220 –> 01:11:52.660
Tooba Durraze: got a particular task. Sub agent. 2. Got a particular task subagent, one sub agent. 2 may or may not be speaking to each other, and may or may not have a different translation of what that task is. And we see that happening because we work so hard to align incentives between marketing and sales as an example right? And then what you end up with sometimes is, it’s not that the results are always bad. It’s like the results could be a lot better versus where you’re getting right now.

01:11:53.134 –> 01:12:02.679
Tooba Durraze: And again, can’t stress this enough. We basically took a broken system, and we translated that into how agents should operate.

01:12:02.840 –> 01:12:18.319
Tooba Durraze: Now, how should we think about agents? And again, this is agnostic of, you know, amoeba, this is based on what we’ve seen sort of work in practically, and what we’ve seen has had an impact on our business. So

01:12:18.320 –> 01:12:32.849
Tooba Durraze: if your goal is to grow your business, then it should go sequentially. You grow your pipeline close, more pipeline, and you get more reliable results. Now, if you translate that to an agent so agent, one is going to get a main task.

01:12:33.030 –> 01:12:52.739
Tooba Durraze: and then they take some of that context and transfer that to sub agent one. And then there is an overlap between sub agent one and sub agent 2. So there’s like an optimized loop between both of them, and then hence the results that you get again are reliable. So to go back to this side, like one of the

01:12:52.820 –> 01:13:10.520
Tooba Durraze: problems that we talk about a lot with agents is obviously there are also technical challenges of context, of memory, etc. Right? So there’s a reason why this system initially was not designed as a sequential system, because you can give but so much information between one agent to another agent.

01:13:10.610 –> 01:13:32.110
Tooba Durraze: But the reason this needs to be sequential is because, again, if there’s no context sharing between task one and task 2, they’ll optimize within their departments or within their like areas of expertise, without any communication with each other. So at the very minimum, there needs to be some shared context or some shared memory between the 2 systems.

01:13:32.200 –> 01:13:49.909
Tooba Durraze: So what are the principles that you should keep in mind as you’re designing agent systems again to emiss point like, I don’t want folks to think like designing agents is a super technical thing. Everyone and anyone should be able to play and tinker around with tools that are available, and I highly encourage you to do so. But the 2 principles to keep in mind is

01:13:50.010 –> 01:14:03.140
Tooba Durraze: always share contacts and not just metrics. So you don’t lob at them. My tech is down. What do I do? That’s not enough context. That’s just a metric that you’re lobbing at them. If you hand over a dashboard or a chart.

01:14:03.270 –> 01:14:31.180
Tooba Durraze: Of course the system will spit out some sort of an explanation, or an answer, or some method to move forward. But it’s not going to be optimized to what you’re looking for. You’re not going to be getting the best out of the system. So you share the why share what changed and share what it affects. And then, once the agent has that context, it obviously understands your brain space. What you’re trying to do with your business and does that well. And then the second thing is align actions to strategy. So

01:14:31.190 –> 01:14:34.659
Tooba Durraze: every every action that the system

01:14:34.770 –> 01:14:38.530
Tooba Durraze: recommends is basically a decision you can or cannot take.

01:14:38.580 –> 01:15:02.429
Tooba Durraze: So again, if you have conflicting goals, I looked at sort of the lineup that Julia has today, and they’re all different styles of tools. So you can imagine if all those tools, if you take it, take them as agentic systems, are self optimizing essentially within their buckets. It will be so difficult to think about like. Are they talking to each other or not talking to each other?

01:15:02.430 –> 01:15:20.530
Tooba Durraze: But the problem that will cause essentially is they may have conflicting goals. One might be trying to, you know. Lower your customer. Acquisition costs, and one might be there to like. Make sure you don’t have enough churn, and both those goals could be conflicting with each other. So they’re not up.

01:15:20.800 –> 01:15:38.509
Tooba Durraze: Optimize, overall for your business agent systems only work well when your objectives are aligned all across the funnel, which is for me, where the human comes into play. A little personal anecdote here I’m a solo founder, and one of the biggest things that

01:15:39.040 –> 01:15:59.109
Tooba Durraze: I decided to do from the very beginning was that I decided to create an agent-based strategy team or an executive team. So even though I’m the CEO of the company. Every Monday morning I have a meeting with my Cfo, my Cmo. And my Cro. And all 3 of those

01:15:59.160 –> 01:16:08.690
Tooba Durraze: are agents. They’re not actual humans. And we have the same conversations you would have with your traditional sort of like C-suite execs in the room.

