SciFin came out of stealth on September 1 with $44M and a thesis: the revenue stack was built on data salespeople never wanted to enter. We spoke with Krishnan "Krish" Badrinarayanan, the company's VP of Marketing, a few days before Dreamforce, to understand what "context convergence" actually means, and how you build demand for a category nobody has a name for yet.
Until SciFin, it took three years and teams of engineers to answer a question every sales leader asks every Monday: what is actually happening in my pipeline?
"It was a multi-year project," Krishnan Badrinarayanan told me. "Teams of engineers build a data warehouse, then dashboards on top of it. And the problem with that whole approach is that you're pulling all the data, but is the data clean? There was no process enforcing data cleanliness."
Badrinarayanan ran product marketing at Nutanix, and Zscaler before joining SciFin as VP of Marketing two months ago. He is now the person tasked with explaining, to a market that already owns Salesforce, Gong and Clari, why it needs one more thing. His answer is that it doesn't need one more thing. It needs a layer that draws from all of them and more.
The round itself is a $44M seed, co-led by Altimeter and Madrona, with Foundation Capital, S32 and Zetta Ventures participating.
The Rolodex problem
The way SciFin tells the story, the modern revenue stack grew one layer at a time, each layer solving the problem the previous one revealed.
"When Salesforce came along, it was fantastic," Badrinarayanan said. "All those Rolodexes with business cards, you could digitize them. That's basically what Salesforce did. It digitized your Rolodex. Contacts, leads, and then activity on top of it: opportunities, from start to finish."
The CRM did exactly what a system of record is supposed to do. But the sale itself kept getting more complex. In B2B, a deal takes six months and touches a dozen people on the buying side, and a record of what happened is not the same as an understanding of why. Sales teams were missing signals. So the industry bolted on insight tools: Clari for forecasting, Gong for call recordings and sentiment, data warehouses to hold history across time, dashboards to read it back. Each one, in his words, tells "its own small story, based on the data and the context it has."
And every one of them rests on the same shaky foundation: a salesperson typing things into a form.
"What do salespeople like to do? They don't like to enter data. They're not data entry specialists. They want to spend time with customers and make money, for the company and for themselves. So the data is stale, the data is missing, sometimes it's just plain wrong. And then you're building context on top of that."
That distance between what the systems say and what is actually happening in the field is what SciFin calls the Context Gap. It is the problem the company was built to close, and the phrase it intends to own.
Infrastructure, then data, then context
If the diagnosis sounds familiar, it's because SciFin's founder has made a career out of it.
Mohit Aron was a lead engineer on the Google File System, the layer that made thousands of scattered disks behave like one. He co-founded Nutanix in 2009 and pioneered hyperconvergence, collapsing compute, storage and networking into a single software-defined box. He founded Cohesity in 2013 and built SpanFS, which turned a dozen incompatible backup silos into one platform with search on top. Two of those companies became decacorns.
"He first started by converging infrastructure," Badrinarayanan said. "Then data. And what sits on top of data? Context. Now he's converging context."
The origin of SciFin is, by his account, personal. "The reason he started SciFin is that it directly addresses the problem he faced at Cohesity. He found himself sitting in meetings, hours after hours, collecting data, having conversations with sales leaders and sometimes with sellers in the field, just to understand what was really happening. In this day and age, with so much sophistication all around us, it seems ridiculous that you have to go through that process."
Aron started building at the end of 2024. The company stayed in stealth for nearly two years, which Badrinarayanan attributes to one thing: "It's a hard problem to solve."
Not a CRM, and not a replacement for one
Here is where SciFin's positioning gets interesting, and where it departs from most of the AI-native startups circling the same market.
"All the other AI-native companies coming up in this space are saying: replace everything you have with what we have," Badrinarayanan said. "And that's a scary thing. You don't want to disrupt something like revenue operations."
So SciFin sits on top of the existing stack rather than ripping it out. It reads from the CRM, the email, the calendar, the Slack threads between a rep and a sales engineer, product usage data, and support tickets, and assembles all of it into what the company calls a revenue model. The pitch is that a customer can "live in both worlds" during an AI transition instead of betting the quarter on a migration.
But the more important move is on the input side. SciFin ships its own conversational intelligence and its own note-taking. It listens to the calls. And it uses what it hears to fix the data problem at the source.
