One Context Layer for Autonomous Revenue

Julia Nimchinski:
Please welcome Deepinder Singh Dhingra, founder and CEO of Rapture AI, and today, Deepinder will demonstrate the context layer for Autonomous Revenue. What a pleasure. Always love featuring new Deepinder. I almost became a voice of reason.
Deepinder Singh Dhingra:
Great, great to be back, thanks for having me.
Julia Nimchinski:
Amazing, let’s dive in.
Deepinder Singh Dhingra:
Perfect. So, hi everyone. I’m Deepinder, I’m the founder of RevShore, and what I’m going to be speaking about today is the missing context layer in your GTM motion, right? If you’re an upper-mid market, if you’re an enterprise, a B2B company who’s trying to deploy Agentic AI across your GTA Motion.
And you’ve struggled to do that because you have disparate tools, disparate systems, you have every team trying to run their own agents by themselves, little collaboration. This is the context layer that you need to make Agentic AI streamlined and coordinated. So why do we need that?
If you think about what’s happening, over the past 10 to 15 years, we all have invested in a number of tools across the GTM motion, right? We have tools for marketing, tools for sales, tools for the SDR BDMotion, tools for the AE motion. It’s called the frag… it’s not called the fragmented GTM and the Frankenstack for no reason, right?
Disconnected tools, messy data. multiple buy handoffs, which leads to a lot of context loss. Each of these tools are great, you know, in their own purpose, but they were never built to connect with each other. And neither were they built for agents, they were built for human beings.
So any effort that connects these tools are more like band-aids from an Agentic AI perspective. And if you try to deploy Agentic AI on top of this GTM tech stack, it only amplifies the chaos. Because without complete context across what’s happening across your buyer motion, across the marketing motion, the SDR, BDRA motion.
Each agent that is running, let’s say you have an SDR agent that’s trying to drive outbound, and if it has no context of what happened with that buyer, with the lead account, your prospects, your visitors, and the marketing motion, that just amplifies the inefficiency. It automates the 10X, but also amplifies the inefficiency by 10x.
So our perspective is. there is a missing layer in the GTM tech stack. Now, what we find is that a lot of enterprises are trying to figure it out themselves, right? They’re trying to… you know, deploy data integration software, data lakes, CDPs, and the plethora of Agentic orchestration tools, the frontier models, etc.
But they also facing a severe talent gap, right? There’s over 3,000 open positions for GTM engineers, right? Which is an interdisciplinary skill set. You need GTM process, architecture, Agentic AI engineering. Agentic AI is not static software, they’re living, breathing systems. So bringing all of this together itself is a monumental task.
for any organization to undertake. So our perspective is, we want to make this task easier. We want to help enterprises with complex GTM tech stacks become Agentic AI ready. And that’s what this missing context layer does. What we do is we unify all of the data, the intelligence, across your fragmented GTM tech stack.
We are not replacing or disrupting your current GTM tech stack, we are just a layer on top. This unified context layer is just a layer on top. that seamlines, unifies all of the fragmented context across the GTA motion. brings it into one AI brain, which is the context graph.
Which further extracts signals and learnings, and then exposes it in one unified way to power Agentic actions, where every agent or every Agentic deployment has access to the full context and is aware of every other agent. And that’s the key.
If you want to deploy coordinated Agentic AI across your GTM motion, you need this unified context layer that becomes this unified brain. Now, how does this work? It works by integrating all of the touchpoints, all of the data, all of the interactions, your business context, ICP, segments, channels, products.
the billions of interactions that are taking place in your GTMotion across sales, marketing, across the SDR BTMotion, your chat interactions, email interactions, call, both structured, unstructured, all of the activities that you’re doing across your GTMotion, and then brings it together into this context graph.
harmonizes it, identity resolves it, unifies it, so that now you have one view of what’s going across the GTA Motion. what’s going on across accounts, what’s going on across each lead, each prospect, each visitor, what’s the impact of your campaigns, etc.