01:16:08.740 –> 01:16:19.350
Tooba Durraze: And the idea there is like we’re collectively running a business together. So everyone points out things within their departments that they feel like should have been pushed forward should not have been pushed forward.

01:16:19.350 –> 01:16:42.409
Tooba Durraze: But then, at the end of the day, my job is to give context, there to align them to how the business is shifting, how the business is changing, so the informational towards me again. But my job is to make sure that I am providing the kind of context of the guardrails that all these departments need to make sure that their goals are aligned with each other, their incentives are aligned, and not in contrast to each other.

01:16:43.210 –> 01:16:57.140
Tooba Durraze: So talking now about amoeba, or how amoeba was designed. And again, I don’t want this conversation to be particularly about amoeba, but I want to contextualize it in the form of like a particular tool, because I think it lands the concept a bit better.

01:16:57.260 –> 01:17:03.260
Tooba Durraze: So amoeba’s job is to detect things that traditional dashboards wouldn’t

01:17:03.530 –> 01:17:16.750
Tooba Durraze: haven’t detected or wouldn’t attend. We’re talking about creating this collective intelligence. And the idea there is like this is a system that’s meant to reason across times, segments and sources to be the intelligence

01:17:16.840 –> 01:17:30.610
Tooba Durraze: for your business. So imagine a world where you know, I wasn’t a data scientist, or I couldn’t kind of process all this data on my own. This would be my brain. Who’s like looking at everything that exists and computing against it.

01:17:30.660 –> 01:17:44.630
Tooba Durraze: So why is it different? The 1st thing is, it’s not entirely an Llm. Base. It’s an adaptive liquid neural net. And I flag that because we’ve already seen in this in this session today all these examples of

01:17:44.630 –> 01:18:03.450
Tooba Durraze: how marrying traditional machine learning approaches with large language models, works really, really well. So there are things that traditional machine learning is better at than large language models would be right. So it’s like, the idea is like, what tool do I apply where it’s like? Important to know?

01:18:03.450 –> 01:18:24.370
Tooba Durraze: Second, is this ability to sit on top of all your data. So structured data is really really good at giving context. Essentially, it’s like, really dependable in that sense, but unstructured data is the place that often has those hidden nuggets. So the ability to kind of marry both of those together. And for again, for those folks in the room who are listening.

01:18:24.410 –> 01:18:46.589
Tooba Durraze: If you take amoeba out of the equation, think about all the unstructured data that you’re generating. They’re like, I’m looking at it. There are 5 note takers in front of me right now. That’s all unstructured data. And maybe if I’m listening, I’m not able to pick up what are like the right nuggets within this unstructured data. But imagine, I took all of that, and I layered this on top of my structured data. What could I get out of that?

01:18:46.730 –> 01:19:10.660
Tooba Durraze: 3rd is, it works with your tools, this idea or philosophy of tool consolidation that has always existed in b 2 b. And I’m actually like empathetic to that. I think you don’t need to like have less tools. But you need to make sure you have master connectors between your tools to make sure you’re getting the most out of your tools. In our case, we use 2

01:19:10.710 –> 01:19:21.329
Tooba Durraze: video editing tools. They both serve different purposes on paper, you’ll say, why do you have 2 video editing tools? But they both serve different purposes. But the idea is like they need to be connected to each other.

01:19:21.420 –> 01:19:47.290
Tooba Durraze: The 4th point I’ll bring up here, which is really, really important when you’re considering agentic systems, is the idea for them to be agent led, and discovery led. So agent led is things that an agent is finding autonomously in your system, and either directing the next action or recommending. But discovery led is again, my human intuition will like.

