Strategy as Software

The problem it exists for

Every quarter, a board reads a pack that says the business is on track. The pack is assembled by the people whose performance it describes. This is not fraud; it is arithmetic. Nobody writes their own report card in a spirit of forensic self-criticism, and nobody has the time to read four hundred pages of budget lines, hiring approvals and contract renewals against a strategy document from eighteen months ago to see whether the two still agree.

So drift happens the way drift always happens — one reasonable decision at a time, none of them wrong on the day, all of them quietly pulling away from the plan. A concentration limit written as 20% becomes 26% because each individual deal was a good deal. Nobody lied. Nobody noticed either.

The evidence was always there. It just sat in different documents, read by different people, none of whom were reading it against the plan.

What it actually does

The strategy goes in as typed claims rather than prose — an objective, a hard constraint, a ranked priority, an assumption — each with the kind of evidence that would test it. The operating documents go in as sources: budgets, board packs, org charts, hiring plans, contracts, the reporting pack.

Then it runs. Coverage first, which sorts every claim into evidenced, unevidenced or untestable and produces a request list for the gaps. Observation mining over the evidenced ones. Pairing, evaluation, mechanical citation binding, a blind second opinion, and a verdict lattice that knows about documented exceptions and takes recurrence seriously. What comes out is a small number of candidate findings — at most five on the front page, each novel, quantified or recurrent — with the strategy health index, the drift since last quarter, the aspiration pile, and an annex listing the exact filenames, versions and checksums the report was built on.

The whole thing costs single-digit dollars of model spend per run, which is a fact best kept in the margin column rather than the pitch, because it invites exactly the wrong comparison.

What it deliberately is not

It is not AI that writes your strategy. That product would be easy to sell, briefly, and impossible to defend the first time a board followed it off a cliff. The machine generates candidates; a named human promotes them to conclusions; the record shows who did which.

It is also not a general assistant with a board pack pasted into it. That is fast, already paid for, and completely unaccountable — no typed model of the strategy, no citation binding, no blind second check, no memory of last quarter’s explanation, and the documents leave the building on their way to being helpful.

Why I built it

I have sat on both sides of this. As a CTO watching a strategy arrive at the work as a translation of a translation of a translation, and as a founder writing the pack that a board would read as though it were an objective account of reality. The most useful thing anyone ever did for me was disagree with the pack using evidence I had supplied myself.

That is a service, and it does not scale, and it depends entirely on there being someone in the room brave enough and free enough to do it. Encoding it was the obvious move. The essay On Track, According to the People Being Judged is the longer argument.

Status

Beta. Stage one is functionally complete and has run live against a real model on a synthetic client seeded with planted contradictions and deliberate traps: five of five contradictions found, zero traps taken — a documented exception surfaced as a caveat, an untestable aspiration flagged as untestable, neither reported as a breach. Eighty-five automated tests, an eval gate that blocks any prompt change failing precision and trap thresholds.

Phase 0 is three manual engagements — a PE-backed business, an incoming CEO, an owner-managed company — with a written kill criterion attached: if three engagements produce nothing the CEO did not already know, the premise is wrong and the product should stop.

There is a public overview and a readiness check if you want to see whether your own organisation has the documents this needs before anyone talks to anyone.

On the name

“Strategy as Software” is the thesis, and it is a terrible product name. It tells a sceptical chief executive that software is now running their company, which is both untrue and precisely the objection you least want to hand someone for free. A CFO has to be able to say the name out loud in a board meeting without sounding like they have handed over the keys. The naming sprint is on the list, somewhere below “find out whether the thing works”, which feels like the right order.