And it not just captures all of these entities, it also captures all of the events, the interactions, the metrics, and the decision traces. that are coming from all of these data sources. And then, it applies reasoning on top of that. So imagine you have millions of visitors, millions of contacts, leads, you have hundreds and thousands of accounts.
depending on whether you’re a PLG motion, an ABA motion, inbound, outbound, product-led, marketing-led, sales-led motion, realistically, with the enterprises that we work with, all motions are operating all at once, right? So imagine you want to bring all of this context together.
And you have the ability to bring this into a context layer and a context graph that harmonizes everything. And I’m going to make each part of what I’m saying real very quickly, and then exposes to AI agents, and then can drive coordinated GTM actions. So now I’m going to switch to showing you how this actually works.

Before I move forward, I just laid the foundation. So, for example, here you’re looking at Refshore’s power user interface, and what you’re seeing here is the first step in doing anything and building this missing context layer and unified context layer is integrating all of the data across your GTM tech stack, right?
So here, for example, you’re integrated with your marketing automation system, the CRM system. If you’re running gifting campaigns, G2 reviews, a first-party pixel that does cookie or cookie-less tracking, based on consent, fingerprinting, server-based tracking, all of your paid ad campaign tools, de-anonymization. Twitter.
your demo tour interactions, call recording interactions, product usage, email, chat conversations, all of those data sources across your marketing motion, your sales motion, your SDRB radio motion, your customer success motion, right? And in the enterprise, it can get really messy.
We have customers Where on average, they have tools between 20 to 75 tools across the GTM tech stack. Yes, it can go as large as 75. That’s our largest deployment as of today. In the enterprise, if you’re just focused on sales, maybe you have 10 to 15 tools, but when you go across the GTM motion.
the heterogeneity and the number of tools that you’re working with just explodes, right? And you can integrate and harmonize this, because if you want to build a unified context layer, you want to be able to bring all of this together. One might have multiple CRM systems, one might have multiple marketing automation systems.
We find those situations very often. We find people transitioning from one marketing automation system to another marketing automation system, acquisitions happening, new systems getting adopted and acquired into the GTA motion, and all of that gets harmonized. Once I have this, what it does, it actually builds the data graph.
every part of the GTA Motion and the tech stack is now harmonized identity resolved into a graph. So, for example, if I look at all of these leads, accounts, campaigns, activities, they are harmonized into one GTM data graph, linked, deduped, and identity results. So, for example, click on leads.
All of the leads and contacts across all of these systems, the marketing automation systems, the CRM systems, people engaging with you on organic social. The visitors getting deer and mice. you know, users, from product usage, from calls, from emails, people that are not even in a CRM and marketing automation system getting identified.
Similarly, getting All of the accounts across all of the systems, and your prospecting motion, in the marketing motion, accounts getting dearized. Similarly, all of the campaigns, all of the activities, all of the visitors, etc. are getting harmonized. All of this comes and gets stitched together. into this context layer. Now, how does this look?
Right? Once I have this context layer, I’m able to now, for any lead account or opportunity, I’ve stitched together all of the interactions and all of the context all of the activities, campaign touches, funnel movements, so imagine the millions, like, we have customers who have billions of interactions across their GTM ocean, right?
Call interactions, page search interactions, email interactions, all of this. that I stitched together from all of these data sources are now available. Right? This particular lead has 79 touchpoints. You see interactions across digital, non-digital, chatbot, web form interactions, ABM interactions, email interactions.
For every lead, similarly, all of the interactions are there. For every visitor, for every prospect, etc. Now, once I have these interactions, I can further extract signals. Right? I’m extracting signals to enrich my context layer. For example, what competitors did this lead mention? What is the buyer persona?
Everything is getting inferred and enriching the context layer for what is the propensity that this particular lead will convert into pipelines this quarter, next quarter, next to next quarter? What are the factors driving this propensity? What should be my next best action for this particular lead, or for this particular account?
