The Most Important Process in Your Company Has Never Been Written Down

Ask a good salesperson how they knew a deal was slipping. They'll tell you something like: the champion took two days longer to reply than usual. The CFO got cc'd on a thread where she'd never appeared before. Someone mentioned a competitor in passing, in a sentence about something else entirely. None of these things means anything on its own. Together they meant everything, and this salesperson noticed.

Now ask where that process lives in your CRM... It just doesn't.

Your CRM has a pipeline with six stages, a probability field somebody fills in on Fridays, and a "next step" box that says "follow up". The actual process — the one that produces revenue — happened in someone's head, in about four seconds, and left no trace anywhere.

I've spent a long time thinking about this gap, and I've come to believe that it's the thing that will actually change about how companies run over the next few years. Not "AI in the CRM". Something quieter and larger than that.

We've been documenting the wrong processes

Every company has a stack of documented processes: how a lead becomes an opportunity, how a quote gets approved, how an order becomes an invoice. These are useful. They're also the boring part. They describe how information gets routed, not how it gets understood.

The processes that actually matter are the ones nobody documented because nobody could. How does a great account manager decide which of forty open threads to touch today? How does a founder sense that a partnership is going cold before anyone has said so? These are reasoning habits, not workflows. We've called them "intuition" or "experience" mostly because we had no way to compute them, so we filed them under things that can't be computed.

I don't think that's true anymore. But I also don't think the answer is what most people expect.

A borrowed idea from physics

Stephen Wolfram has a concept he calls "computational reducibility". Roughly it means: most of the world is too complex to shortcut — you have to run the whole thing to see what happens at the end. But scattered through that complexity are pockets where a simple rule predicts the outcome. Science exists because we found those pockets as "laws of physics". Weather is mostly irreducible; the fact that water boils at a hundred degrees is a pocket.

A salesperson's judgement is, taken as a whole, irreducible. It's situational, relational, built from twenty years of being in meeting rooms. You cannot write it down as a rulebook and you shouldn't try.

But inside it there are pockets. "Disputed invoice, plus champion silent for three weeks, plus a competitor's name in a forwarded email" is a pocket. Each signal is faint. The combination is a pattern that reliably predicts what the experienced human would conclude. And a pattern like that can be found, tested and, eventually, run by a machine — cheaply, consistently, on every account at once, at three in the morning. Thanks to AI tech ...

That's the shift. We're no longer designing business processes. We're prospecting for pockets of reducibility in how our best people think, and turning the ones we find into something the company can run.

Where the record comes from now

Today the record comes first. A human types in the client, the contact, the deal, the stage — and then, maybe, something is done with it. The typing is the tax you pay to have a memory.

In the version we're building, the record is a by-product. The system reads what's already happening — the mailboxes, the chats, the shared drives, the meeting notes — and derives the clients, the contacts, the deals, the tickets, the risks, the opportunities. Nobody types. The record assembles itself, and it stays alive: recomputed when new evidence arrives, carrying its own reasoning so you can see why it believes what it believes.

Some records still have to be frozen. An order, an invoice number, a contract — those are commitments, and a commitment can't quietly change its mind overnight. So there's a distinction worth being clear about: living records the system keeps deriving, and frozen ones a person (or a rule) has crystallized into a commitment. Legacy tools only had frozen records and made humans do all the crystallizing by hand. That's most of what "CRM hygiene" ever was.

What the humans are for

The point is not to replace the salesperson with a model that "thinks about everything" on its own. That's the easy, expensive, wrong version.

A system that reasons from scratch over every email is slow, costly, and gives you a slightly different answer each time you ask. It also can't explain itself, which means nobody trusts it, which means nobody acts on it.

The discipline is the opposite. Find the pockets. Make them cheap, deterministic, and legible. Run those constantly. And bring the human in exactly at the boundary where the pocket ends — where the situation is new, political, or ambiguous enough that judgement is the only tool that works. Humans close the loop in human to human conversations making final decisions.

A lot of what a great salesperson knows never touched a recorded channel. It happened at dinner. The system computes over what leaks into communication — which is more than most people think, but not everything. This is instrumentation, not replacement.

What this does to running a company

Wolfram makes one more point that I keep coming back to: which pockets you can see depends on what you're looking for. Reducibility is relative to the observer.

For a company, that means something concrete. Deciding which signals count, which combinations compound into something worth acting on, and where the line sits between "the system handles this" and "a person decides" — that's no longer a configuration task. It's the operating model. Management stops being the design of flows and becomes the curation of how the company reasons.

For decades we've asked people to be the sensors and the filing clerks of their own organisation, and then wondered why the tools were hated. If the machine can hold the memory and spot the patterns, the humans get to do the part they were actually hired for.

Conclusion

The most important process in your company has never been written down. I think we're about to find out what happens when it can be.