The Latent Pipeline

B2B sales is circular, not linear

Everyone draws the pipeline as a funnel: leads pour in the top, deals fall out the bottom, and the job is to widen the top and reduce the leaks. It's a useful picture for the 20% of your pipeline that's already hot. For the other 80%, it's misleading.

Most of your winnable revenue isn't in the deals that are closing this quarter. It's in the ones that stalled, went quiet, got deprioritized, or never quite started – the latent pipeline. That part of the business isn't a funnel at all. It's a set of relationships that keep changing, where nothing is ever truly lost and nothing stays won, and where the real work is deciding, over and over, what deserves a move next.

My claim is narrow: managing the latent pipeline is fundamentally a summarization and prioritization problem – one almost every company has left unsolved, because its difficulty compounds exponentially in a way that neither your CRM nor an agent turned loose on it can handle on its own. That's the problem I'm building for.

What a "deal" actually is

A "deal" isn't a real thing. It's a construct that joins at least two real things: a target company and at least one person inside it. Your pipeline is the sum of all those constructs. At that conceptual level every B2B company works the same way – no matter how the CRM ends up recording it.

Where does any given deal actually live? Rarely anywhere useful in the CRM. It's spread across inboxes, call notes, someone's memory, and history. Most deals are latent most of the time – and you'll know when they are not.

For any deal in the latent pipeline, there are only two valuable outcomes:

1. Keep it alive – make the next move that keeps it heading toward the hot zone.
2. Disqualify it – decide, honestly, that it isn't worth attention right now, and stop spending on it.

Everything else is drift. And drift is expensive, because attention is the scarce resource, not leads.

Why following up well is genuinely hard

This isn't unsolved everywhere. The best-run companies have quietly solved it – and, tellingly, the rest haven't copied them, even though none of it is secret. It resists broad adoption because it's hard to build and harder to sustain, not because nobody knows how. Here's what makes it hard:

• No deal is ever truly lost. The reflex is "I need better leads." You almost never do – you already have more routes into more accounts than you can work, especially in enterprise, where there are tens or hundreds of ways into a single company. Unlimited routes sound like good news; in practice they produce maximum confusion, because nothing forces a deal closed and everything stays half-open forever. Today that gets sorted by gut feel – and gut feel good enough to trust is rare and expensive.
• No two deals are equal. To prioritize honestly you have to compare deals that aren't comparable – different sizes, stages, people, and odds. That's cheap to eyeball for an early-stage deal and genuinely hard the further down the pipeline you go.
• Every action changes reality. Take Sarah, the seller who owns an account, and Jerry, a prospect inside it. Every message either of them sends opens a loop with its own set of possible next moves. One deal like that is trivial to manage. The reality is Sarah has dozens, each with several stakeholders, and every action reshuffles what the best next move is across all of them.
• Complexity compounds. Put those together and the decision space grows exponentially with pipeline size – with or without AI. Turn an agent loose on the whole CRM and it's expensive and mostly produces noise; it only works behind guardrails: intermittent data checkpoints, compression, and firm guidance at the steps that matter.
• Your CRM was built for humans, not for this. Stage models, rules of engagement, thresholds, checklists – all of it exists so a person can eyeball where a deal stands. It's a useful simplification, but it doesn't do justice to what's actually going on inside a deal, and it was never designed to drive the next move across hundreds of them at once.

None of this means there are no great sellers who keep their deals under control – there are. But getting that consistency across a whole organization is hard: it takes people who are exceptional at running processes and have every other trait of a strong seller. Those people are rare and expensive, so they're reserved for the hot part of the pipeline – and even they get more done when the system behind them is good.

The limits of AI judgment

Needless to say, I use AI for most of what I build for clients – and you probably do too. It's astonishing on precise problems: code compiles or it doesn't, a translation is faithful or it isn't. Relationships aren't precise. Reading a deal is a matter of taste and judgment, and that taste can be encoded – in prompts, configurations, and the data the system reads during onboarding and daily operation – but doing it well is a system-design problem, not a weekend of vibe-coding (which I tried).

So the rule isn't "automate the relationship." Automate most of acquisition and parts of the pipeline work; but the moment a relationship gets serious, AI is a heavily-guided helper at most, and never the thing that decides the exact words that go to a customer. The human stays the first-class, final decision-maker.

Point an AI at this without that structure and you hit what I call the AI Double Trap – it catches you twice:

• The output looks good. Models are trained to produce confident, fluent answers, so every response lands like progress even when you're far from done or a bad assumption just slipped in. That's the trap you can see.
• The model tunnels. A model knows an enormous amount in general – it read most of the Internet, after all – but the moment you put it on a track it commits to that track and stops surfacing everything else it knows. Type "pink elephant" and it will happily discuss pink elephants; it won't spontaneously raise what pink elephants mean for the risk in your sales process. Aimed at a live pipeline, that same tunnel vision quietly drops the considerations that matter most.