All of this is further enriched into the context layer. The context layer is not just a data layer. It’s a layer where you harmonize all of the data, but you’re extracting signals to make it available for Agentic AI reasoning, and all of these signals All of these interactions are coming and getting harmonized from across your GTA motion.
It is not one… system that it’s getting organized.
For example, if I look at accounts now, all of the interactions across at the account level, all of the contact level interactions, all of the account-level interactions, what is the unique timeline journey, what is the influence journey, all of these signals are getting processed within the context layer.
The UI is a way for us to help understand what’s happening across time, over time, the ability to do time travel, the event traces, all are coming together, even the decision traces and the reasoning traces on the learning engine. Once I have this, now I have a unified context layer along with signals across my GTA motion. Right?
I have signals at the lead level, account level, opportunity level, I have meeting signals, I have chat signals, I’m extracting intent, upsell signals, propensities, etc. Now I can make these available To… for agents to act on. So, for example, and you could make this available as an AI brain to any Agentic orchestration layer.
For example, you could do it on Claude, you could do it on ChatGPT, you could do it on Glean, on Codex, etc, right? So, for example, I’m now… I’m here in my Cloud environment. I have the RevShow connector set up. With this RevShow connector. every aspect of my context layer is now exposed through MCP tools, right?
All of the account signals, account details, account engagement trends, account journeys.
the campaign… the impact of campaigns on accounts, there’s a whole set of engines, attribution engines, signal extraction engines, AI learning engines, propensity engines that are informing the context layer, and every aspect Hoff… context, and the reasoning on top of it is now exposed as an MC2… as MCP Tools.
These are MCP tools that span data, span insights, and intelligence, span signals, as well as span action primitives, right? So now, instead of trying to access 10 MCP servers.
I see a lot of companies are very glad when they say, you know, in real time, I can access data from Salesforce, and I can access data from HubSpot, and I can access data from… you know, some sales automation tool, etc. Why do you need to do that? Because the more MCP servers you’re interacting with, the more latency.
The more identity and entity resolution problems, that will require token usage. whatever tool you’re using, whether it’s ChatGPT or Cloud or Glean, etc, will incur.
Instead, go after one MCCP server, because that is providing you a unified context layer, which is secure, which is… Behind authentication, and authorization, gateways, so that you can just access one MCP server. So just like Stripe. collapsed all of the payment interfaces into one API.
what the unified context layer does is it abstracts all of the MCP’s servers that you need to interact with into one MCP server with a rich, a very rich set of tools that can help you analyze and take action on any aspect of your GTA motion. So once I have this. Now one can go crazy with that, right?
You can customize, you can build your own skills, right? You have skills that are deployed. right, across the GTA Motion, skills for lead prioritization, account prioritization, skills for prospecting, skills for enrichment, skills for understanding your pipeline health, skills for orchestrating sequences. etc. You can drive skills on top of it.
So, example, here, I’m showing you one example of my skill that helps me do my lost deals, win… play… win playbacks, right? And what you’re seeing in the skill is very interesting. Right? And what I’ve done is I’ve masked all of the accounts and lead names, right, as much as possible.
But what this skill is telling me is for every account that I recently lost. right in an opportunity, how to win that back. But just this one skill is accessing all of these tools from the unified context layer.
The whole tool trace, the, you know, it’s getting my funnel stages, there’s a memory, what are my… what are my opportunity details, you know, what are the reasons I lost.
All of this, then it’s actually going and say, for that particular opportunity, for that account, what are the leads and prospects I can go after, enrich those leads, get more prospects, right? Then access more agents. You know, maybe I want to access a trigger agent that’ll help me research the account.
Not only from the data that I have, but from the web data, and create an action plan, and then run that agent. So, I can use skills now, and I can orchestrate coordinated GTM actions, and I can get insights, but I can also act on it. Right?
