RevOps Just Became the Most Important Function in Revenue. Here's Why.

TL;DR
- Gordon Ritter's Above the Model argues the durable edge in the most automated decade in business is how your people think. In revenue, that's GTM judgment: the tacit calls your best reps make, which walk out the door when they leave.
- Capturing it takes more than an LLM. Loop captures the traces and trains models on your own won and lost deals (regression, XGBoost, KNN, and graph models), then Olli, the default agent in Loop, and your other agents run the plays and feed the results back, so the loop closes on every deal.
- RevOps already sits where people, process, and systems meet, which makes it the engineering function for revenue. Ritter's eight-point platform scorecard is the rubric for whatever you build or buy.
RevOps Just Became the Most Important Function in Revenue. Here's Why.
Gordon Ritter, the founder of Emergence Capital, published a piece last month called Above the Model. In it, he makes a case that runs counter to most of what the AI industry has been saying:
The most automated decade in business history is less about rewarding companies that replace their people. It's more about capturing how their people think.
He's right. And if you run RevOps, it's your new job description, in short.
It's what we fundamentally believe and have built around for the last 4 years, too: the most valuable thing in a revenue org isn't a tool, a process doc, or even an individual "closer." It's the accumulated judgment of how your team sells.
Here are eight takeaways we'll walk through:
- Rep knowledge is the most perishable thing in revenue, and it doesn't have to be.
- Your sellers aren't leaking data into public AI tools. They're leaking process.
- "Above the model" has a specific meaning in selling: we'll show you where it runs in a sales cycle.
- AI is not just LLMs. The market has gotten dangerously narrow in how it defines intelligence.
- A flywheel that doesn't complete its loop is just potential energy. The pieces exist... but they're not flying yet.
- Ritter published the platform scorecard. Eight requirements for the layer that compounds. We'll grade ourselves against every one.
- This has to be a vertical vs. horizontal system design. Horizontal tools can't capture tacit selling knowledge.
- RevOps is the engineering function for revenue.
Let's get into it.
1. Rep knowledge is the most perishable thing in revenue
"The durable layer is how your people think, decide, and work together, captured and sustained within the company."
– Gordon Ritter, Above the Model
You've lived this. Your best AE leaves. They take with them instincts about which deals are real and which are theater, when to push a champion wider and when to hold, how to read a procurement stall caused by budget versus one caused by politics. None of it is written down. Not in your CRM. Not in your playbook. Not in the call recordings nobody rewatches.
What makes it urgent now: it no longer has to be this way.
Think about your last QBR. Someone told a story about saving a deal by going around a blocker to a different stakeholder. Everyone nodded. By next quarter, that insight is gone. The rep might be too. Now imagine that pattern, going around a blocker at a specific stage when certain signals are present, gets captured and surfaced to every rep facing a similar situation. Not as a training video. As a recommendation inside the deal they're working right now.
14% of reps generate 80% of revenue. What they know, nobody else does. The institutional selling intelligence of your organization is the most valuable thing inside revenue. It far outlasts anyone's tenure. RevOps is the natural owner of the infrastructure that captures and compounds it.
2. Your sellers aren't leaking data. They're leaking process.
"The exposure most leaders worry about is data, the sensitive file leaving the building. The subtler exposure is process."
– Gordon Ritter, Above the Model
In revenue, this exposure is worse than almost anywhere else. Reps are running deal strategy through ChatGPT. Pasting in competitive positioning, pricing frameworks, negotiation logic. Building business cases that encode your entire commercial approach. And 77% of employees paste data into generative AI tools, with 82% of that running through personal accounts nobody manages.
| What leaders worry about | What actually leaks |
|---|---|
| Sensitive files and documents | How your team sequences a deal |
| Customer lists and pricing sheets | Your competitive positioning logic |
| Internal communications | The decision framework your best reps use |
| PII and compliance violations | The coaching patterns your managers deploy |
A multi-step AI session doesn't just share information. It encodes your sequencing, priorities, and decision logic. RevOps owns the systems where this happens. Not by locking tools down (your reps will just use personal accounts). By building internal infrastructure that makes public tools unnecessary for the high-judgment work.
3. "Above the model" has a specific meaning in selling
"AI is extraordinary at optimization. What it does not do is decide which goal is worth pursuing or make the judgment call in the moment when the model has no answer."
– Gordon Ritter, Above the Model
Ritter draws a line between optimization work (below the model) and judgment work (above it). In go-to-market, that line runs right through a deal cycle. Get it wrong and you automate the stuff that doesn't matter while ignoring the stuff that does.
| Below the model (automate it) | Above the model (capture it) |
|---|---|
| CRM hygiene and enrichment | Reading whether a champion is truly bought in or just being polite |
| Activity capture and logging | Coaching a champion through resistance you can't see |
| Forecast roll-ups and pipeline math | Changing the narrative mid-deal when a competitor shifts eval criteria |
| Meeting summaries and follow-up drafts | Knowing when to walk from a deal that looks good but will churn |
| Lead scoring and routing | Recognizing this deal is on the same death spiral as three from last quarter |
Everything on the left will be automated soon. Everyone will have it. Table stakes.