Escaping the trap isn't about picking a better model. It's about surrounding the model with enough structure – frameworks, architecture, checkpoints – that its judgment stays constrained, checkable, and subordinate to a human's. That's the whole game, and the reason this stays hard even as the models keep improving.

The answer is circularity

If the problem is exponential decision-making over a system that never stops moving, the answer isn't a smarter one-shot. It's to close loops, on repeat.

Concretely, doing it well takes four things working together:

• Frameworks – a map of your specific revenue engine, so "what matters" is defined for you, not in the abstract.
• Architecture – the wiring that holds the data and the loops together at scale without drifting. The data doesn't fit in a context window; some of it wants to be a graph.
• Taste – your judgment encoded during onboarding, plus a human deliberately kept in the loop. Part of taste is unglamorous: every CRM is set up differently and keeps changing, so anything automated has to get a fresh, correct read of the data model every single time. That detailed understanding gets captured once, in config prompts, in the places I know it's needed – and reused by everything downstream that depends on it.
• Circularity – run it continuously, so every closed loop feeds the next decision instead of resetting to zero.

Underneath all four sit the two primitives from the start: summarize the context down to what matters, then prioritize the next move honestly – and repeat. This is also why your CRM's built-in AI overview falls short: good summarization gets complex fast, and prioritization is only ever as good as the summary underneath it.

Get those two right and something useful falls out almost for free: the forecast. Once every deal is honestly summarized and consistently prioritized, forecasting revenue is mostly a read on data you already trust – the hard upstream work is what makes the number believable in the first place. To be clear, that's a consequence, not my offer. Forecasting can be done a dozen ways and plenty of tools already do it; I'm not selling a forecast. I'm pointing out that a trustworthy one is a symptom of having the summarization and prioritization right underneath it.

Anyone can build this

Let me be straight: none of this is secret, and there's no trick I'm guarding. Any competent team can build it. That's the honest starting point – and it's also how I tell the real work apart from the noise.

Two kinds of noise, specifically. The first is whoever – an internal champion or an outside vendor – tells you this is a weekend of "skills," or whatever the current fashion is. The ideas are simple, so it sounds plausible; but the implementation is where the whole thing actually lives. The second is the plugin that promises to "win your deals faster" – a smaller, different claim than getting the latent pipeline under control. Easy to think through, hard to implement well: that gap is the entire game.

I'll be just as honest about how the building goes. The first attempt never lands perfectly. I wire up the biggest bottleneck, watch it against real data, and iterate until it genuinely holds – and only then move on to something more complicated. And it isn't "buy module A, B, and C." It's taking the philosophy in this piece and adapting it to your biggest constraints, in the way that's right for you.

So build it yourself if you can – genuinely. What matters isn't who builds it; it's that the loops that shouldn't be open get closed. If reading this sends you off to do it internally, good. If it makes you realize it's too far from your core to build well and you can't get it elsewhere, let's talk.

What this is not about

To be precise about the boundary, because the word "sales" attracts the wrong associations:

This is not top-of-funnel or lead generation – that's largely solved and commoditized. Not a "close faster" tool. Not GTM advice or a fractional exec. Not a data vendor. Not a CRM plugin. It's the hard, unglamorous middle: the latent pipeline, and the judgment layer that runs it.

Where I've watched this play out

I didn't reason my way to this from a whiteboard. I've built pieces of it under fire, and the wins came from judgment calls about what to read and what to ignore – not from working harder.

A US tax-credit firm. Its collections floor chases money through two failure points: the IRS paying the client, then the client paying the firm. The leverage wasn't a nicer dashboard. It was refusing to build on the CRM everyone assumed was the source of truth, anchoring every money-critical field to the system that was actually correct, and shipping per-client summaries before the dashboard. What that bought the operator: roughly $2.7M in fees and refunds made visible to act on, caller prep cut from 15+ minutes to about 30 seconds, and a scored baseline across 10,000+ graded calls where there had only been anecdote. The CEO stopped managing by what the CRM said and started managing by what was true.

A Series A B2B SaaS startup. The real constraint wasn't pipeline volume – it was the founder's bandwidth, and the politically expensive call was to trust the data over his hunches when the two diverged. Betting on the quality of scoring and personalization instead of adding channels or reps produced 4× closed revenue in four months and lifted reply rates from ~10% to over 30% – without hiring another rep.

Different companies, same shape: summarize what matters, prioritize the next move honestly, and the numbers a CEO actually underwrites – cash visible, revenue closed, capacity not hired – follow from that.

Coda: never won, never lost

I keep coming back to the ensō – the circle drawn in a single brushstroke, complete and open at once. A deal is like that. It's never permanently won and never permanently lost; it's open, or it's closed for now, and the work is to keep making the next honest move around the circle.

Most companies leave most of that circle unattended. That's the opportunity, and it's the thing I want to build for someone who feels it as sharply as I do.

If that's you, see how I work or book a conversation.