So, for example, this one is giving me all insights on what are my real loss reasons, what are my champion verification, what should I do next, what should outreach emails I should kind of write. Similarly, here’s another one.
that, you know, I want to kind of research one account, and it’s actually, you know, gone out and told me what I need to do, right? What are the signals on the account? What are the outreach drafts? It’s researched the full spectrum of leads and prospects.
the emails, phone numbers, everything that I need, and so that I can reach out to them, right? And it can also take action. Right? Or, for example, I want to find my high probability leads, and I have, you know, leads that I ran a skill from high probability, and it’s actually showing me all of the MCP tools that it’s used.
The great part of this is that it’s a unified context layer, so the MCP tools are coming from one MCP server, because all of the context has already been integrated. You don’t need to go after 10 MCP servers again, right? And that also leads to reduced token usage and reduced costs. Faster, and faster agents, right?
All of these skills I can do, I can also orchestrate within the unified context layer, and you turn the unified context layer into a system of action. So, for example, here. we have our own agent orchestration layer, where you can deploy agents.
You can deploy agents in Cloud, you can deploy agents in Glean, you can deploy agents in… cursor, you can deploy agents in OpenAI, etc, whatever your favorite tool of orchestration is, and Manus, and OpenClaw, etc. But you could also deploy them in an Agentic hardness. that also applied guardrails, right?
Because you want security guardrails, you want PIA reduction, and RefShow’s Agentic orchestration layer provides the necessary guardrails in which you can run Agents in a safe and secure way, for example, and you could deploy agents as a team of agents.
The great part is that each of these agents is accessing the same context layer, so each agent is aware of the same context that happened to every lead, account, opportunity, campaign, channel. within your GTA motion, as well as each agent is aware of the other agent’s action. That’s a very important aspect.
So if I have an outbound agent, right, like, do I really know whether marketing is targeting and nurturing that lead, and what interactions are happening? Right? Mostly no. But in Refshore’s case, because of the unified contact letter, every agent knows what the other agent is doing.
And you could run prospecting agents, SDR agents, you could LinkedIn campaign refinement agents, you can run list enrichment agents, you can run signals-based outreach agents, you can run SDR conversation agents.
Agents across the spectrum, across the sales motion, marketing motion, SDR, BDR motion, form-filled responder agents, personalized email agents, a team of agents approach that’s sharing the same context prevents agents from growing rogue and maximizes the value or from your Agentic GTM ocean, towards your Autonomous GTM. initiative.
So, for example, I might run, for example, an account research agent here, if I come back, right? If I want to run, I could just run this agent, enter Any account name. You see, I did the same in Claude, but I can do it here. I can say I want to, you know, research one of these accounts, Harvey, and I can click run. I wouldn’t do that now.
I have my run history, which keeps a track of all my… Agents, so I can just click on Harvey, and it’ll actually bring me all of the account research, in addition to real-time web signals and web monitoring.
I have the full details of the account, the key stakeholders, the strategy I need to follow, the competition, the most recent web signals, what’s been happening within that account, and, you know, over the past 6 months, the past few days, you can go as granular. What is the LinkedIn profile? What is the action plan?
Who are the key stakeholders that I need to take action with? If I wanted to, what are the intent signals? What is the task list? And I could run this through an agent, right?
So you can build Agentic orchestration right, on top of this unified context layer, in any Agentic environment, or you can build it on top of RevShow’s Agentic orchestration harness, which has all of the context already, but can also do it in a safe. and secure way, where you don’t have to worry about authorization.
You’re using one unified context layer. So, coming back, that’s what the unified context layer enables for you. It enables a way for you to deploy Agentic AI in a coordinated way, on top of a unified context layer. This becomes the AI brain. It could drive agents, right, in any Agentic orchestration layer. that you might want to use.
Also, Refshow, we have our own Agentic orchestration layer, and we have our own ability to drive autonomous bots on Slack, etc, give you notifications, alerts, hot leads, heart accounts, etc. Drive agents across marketing, SJR, BDR, sales, customer success. But more importantly.
