By Lolita Trachtengerts, GTM Ops & Growth, Spotlight.ai. Every revenue org I talk to right now is building an agent.
A discovery bot. A forecasting copilot. An MEDDICC scorer. A research summarizer wired to their CRM. The pitch decks all sound the same: “We’re going agentic.” The demos all look the same. And six months in, the results all start to look the same too.
The agent hallucinates a stage advance. The forecast comes out cleaner than the pipeline actually is. The “champion” the system flagged turns out to be a mid-level IC who attended one demo. Reps stop trusting the output. RevOps quietly turns the workflow off.
This is not an LLM problem. GPT-4, Claude, Gemini are all extraordinary general-purpose engines. The problem is that nobody on the buying side of enterprise sales wants a general-purpose answer. They want the answer their best CRO would give. And no off-the-shelf model has ever sat through 200,000 enterprise opportunities, watched which signals predicted a slip in week four of the quarter, and learned the difference between a buyer saying “budget” in a hallway and a buyer saying “budget” while asking for a business case.
That is a knowledge problem. And it is the one thing nobody in the agent gold rush is solving.
The Brain Is The Moat. The Agent Is Just The Interface.
When we started Spotlight.ai, we made a bet that the model layer would commoditize. It has. What does not commoditize is domain knowledge structured for machines.
So we built one. We call it the Spotlight.ai Knowledge Graph. Today it holds more than 40 million atomic sales signals, drawn from $8B+ in managed enterprise opportunities. It is built in three compounding layers:
- An enterprise sales layer encoding how complex B2B deals actually get won and lost. This is not scraped from the internet. It comes from our founding team and from hundreds of CROs we worked with before we wrote a line of product code.
- An industry layer that knows how deals are structured in your vertical, how your competitors run their playbooks, and what signals mean something versus what is noise.
- A company-specific layer that gets smarter every time your reps run a deal through it.
A keyword search can tell you the word “competitor” was said on a call. The Knowledge Graph can tell you who said it, what stage they said it in, what role they hold, what their engagement pattern has looked like for six weeks, and whether that pattern matches the last 4,000 deals where a champion went dark and the deal slipped. One of those is data. The other is judgment.
Reps do not need more data. They need judgment that scales.
Why we put it on MCP
If you want autonomous selling to work at scale, two things have to be true.
The brain has to be portable. And every team building an agent on top of it has to get the same answer the platform gives.
That is the bet behind exposing the Spotlight.ai Knowledge Graph through the Model Context Protocol. Anthropic’s MCP standard does for AI context what REST did for APIs. It gives any agent (yours, ours, a partner’s, a Claude or ChatGPT instance running inside Salesforce) a clean, structured way to ask the brain a question and get a grounded answer back.
Concretely, when you connect to our MCP server, your agent can:
- Ask whether a specific contact looks like a real economic buyer, given their actual engagement pattern, not a title match.
- Ask whether a deal is at risk of slipping, with the signals that justify the answer.
- Ask what the next best step is on a stalled opportunity, scored against thousands of similar deals in your industry.
- Ask for a business case grounded in the actual buyer’s metrics, not a template.
You do not have to retrain a model. You do not have to build the data pipeline. You do not have to talk your CISO into pumping every customer call into a foundation model and praying it does not memorize anything. The graph is the layer in between.
What This Changes for Builders
I have spent the last year talking to RevOps leaders, AI platform teams, and the consultants now sitting between them. Three things keep coming up.
First, agents fail in production for the same reason chatbots failed in 2019. No grounding. The model sounds confident. The output is wrong in ways the rep cannot easily detect. By the time the manager catches it, the deal is gone. A purpose-built knowledge layer is not a nice-to-have. It is the only way the rep ever trusts the system.
Second, the cost of building this in-house is wildly underestimated. Encoding enterprise sales judgment is not a sprint. It took us years and several hundred CRO conversations before we processed our first customer deal. Most teams trying to replicate it from scratch will burn 18 months building a graph that is shallower than the one their competitors are renting.
Third, the winners will be the teams that stop building the brain and start building the experience. The same way nobody writes their own database engine to ship a SaaS product anymore, in two years nobody serious will be writing their own sales reasoning layer. They will be plugging into one and competing on the surface above it: the rep workflow, the manager dashboard, the customer-specific playbook, the design.
That is the world we want. We will keep shipping our own agent squad on top of the graph because we know what good looks like. But the graph itself is meant to be infrastructure. Use it.
The Honest Part
I do not think every sales team needs a Knowledge Graph. Small teams selling simple products into known buyers do fine with a CRM and a good manager.
The companies that need this are the ones running real enterprise sales motions. Multi-stakeholder deals. MEDDICC. Long cycles. Deals that slip and nobody can quite say why. If that is your world, the question is no longer whether to put AI in front of your reps. It is whether the AI you put in front of your reps actually knows what it is looking at.
A confident agent grounded in nothing is worse than no agent at all. It will make decisions, and your reps will follow them, and at the end of the quarter the forecast will still be wrong.
Build on a brain.
Media Contact:
Lolita Trachtengerts
VP Growth & GTM Ops, Spotlight.ai
Email: in**@*******ht.ai
Spotlight.ai
LinkedIn
Twitter
Facebook
Instagram