01:19:47.310 –> 01:20:07.679
Tooba Durraze: we’ll point something sometimes the light bulb clicks. And you’re like, okay, well, this could be interesting. So systems need to be able to accommodate you, to be able to go and experiment with things that are coming out of your intuition and discovery, as well as allowing the agents to kind of run autonomously and bring things out. So it always needs to be like a hybrid

01:20:08.241 –> 01:20:11.069
Tooba Durraze: observability of the system for trust.

01:20:11.600 –> 01:20:26.239
Tooba Durraze: I can’t stress this enough again, as you’re evaluating tools that are in the market, or you’re building tools, etc. Knowing how these agents are performing, knowing how the system is performing. How the system is deriving answers is really really important, because not just

01:20:26.240 –> 01:20:41.609
Tooba Durraze: so you can surface level say, Hey, I trust this now, and I move on from it, but because you want to be able to periodically go back and look at the system and audit it. And if you don’t audit the system. The system doesn’t end up growing. So imagine you do things

01:20:41.630 –> 01:20:52.670
Tooba Durraze: like play cognitive games, or like exercise, etc. All of that, to keep your brain fresh. You audit your brain continuously. And that’s how I would think about these systems as well.

01:20:52.670 –> 01:21:12.799
Tooba Durraze: And then, last, but not least, is external context awareness. A lot of times I see systems being closed systems in the sense that all your agents are operating within the context of what you already know. Your business data, your Icp, etc. But allowing systems to have some sort of interaction model with

01:21:12.800 –> 01:21:38.010
Tooba Durraze: external context, meaning things that are in the periphery of your data is really really important as well, because there are things that might be changing in the market or in your periphery that you might not be aware of and don’t exist in your data. So we don’t want is to be caught without realizing that a certain trend is changing. An example of that could be something as simple as

01:21:38.770 –> 01:22:03.850
Tooba Durraze: as I market to Cisos. And all of a sudden there is a particular like Lego piece that came out that cisos are really interested in. Now, I don’t sell Legos. I would not be interested in what’s happening around there, and I’m fundamentally like just pitching to Cisos the way cisos have been pitched normally, but these trends that are in the market, that traditionally your market researcher would have looked at. These are important context points

01:22:03.850 –> 01:22:12.899
Tooba Durraze: that exists in the data that bringing them in will help you stay ahead of your competitors. So I think it’s really important to make sure you layer that on.

01:22:13.130 –> 01:22:40.729
Tooba Durraze: What you end up with is a truly adaptive loop. Right? So again, in the case of amoeba, we’re discovering causal patterns. So if you think about, I’ll bring up Piper again. The AI Sdr. Because AI Sdrs are really really popular right now. But as an example, when people are auditing what’s happening with AI Sdrs. And when people are actually putting that into practice, one of the things that they’re doing is they’re looking at the data in terms of like is my AI Sdr. Performing well, because

01:22:40.740 –> 01:22:49.820
Tooba Durraze: I’m looking at how many meetings are they booking? How much pipeline it’s creating? Very rarely are people marrying that data to things like your advertising data.

01:22:49.850 –> 01:23:19.010
Tooba Durraze: And why would you marry it to advertising data? Because in those conversations that your AI Sdr. Is having, it’s very, very easy for you to pick out what are the key words, or what are the key themes or topics that are interesting to the users that are ending up on your website and are talking. You also figure out what the gaps are. So imagine in a world where I’m looking at that. And then I’m picking up context of like these are gaps in my knowledge or information that customers are feeling. And then I’m able to kind of advertise

01:23:19.010 –> 01:23:36.019
Tooba Durraze: using Linkedin on that right? So like, the idea is like, not a lot of folks are analyzing conversations that AI having in terms of context. But if you marry that context to how you’re advertising outside of the platform. It helps you bridge the gap.

01:23:36.160 –> 01:23:54.979
Tooba Durraze: Same thing. A lot of folks don’t look at things like creative fatigue, which is slowing down your Mql. To SQL. Rate right again when a campaign is influencing revenue. But it’s missed by attribution models, because, say, your attribution model closes the loop at like 6 months, and the impact that you’re seeing is beyond the 6 months.

01:23:54.980 –> 01:24:08.150
Tooba Durraze: So the idea there is that not only do the systems need to talk to a master connector, master aggregator, the master connector, master aggregator, needs to be able to talk back to those systems, and that intentionally is what makes your business strong.