It also helps you interact with other ecosystems, because once you have this unified AI brain, you can drive different aspects of your GTM motion and interact with your current GTM tech stack. There’s a bi-directional flow between the unified context layer and other aspects of GTM tech stack.
So you’re bringing context from the GTM tech stack, but you could also send actions onto… back to those GTM tech stacks. So you keep your GTM tech stack, don’t disrupt it, because you don’t want to. If you’re an enterprise, you’ve already spent millions of dollars.
on your current GTM tech stack, and you have heavy investments in workflow and process, etc. So, you don’t need to disrupt any part of your GTM tech stack, and yet you can become Agentic AI ready. Right? By using this missing layer, this one context layer.
which is identity resolved, governed, and shared by every agent, you can deploy agents that will help you accelerate your GTA motion.

Julia Nimchinski:
Thank you so much for the presentation. So many insights here. Deepinder, have a lot of questions, but I’d like to start with a point of context. A lot of vendors in the industry, they claim to have context, or something that is called shared context, so can you just share your philosophy at Rapture?
What’s the actual difference in how an actual shared context layer would differ from some other vendor claiming that they have it?
Deepinder Singh Dhingra:
Yeah, so our perspective is that the shared context layer needs to be able to, first thing, I think some of the other speakers also talked about, data being the most important aspect for agents. Today. Right? It’s not the Agentic orchestration layer that’s a bottleneck, you know, there are many, it’s getting fragmented.
It’s not the frontier models that are bottleneck, anymore, they converge every 12 months anyways, right? What the key mode and the key ability to drive Agentic AI with effectiveness is the context. Now that context layer needs to be able to bring in all kinds of interactions across your GDMotion, right?
You need a common view of what’s happening with every lead account, opportunity, campaign. with every prospect in your GTM motion, a common view, that’s the first aspect, right?
And that itself requires… the promise of CDP, which most CDPs have not been able to fulfill, is the ability, because they don’t have a domain-specific, purpose-built GTM model that gets encoded within the context layer. Most shared context layers approach it from a technology problem.
They approach it from a problem saying, hey, you know what, I’ll put a context layer, then I’ll have a semantic data model, and then I’ll have, you know, the ability to track traces and reasoning and decisions and feed that back. That’s all good, in theory, right? And you can build the technology components to do that.
But the art, and the devil in that, is how intimate are you with the specifics and the nuance of the GTA motion? How much are you encoding the GTA motion context? What does pipeline mean to you? Right? What does the lead lifecycle stages mean to you? What do your buyer journeys mean to you?
This is semantic meaning encoded in the GTM… into your context layer, which becomes very important. Yes, tracking events, tracking decision traces is important, having the right memory layer is important. So from a technology perspective. You need to have the right technology foundation But it needs to be purpose-built from a GTM perspective.
So we, when we talk about the context layer, yes, you need all of these layers to make this happen, right? You obviously need the… you need to connect and unify data from across the… technology systems that you had that I showed you. You need to have an enterprise foundation around security, tenancy, region, audit trails, etc.
You need to have a context graph. Right? You need to have semantic intelligence, you need to extract signals and predictions. This is one key differentiation. People just limit the context layer to the interactions, the data, the events, and the metrics, and the inputs and the outputs, but they don’t extract signals on top of the context layer.
You can’t… you have to learn at billion touchpoint scale to understand the impact of a campaign on progression of a lead journey. You can’t just, you know, have an MCP call on that.
You need to… you need to unify context and learn at the billion touchpoint scale, or the hundreds of millions of touchpoints to understand attribution, or to understand propensity, right? If you want to be productive. and then you want to put an activation layer. These are the technology building blocks that are important.
But the devil is in how well you encode the GTM model, and the GTM motion context, and how you encode the GTM your specific GTA motion into the context layer, and that’s where most context layers fail.
Most horizontal context layers will fail, or will find it very tough to adapt to the specifics of a vertical, and that’s what our key differentiation is. We are purpose-built for the B2B go-to-market motion.
