.png){kind=small}
I recently joined Warren on the Selling AI podcast to talk about something I have been saying to every client for the past year: most revenue operations teams are solving the wrong problem. They are optimising dashboards, cleaning CRM fields, and building automations on top of a data architecture that was never fit for purpose.
The conversation covered a lot of ground, from close-loss data and ICP definition through to signal layering, conversation intelligence, and where AI implementations are quietly destroying value. What follows are the core arguments from that episode, expanded with the context and examples I did not have time to get into on the call.
If you would rather watch the full conversation first, the episode is linked below.
Watch the full conversation on the Selling AI podcast: Watch on YouTube
Ask your team why deals are lost. Most will say "no decision." A few might say "budget." What they won't say is the truth, because the truth requires a system designed to capture it.
Close-loss data is broken everywhere because it's treated as a checkbox instead of a diagnostic tool. When sales logs a loss as "no decision," you've learned nothing. You've documented failure to understand your own funnel. The real signal lives elsewhere: how long did the deal sit in evaluation? Who stopped responding? What gap emerged between your solution and their stated problem?
A scalable system requires close-loss data that's actually useful. That means forcing specificity. It means knowing the difference between "customer wasn't ready" and "customer never wanted what you were selling." These aren't minor distinctions. They're the difference between fixing your pitch and fixing your ICP.
This is where most revenue operations programmes fail. Companies define ICP as the entire addressable market, minus obvious disqualifiers. This is not an ICP. This is a guessing game.
Real ICP is significantly smaller. Not 80% of your market. Closer to 10%. These are the accounts where you solve an urgent problem, where your solution aligns with how they actually work, and where deal economics work in your favour. Everything else is noise.
We worked with a company that defined their ICP based on industry and company size. They were targeting 25,000 accounts. When we layered in signal-driven criteria: companies using specific tools, hiring in certain roles, showing expansion intent within existing customers, the addressable set collapsed to 1,000 accounts. This wasn't a loss. It was a focus tool.
That narrower set closed at a higher rate, faster sales cycles, and significantly better unit economics. They didn't lose market. They found their actual market.
The principle is this: a scalable system requires constraint. Your revenue architecture isn't broken because your target is too small. It's broken because your target is too large.
Single-source intelligence is always incomplete. Your intent data tells you someone's interested. It doesn't tell you whether they can actually buy. Your account intelligence tells you the company is growing. It doesn't tell you whether the pain you solve is their priority right now.
Revenue operations moves from broken to functional when you stop relying on one signal and start building architecture that layers them. Intent plus account expansion plus hiring patterns plus technology stack plus conversation velocity equals a comprehensive picture of purchase readiness.
This isn't automation. This is intelligence. Combining data points to see what one data point alone will never show you. A company is hiring three engineers in the exact team your solution serves. That's signal. They're also visiting your pricing page repeatedly and their main competitor just won an award. That's signal. Now they're also in a contract renewal cycle with their current vendor. That's the full picture.
Layering transforms your system from reactive to predictive. You're not waiting to find out if they're qualified. You're understanding their context before the conversation starts.
Here's the uncomfortable truth most AI vendors won't tell you: automating a broken process just makes it break faster.
Sales teams don't fail because they lack automation. They fail because they lack context. A sales rep with three relevant data points about an account will outperform a fully automated email sequence every day. They'll personalise, adjust timing, know exactly what problem to lead with.
The vendors selling you AI-driven email automation are solving their problem, not yours. They're making it easy to do something at scale. What your team needs is the opposite: harder, fewer, more contextual interactions.
This matters because operational friction isn't always bad. Sometimes friction is a feature. A sales rep forced to think about why they're reaching out develops better intuition. A process that requires actual research generates better conversations. The companies winning right now aren't the ones that automated everything. They're the ones that automated the drudgery and forced thinking back into the important parts.
Meeting transcripts are a source of truth that most organisations completely ignore. A single call between a buyer and your seller reveals more about actual fit than months of intent data.
The powerful move is rolling this up across your entire pipeline. Every call recorded, transcribed, indexed. Not for compliance. For intelligence. What topics come up repeatedly across your best deals? What objections do your best reps overcome? Where does deal momentum actually stall?
This is the real intelligence layer. Not predictive AI that guesses at buyer readiness. But systematic analysis of what's actually happening in the conversations your team is already having.
A breach specialist uses e-discovery tools to roll up conversations across millions of emails. They find patterns invisible at the single-conversation level. Sales organisations can use the same principle with meeting intelligence. It's not about recording reps. It's about building a system that learns from what works.
This is the contrarian take that most growth teams won't admit: AI implementations in sales operations are frequently value-destructive.
Not all of them. But many. Here's why: they optimise for speed instead of quality. They automate at the wrong layer. They create dependencies on platforms that don't integrate cleanly with your actual source of truth.
A company implements AI-driven lead scoring to "accelerate qualification." It saves time in the sales team's calendar. But it also creates misalignment between what sales thinks is qualified and what revenue operations measured as qualified. Now you've got duplicate work, manual overrides, and a system that's actually slower than before.
Another example: a company deploys AI email generation to "personalise at scale." The tool learns from your best templates. What it actually learns is your average templates. Sales gets faster at sending more emails. Conversion rates stay flat. You've converted a constraint into a scale problem.
The companies getting real value from AI in sales aren't the ones buying AI for AI's sake. They're the ones solving operational friction first, understanding their system, and then applying AI to the specific problem they've diagnosed. That's the opposite of most AI sales implementations.
This matters because building a revenue architecture that actually works requires working with your sales team, not against them. Most revenue operations initiatives fail because they're implemented as constraint. New process. New approval gate. New data you have to log.
The ones that work are designed with sales. What information do you actually need to be better at your job? What's friction you'd remove if you could? Where are you making decisions you wish you had better data for?
Your sales team is the only people in the organisation who understand your revenue model at operational depth. Ignoring their input while building your system is ignoring your best source of truth.
If you're building towards scalable systems, these are the questions worth asking right now:
1. Is your close-loss data specific enough to be useful, or are you documenting failure to understand?
2. Is your ICP actually an ICP, or is it the entire market minus the obvious disqualifiers?
3. Are you building intelligence from single signals, or layering data to build actual context?
4. Where are you automating instead of contextualising?
5. What conversations are happening in your pipeline that you're not systematically learning from?
6. Which AI implementations are solving real problems, and which are just making broken things faster?
7. Is your revenue operations team building systems for sales, or constraints on sales?
These questions aren't theoretical. They're operational. They determine whether you're compounding revenue or just compounding complexity.
If your revenue operations system feels like it's fighting you instead of enabling you, the problem isn't that you need more tools. It's that your system is built wrong. Start by understanding what's actually broken. Build signal-driven architecture from there.
%20(1).jpeg)