Everything on the right is where revenue is actually won. We call it GTM judgment. The job isn't to automate judgment. It's to build infrastructure that frees sellers to live above the line and captures what they do there.
4. AI is not just LLMs
"The traces that encode how your analysts build a model, how your operators make a call, how your team actually decides, belong to whoever builds the system to capture them."
– Gordon Ritter, Above the Model
When most people say "AI for sales," they mean an LLM copilot that drafts emails or summarizes calls. That's useful. It's also a fraction of what's required for the compounding system Ritter describes.
We wrote a full crash course on this because it matters that much. Short version: AI and LLM are not synonyms. It's like calling every animal a cat. The trace flywheel requires purpose-built models across the broader ML landscape:
- Regression (linear, logistic) for calculating explanatory weight. You have 15 deal attributes. Regression tells you stakeholder count contributes 34% of the explanatory weight against closed-won. Stage 2 duration, 22%. Whether the rep sent a mutual action plan, 3%. Now you know where to focus.
- XGBoost / decision trees for scoring. Not one tree. Hundreds of sequential trees, each correcting the last. A new deal enters pipeline and gets run through 500 rounds of gradient-boosted decisions. Output isn't "this deal feels good." It's a mathematically validated weight for each variable against the outcome.
- KNN for pattern matching. A deal hits Stage 2. KNN asks: of the 4,000 deals we've seen, which 15 look most like this one across size, industry, stakeholder count, velocity, and engagement? What happened to them?
- Graph models for mapping how influence flows through a buying committee.
- LLMs for generation and reasoning, but only after the hard math upstream has decided what to say and why.
This is why most "AI for sales" tools plateau. They bolt an LLM onto a workflow and call it intelligence. It's renting a supercar to do the school run. When we were proving this approach, we studied 150,000+ complex B2B deal cycles with this full stack working together. Which is the same stack Loop runs today, except the models train only on your own won and lost deals, and retrain daily as those deals close (it's what sits behind Loop's predictors).
5. A flywheel that doesn't complete its loop is just potential energy
"Better performance drives more use, more use produces more traces, and the gap widens with every cycle."
– Gordon Ritter, Above the Model
The pieces of Ritter's compounding flywheel already exist in most revenue orgs. Somewhere you're capturing interactions. Somewhere you have past deal data. Somewhere you have a way to surface insights. But those pieces are scattered across a dozen tools that don't talk to each other.
A flywheel that doesn't loop is potential energy that never becomes kinetic. This is why we took a platform approach:
| Layer | What it does | Why the flywheel needs it |
|---|---|---|
| Loop | Context engineering and revenue attribution. Structures every interaction into a signal graph. | Without this, traces never get structured into anything a model can train on. |
| Your private models | Trained only on your own won and lost deals, retrained daily as deals close. Regression, XGBoost, KNN, and graph models scoring every variable against the revenue outcome. | Without this, structured data never compounds into predictive insight. |
| Olli + your agents | Olli, the default agent in Loop, and any agent you connect run the plays and put intelligence into sellers' hands at the moment it matters. Every accept, reject, or modify flows back through Loop. | Without this, the feedback loop never closes and the system never learns from itself. |
These aren't three products bolted together, they're one loop: Loop captures the traces and trains models on your own deals, and Olli and your other agents run the plays and feed the results back. Remove any segment and the wheel stops. That's the difference between a platform and a stack of disconnected tools paying the orchestration tax.
6. Ritter published the platform scorecard
Ritter's piece includes eight requirements for the platform layer that sits underneath any agent. Not features. Architectural principles. They read like a grading rubric, so here's our honest self-grade:
| # | Ritter's requirement | How Fluint is built against it |
|---|---|---|
| 01 | Effortless to use. Friction and context switching kill adoption. | Olli and your agents live inside the deal workflow. No separate tab, no manual context entry. Loop already knows the deal. |
| 02 | Spread individual knowledge so one person's right answer becomes the default for everyone. | The core job of the models trained on your deals. One rep's breakthrough becomes everyone's starting point. |
| 03 | Real permission boundaries scoped to the task, not the most permissive level someone can get away with. | Private deployment. Your org's data trains your org's model. No cross-customer bleed. |
| 04 | Learn continually from every interaction, not just when an engineer ships an update. | The flywheel from section 5. Every rep interaction flows back through Loop into your models, which retrain daily as deals close. No quarterly retrains. |
| 05 | Keep traces and evaluations as your own property, not inside a vendor's black box. | Loop is the context engineering layer you own. Traces are yours. Evaluations are yours. That's the architecture, not a feature. |
| 06 | Route each task to the model that does it best. No single model wins everything. | The multi-model stack from section 4. Don't send a frontier LLM to count days in a stage. |
| 07 | Use open source models broadly. Less ability and motivation to harvest human innovations. | Where an open model does the job, use it. Save proprietary models for where they genuinely outperform. Security and cost discipline in one. |
| 08 | Track cost honestly. Agentic systems spend fast. | No orchestration tax. Single flywheel, attributable token spend. You see where the money goes before the bill arrives. |
Use this as a scorecard for any platform you evaluate. Not just ours.