And we have a semantic layer that encodes the GTA motion off of every customer, or every enterprise when we work with them, and that is where the secret sauce is.
Julia Nimchinski:
Thank you. And one more topic I wanted to address is the future of, you know, AI monetization, AI business models, and the reality today. Because again, there is a lot of… various, you know, marketing going on in the industry. There were a lot of claims, and curious, you know, your thoughts. You really became the voice of reason in our community.
So, what’s actually happening? What’s realistic today? How do you approach it at Fapture?
Deepinder Singh Dhingra:
Yeah, so we run our whole, you know, we have our own context layer, we have RevShore on RevShore, right, and we run our whole GTA motion on RevShore. So our perspective is, threefold. On the context layer, you need a unified context layer, so our marketing team, our sales team, our SDR BDRs, our running agents.
And each team knows what the other agent is doing, so they have full visibility of what’s happening, and that should be expected. So that is real today, right? You should not wait for… that’s not futuristic, that is real today. on the Agent AKI side, I think there’s a lot of talk about autonomous GTM, right?
I think we will get there, but there’s a lot of management of the agents that is required, so you need to… these agents are living, breathing systems. So you need to monitor, fine-tune, eval agent as they’re running in production, and so I think there’s still a manual effort to monitor… not to monitor, you can automate monitoring.
But you need to be able to refine agents as you see the performance of agents degrading. I think reinforcement learning is there, but reinforcement learning is not there for every type of agent. Right?
So, different types of agents, some will have automated reinforcement learning, some will need management, which is where the human beings come in, and that’s where the expertise of the marketing person, the marketing ops person, the salesperson, the SDR, BDR, etc. is very important.
We’re also at a stage where you need a team of Agents approach, where it’s not one super agent that will do your job. You need multiple… and when I showed you my… the agent portal, where you had multiple agents, but those agents share the same context, right?
I think there’ll be a future where you might have You know, completely autonomous GTM, but we are believers that today. a team of Agents approach, where you actually right-size the Agentic task and the Agentic goal, compared to trying to achieve… trying to do everything at once. So that’s our perspective.
Julia Nimchinski:
Thank you. And last question, I just wanted to address the, you know, the most boring topic, guardrails and, obviously, harness. So, I mean, there is a lot of talk in the industry about, you know, feeding your alpha as an enterprise. back to the Frontier Labs, so what are your thoughts and recommendations here?
Deepinder Singh Dhingra:
So my thoughts is that you should not give your IP to the frontier models. Your IP needs to be in the context layer. Right? So, right?
Obviously, the guardrails are PII redaction, and no prompt logging, and no prompt caching, and no training, based on your data, but your IP your sovereign IP is the context of your GTA motion, which is inclusive of how you run your GTA motion orchestration, what is your… who your ICP, your business model, your segment, your value propositions, your product.
We have a whole configuration layer on the context layer that encodes that, which I haven’t shown you today, but it kind of encodes the whole GTA motion. That is your IP. The second part of your IP. is the data and the interactions, and that also you should not outsource. Yes, you can use… you can send selectively context. Right?
To the frontier models, or to the LLMs to get back responses for the reasoning, right? But that’s our perspective. You need a clear separation of the context layer versus the Agentic AI reasoning, right?
And so that’s our perspective on that, and that’s why you need sovereign ownership on top of your contacts, and that’s the most important, that’s what we support as part of our context layer, but that’s our perspective on the guardrails, yeah.
Julia Nimchinski:
Phenomenal. Thank you again, and what’s the best way for our community to engage with you? Go to Rateshar, or message you directly?
Deepinder Singh Dhingra:
Yeah, feel free to go to RevShow.ai, R-E-V-S-U-R-E dot AI, or just contact me on LinkedIn directly, or my email, Deepinder at refshore.ai. Would love to chat.
Julia Nimchinski:
Thank you so much.
Deepinder Singh Dhingra:
Great, thank you, appreciate it.