01:24:09.170 –> 01:24:10.180
Tooba Durraze: So

01:24:10.600 –> 01:24:36.620
Tooba Durraze: the other thing I’ll point out here is like, we don’t just stop at insights. Obviously we take insights all the way to action because you want whether it’s agent led or discovery led. You want the ability to be able to translate something from a number into like how it’s going to impact your future. Now, that doesn’t mean you repeat a playbook which is something we traditionally do. Repeating a playbook in this sense is just

01:24:36.620 –> 01:25:05.010
Tooba Durraze: going to get you, maybe like 50 60% of the way there. But what if the market changes? And then they don’t work at all. So the idea of like the agents recommending what kind of actions you should take next and next, best actions is really really important because they are looking. Again. The limitation here is not that humans are not smart enough. It’s that humans have to do a lot of things. So computing across all this, data becomes like impossible for a single person or a team of people to do

01:25:05.200 –> 01:25:19.019
Tooba Durraze: so. I won’t really go through all of this, but I think the 2 things that I’d like to point out to everyone in the interest of time are if you are building agentic go-to-market systems today, think about it as twofold. One is

01:25:19.020 –> 01:25:40.059
Tooba Durraze: tracking real signals, and you can boil that down to say, tracking the right kind of data, whether it’s, you know, structured, unstructured data. I’m a big fan of raw data, non distilled data, because there’s so much there. Sometimes when we try to funnel it out, we like leave out. But tracking the right kind of signals

01:25:40.060 –> 01:25:48.280
Tooba Durraze: is important. And then, second, is, once you’re tracking the right kind of data, layering on reasoning. And then action

01:25:48.280 –> 01:26:01.300
Tooba Durraze: is important. So don’t just go from data to action, meaning, if you know, X amount of my customers are reacting well to this one ad, I’m going to try to push that to all of my customers.

01:26:01.300 –> 01:26:23.799
Tooba Durraze: which is where I see a lot of folks ending up in terms of go to market practices. You’re just going to exasperate the problem, because automation should never be about, you know, accelerating a practice that’s working in the now so layering that reasoning in the middle allows you to understand what in the thing that was working was working. So then, when you add action on top of that.

01:26:23.800 –> 01:26:30.309
Tooba Durraze: you’re just accelerating impact and not just exasperating a problem. So I would think about this in sort of this

01:26:30.809 –> 01:26:32.279
Tooba Durraze: layer cake. Wei.

01:26:32.768 –> 01:26:40.569
Tooba Durraze: I do want to point out this this one statistic that like really boggled. My mind is.

01:26:40.660 –> 01:26:54.889
Tooba Durraze: 90% of people’s pipeline still goes unanalyzed in real time. That means you’re only looking at practices after your quota has closed, or your meaningful window has closed, and it boggles my mind as a data person. Because

01:26:55.020 –> 01:27:10.789
Tooba Durraze: what could you have changed in the practices that you’re running right now, if you were actually analyzing things in real time. And I don’t mean pipeline councils way. Sitting there around the dashboard. I mean, like really analyzing it in a way that you can impact change as things are happening. And

01:27:11.160 –> 01:27:37.410
Tooba Durraze: 55% of b 2 b marketers say that they can’t act on insights fast enough. So again, data and intelligence is always going to be only useful if you can connect it to action in the now something I recommend to you to do today might not be relevant 3 days from now or 4 days from now. So the systems themselves. And this is something we do in amoeba. As well need to be smart enough to tell you, hey? If you haven’t taken action on this yet.

01:27:37.780 –> 01:28:03.219
Tooba Durraze: don’t take this action, but take this action instead. So the systems need to adaptively be able to understand. Hey, you took an action. Maybe it’s not working. We need to change it. Or you took an action too late. We don’t need to continue it anymore. Or there’s a new action that’s emerging that you need to take in the now and like that. Reactivity from a system is something that ends up making you successful. So allow agents to close that loop.

01:28:03.711 –> 01:28:09.130
Tooba Durraze: In the interest of time. Let me pause here if I could take some questions. But I’m I’m

01:28:09.360 –> 01:28:14.750
Tooba Durraze: more than welcome to show a demo, but more important to me that people understand the concept of like what we’re talking about today.