7. This has to be vertical
"The traces are structurally exclusive. This is real intellectual property that grows on its own."
– Gordon Ritter, Above the Model
There's a tempting shortcut: take a horizontal AI platform, connect it to your CRM and conversation intelligence tool, build a knowledge graph, call it revenue intelligence. It doesn't work. The reason is structural.
The judgment that matters in selling is tacit. Converting it into explicit signals is a math problem that requires a schema built for how revenue actually works.
The time-series problem horizontal tools miss:
Revenue data isn't a static relational graph. It's a time-series. When things happen carries as much signal as that they happen.
Procurement responds to your proposal at 8 p.m. on a Friday, same day as your executive meeting. A horizontal system logs: "Procurement engaged." It would log the same thing if procurement responded the following Wednesday at 3 p.m. But those are completely different signals. An 8 p.m. Friday response after an exec meeting? Someone went home, kept thinking about it, came back to their laptop. That's urgency. A 3 p.m. Wednesday reply? That's routine.
| What you need to understand | Horizontal platform | Vertical, time-series-aware system |
|---|---|---|
| Champion building consensus | "Multiple stakeholders in calls" | Champion multi-threading into finance and legal in a compressed window, mirroring patterns from the 15 closest analog deals that closed |
| Deal at risk | "Activity decreased" | Activity dropped 72 hours after a technical eval surfaced an integration gap, matching a sequence that preceded 3 losses this quarter |
| Procurement signal | "Procurement engaged" | Same-day 8 p.m. response following exec meeting: timing pattern with 0.87 explanatory weight against closed-won |
Simply mapping nodes in a relational graph (who talked to whom, what was shared) isn't the same as learning from a time-series against an outcome. A knowledge graph tells you your champion talked to the CFO three times. A discriminant model on time-series data tells you CFO engagement has a 0.92 explanatory weight on closed-won in your org, and champion-to-CFO threading drives a 23% ACV lift when it happens before Stage 3. Not just that it happened, but when, how quickly relative to other deal events, and how much of the outcome variance it explains after isolating it from every other variable.
One is a map. The other is turn-by-turn directions that update every time a deal closes.
So Loop trains a private model on each customer's own won and lost deals, and nobody else's. Your weights and data are never shared with another customer, and the model retrains daily as your deals close. Which also takes care of the cold start: your calendar and email history already holds the source data, so you don't need two years of clean CRM fields (or anyone else's deals) to get going. From day one, the AI points inward, compounding exactly the way Ritter describes.
8. RevOps is the engineering function for revenue
"The human being is the mutation engine in the system. The genuinely new ideas, the moves no model could have predicted, come from creative people finding a better way. The system spreads those beneficial mutations to everyone else."
– Gordon Ritter, Above the Model
Ritter first wrote about this in 2017: software that learns what actually works from a distributed network of people and guides them toward doing their jobs better while they're doing them. That's the transition from enablement to engineering. One we tried to reverse-engineer with off-the-shelf tools to see if it could be replicated. It couldn't.
| Enablement (old model) | Engineering (new model) |
|---|---|
| Training content pushed at reps | Infrastructure that learns from reps |
| Static playbooks updated quarterly | Dynamic intelligence updated every deal |
| Knowledge leaves when people leave | Patterns remain in the system |
| RevOps manages the tech stack | RevOps engineers the intelligence layer |
| Success = tool adoption | Success = revenue attributed to agent actions |
Same paradigm shift as platform engineering for developers. DevOps for RevOps. RevOps owns this transition because it already sits at the intersection of people, process, and systems. The job is expanding from managing the stack to engineering the intelligence layer on top of it.
Build the layer. Keep the mutations yours.
Automate everything below the model. Take the free money. But the edge isn't optimization. Everyone's going to have that.
The edge is the judgment that happens above it. Your people generate it every day. The question is whether you're capturing it or letting it evaporate into tools you don't own.
We built Fluint because we believe the answer should be obvious. The platform to complete the flywheel is here. The function best positioned to own it is RevOps. The only question is whether you start now, or wait and try to buy what someone else's traces already taught them.
That's not a gap you close with a purchase order.
FAQ's on:
RevOps as AI infrastructure owner
Why stop now?
You’re on a roll. Keep reading related write-ups:
Deal-winning context that changes the outcome
“Fluint gave us months of AI engineering work with a single integration. Now everything we build is grounded in how we win, with the right context for every agent and seller.”
“Fluint keeps our team on our sales process and calls out the gaps in every deal. Our win rate is up, and it’s helped us 6x the efficiency of our sales organization.”
“No other platform captures the right context in fewer words than Fluint.”