The example he gave: during a customer call, the system notices the executive buyer is present, that the buyer committed to a timeline, that the tone was optimistic with no hesitation. It concludes the deal is ready for the paper process, proposes a status update, and asks the rep to approve it with one tap. The revenue model updates. No form was filled.
"It's not even CRM anymore," he said. "It's a revenue operating system. It's much bigger than CRM."
The design principle behind it is what he calls a self-driving experience: the product should follow the way business is already done rather than insert a step into it. A rep can use the mobile app, the web app, Slack, or just talk to it. "I'm about to head into this meeting with this customer. Tell me what I need to know and what I need to do for a successful outcome. And it gives you the whole brief, via voice."
The assistant is called Pixie. It is named after Aron's dog.
The token problem nobody wants to talk about
Ask why a company can't just point a frontier model at its data and get the same result, and you get the most technically interesting part of the conversation.
"You're going to see a lot of people in the industry saying: that's why I have Claude, that's why I have ChatGPT. I'll just take all my data sources and throw the data at it, and it'll converge the context for me." He paused. "When you're storing that much revenue data, you're talking about terabytes. Imagine throwing Claude at it. You're going to burn tokens, not in months, in minutes."
The context window problem is the other half. You cannot feed billions of tokens of enterprise history into a prompt, and even if you could, you shouldn't pay for it every time someone asks a question.
SciFin's answer is a layer it calls context cognition: machine learning that continuously maintains a compressed understanding of what sits underneath. Fifteen emails between a rep and a buyer become one extracted conclusion. Forecast roll-ups are maintained as numbers rather than recomputed from raw activity. Hundreds of these scenarios, precomputed and kept current.
When I suggested it sounded like a compiler, he took the analogy further. "Exactly. We're creating the bytecode. If it's Java, we're already creating the bytecode. It's machine-understandable. It's not code anymore, it's ones and zeros."
The commercial consequence is subtle but real: because the expensive reasoning has already been done, the question-answering layer can run on the cheapest models available. Pixie can use Claude, GPT or Gemini, or a low-cost open model, and still return, in his words, "very accurate, powerful, deep and consistent answers." Whether that holds at scale is the kind of claim a customer verifies in a pilot. But it is the right claim to be making in a year when AI token bills have become a line item CFOs actually read.
Closing the Context Gap with Context Convergence
On September 10, the company announced Context Convergence. Convergence, as Badrinarayanan frames it, solves the Context Gap, which is made up of two things at once: what he calls reality divergence, where the data no longer reflects real activity, and fragmentation, where the context that does exist is scattered across tools. "That's the category-defining narrative we'll keep feeding the market with."
Strip away the vocabulary and the bet is this: buyers of revenue software have always bought applications. A forecast. A call recorder. A dashboard. SciFin is asking them to buy a layer, and to trust that the applications on top will be good enough to retire the ones they already have.
Aron has won that bet three times, but with IT buyers, who think in architecture. This time the buyer is a CRO or a head of RevOps, who thinks in quarters. That is the real GTM problem SciFin has to solve, and it is why the company's first moves are so disciplined: one category, one buyer, a few hundred accounts, and a narrative built to make a data-hygiene problem sound like the strategic risk it actually is.
The counterargument is that the incumbents have been promising unified revenue data for a decade, and that breadth of integration is not the same as depth in any single workflow. SciFin's response, in effect, is that nobody fixed the input. Until the capture is automatic, every layer above it inherits the same gaps.
SciFin will be at Dreamforce this week, booth 308. Whether that argument lands with enterprise buyers is a question that starts getting answered this week, in a booth at the Moscone Center.
Three things growth teams can take from this
Fix the input before you build another layer on it. Every forecasting, coaching and reporting tool in the revenue stack inherits whatever the rep typed into a form, or failed to type. SciFin's whole argument is that no analytics layer can outperform its capture layer. Before buying the next insight product, audit what actually reaches the CRM and how.
Sitting on top of the stack is a go-to-market position, not just an architecture. The AI-native competitors are asking a CRO to replace revenue operations mid-quarter. SciFin lets a customer live in both worlds during the transition. When your buyer's risk is operational continuity, being additive beats being better.
Precompute the expensive reasoning, then let cheap models answer. SciFin does the heavy machine learning once, continuously, and keeps a compressed representation current, so the question-answering layer can run on whatever model is cheapest. If your product's margin depends on token spend, the leverage is in what you compute ahead of the question, not in which model you pick.