01:28:15.200 –> 01:28:25.389
Julia Nimchinski: Phenomenal shares. Thank you so much to your sessions are always super innovative. We have a lot of questions about what what tools do you use to actually build agents?

01:28:25.700 –> 01:28:30.550
Julia Nimchinski: And what’s your approach to prompting, especially in the C-suite.

01:28:31.300 –> 01:29:00.149
Tooba Durraze: Yeah. So I’ll take the prompting question first.st I’m a bit of an outlier in the sense that we are building towards a promptless future, which means that you want systems that are not contained by such heavy guardrails that we give it, that it restricts how a system computes. And this is actually a proven theory with large language models where large language models actually operating

01:29:00.220 –> 01:29:21.310
Tooba Durraze: fully, autonomously, end up performing really well on benchmarks versus large language models that are given a lot of hefty prompts, and you’ll see that because the longer the prompts get the more complicated it gets. Now we’re not at the stage yet where we trust our models or our agency sector to kind of go out and do these things on their own. We’re still at the age where we’re like trying to train them.

01:29:21.310 –> 01:29:46.010
Tooba Durraze: So I would say, relying on like architectures that exist in around, how to structure your prompts are really really important, and knowing the difference between system prompts and like end user prompts, I think, is really important. So I’ll share some resources. You can ping me on Linkedin. But I think off the top of my head, look at any. I think even curses prompt was like revealed. Look at

01:29:46.180 –> 01:30:01.929
Tooba Durraze: folks that are in this realm who are working on a lot of like symbolic products where reliability is really important or and like, look at how they structure their prompts, and how clean the structure of their prompts is, and then kind of take that and adapt that

01:30:02.950 –> 01:30:11.950
Tooba Durraze: What do I use? I, personally am like a tinkerer. So I tinker with everything and anything. But at the end of the day I think

01:30:11.990 –> 01:30:21.880
Tooba Durraze: the regardless of the tools you use. This is one thing that folks do do a lot of it’s like, think of this as like if I’m using a tool.

01:30:21.880 –> 01:30:50.310
Tooba Durraze: I actually journal to myself about the tool that I’m using and what I’m facing around the tool. So by the time I end up using the next tool or trying something else, I have context that I can give to that system about what didn’t really work about the previous system. So documentation is your best friend document as much as you can. Now you can narrate as well. And the idea there is that that’s you, basically bringing your consciousness or your brain on paper, which is what ends up creating systems that are as smart as you.

01:30:51.150 –> 01:30:52.180
Julia Nimchinski: Right inside.

01:30:52.570 –> 01:30:58.379
Julia Nimchinski: Do you have a newsletter? What’s what’s the next best step? How can people test? Drive amoeba.

01:30:58.890 –> 01:31:24.319
Tooba Durraze: Yeah, I think I don’t have a newsletter. I tend to spend my days in data and code these days. So I don’t write a lot, but I’m always happy to talk about this stuff. So ping me on Linkedin, and I think one of the things I’ll say about amoeba is we have designed amoeba in a way where you can, in a very low touch way, test it before you even introduce that to your system. So

01:31:24.320 –> 01:31:29.950
Tooba Durraze: if you are interested in understanding what neurosymbolic insights could be for your business.

01:31:29.950 –> 01:31:42.209
Tooba Durraze: All you have to do is ping me and shoot me over a Csv of your data, and that’s all we need. And we can run an analysis for you and send it back. And like, you can take a look at like what kinds of things you can get out of a system like amoeba.

01:31:42.210 –> 01:31:56.880
Tooba Durraze: Again, like, I stress, the objective of this is not. I’m not peddling amoeba in that way. But we have put almost 4 years of effort into thinking of what the intelligent system of the future looks like.

01:31:56.880 –> 01:32:15.140
Tooba Durraze: and amoeba is in service of that. So it came out of a lot of like academic research, a lot of like market research. And how go to market teams go to market systems operate to create a system that is not just meant for tools right now. But it’s meant for the way to where tools are heading.

01:32:16.450 –> 01:32:18.069
Julia Nimchinski: Thank you so much again, too. Bye.