On Track, According to the People Being Judged

A new chief executive inherits a strategy they did not write and a reporting system built by the people it evaluates. An investor backs a plan they will never watch being executed. Both are told the same thing every quarter, and neither of them can check it.

Picture the first Monday.

You are the new chief executive. There is a laptop that does not have your fingerprint on it yet, a calendar somebody else filled in, and a shared drive containing eleven thousand documents, of which four matter and nobody can tell you which four. On the desk is the strategy. Forty-one slides, approved eighteen months ago by a board that has since changed by half, written in the confident future tense that all strategy documents are written in. We will become the trusted partner of choice. We will expand into adjacent segments. We will maintain disciplined capital allocation.

And on top of it, the most recent board pack, which reports — with a small green tick beside each pillar — that the business is on track.

On track according to whom, exactly?

According to the people whose bonuses depend on it being on track. Not because they are dishonest. Because they are human, and because you have handed them the job of writing their own report card and then given them no time at all to write it. Even a saint, given a template with a traffic-light column and forty minutes before the deadline, will find a shade of amber they can live with.

This is the position every incoming chief executive starts from, and the position every investor occupies permanently. You are being told the state of a system by the system itself. And the strange part — the part that ought to be scandalous and is instead simply normal — is that the evidence which would settle the question already exists, in the building, in files you already own, and nobody is reading it against the plan.

The pack is a self-portrait

Let me be precise about the failure, because it is not lying and treating it as lying gets you nowhere.

A board pack is an act of compression. Twelve weeks of a company’s life, several thousand decisions, tens of thousands of transactions, reduced to thirty pages by a small team who must decide what matters. Every act of compression is an act of judgement, and every act of judgement made by an interested party bends, very slightly, in the direction of the interest. Nobody deletes the bad number. Somebody chooses which chart it appears on, next to which comparison, under which heading, with which footnote. Do that four times a year for three years and you have a document that is entirely true and systematically flattering.

Meanwhile the strategy sits in a different file, in a different format, written in a vocabulary the pack does not use. The plan says disciplined capital allocation. The pack says EBITDA bridge. Nobody has ever put the two documents on the same desk at the same time and asked whether the second is evidence for or against the first, because that is an unpleasant, unrewarded, forty-hour job requiring someone senior enough to understand both and junior enough to have forty hours.

So drift happens the way drift always happens: one reasonable decision at a time. Take a concentration limit — the plan says no single client above twenty per cent of revenue, and it says so for a good reason that everyone has since forgotten. Then a client wants more work. It is good work, at good margin, from a client who already trusts you, and turning it down would be strange. That happens four times over two years. Nobody ever approved breaching the limit. Nobody had to. The number is at twenty-six per cent, the pack still says on track, and the sentence about concentration is still in the strategy because nobody deletes sentences from strategies. They just stop being true and carry on being printed.

You will find this out. The only variable is whether you find it out in month two or in the month the client leaves.

Three things a new chief executive cannot do

The honeymoon is not a period of grace. It is a short window of unusual permission, and it closes.

In it, you cannot read everything. The evidence is genuinely there — budgets, hiring approvals, contract renewals, the reporting pack, the org chart, the last three years of board minutes — and it is genuinely too much. Four hundred pages a quarter for three years is not a reading problem, it is a career.

You cannot ask freely, either, which is the part nobody warns you about. Every question a new chief executive asks is a signal, and the organisation is reading you far more carefully than you are reading it. Ask twice about the concentration limit and by Thursday there is a workstream, a slide and a nervous director. You have a limited budget of questions before you have accidentally announced a strategy, and you must spend it before you know which questions are the good ones.

And you cannot trust the answer, in the specific sense that you cannot distinguish a confident answer from a correct one. The people telling you things are competent, plausible and have been here for years. That is exactly what makes it hard. A liar you could catch. A well-meaning colleague reproducing a shared belief that stopped being true in 2024 is invisible, and there are eight of them, and they agree with each other, which you will mistake for corroboration.

Put those three together and you get the standard outcome. The new chief executive spends ninety days conducting a listening exercise, arrives at a view assembled largely from what the confident people said, announces a strategy that is the old strategy with new adjectives, and discovers the real problem in month fourteen, by which point they own it.

What the investor is actually buying

Swap seats. You are writing the cheque instead of receiving it.

You did financial diligence, which read the past and read it well. You did commercial diligence, which read the market and produced a projection with a defensible-looking wedge in it. Between the audited past and the projected future sits the thing that decides the outcome — whether this company, with these people, can actually do what the plan says — and there is no established discipline for reading that, no line item for it, and often nobody on the cap table who could.

Then the money goes in, and your instrument for knowing what happens next is a quarterly pack written by the management team you just funded.

I have watched this from close enough to still feel slightly ill about it. A friend of mine — one of the best engineers I know, and he would tell you himself he is not a businessman — took a six-figure cheque and was left to it. Scope crept, because building the more interesting thing is what engineers do. The market moved, because markets do that when nobody whose job it is to watch is watching. The post-mortem records ran out of cash, which is true and almost entirely beside the point. The cash ran out because of eighteen months of small invisible things, and the bank balance was simply the last of them and the only one anyone wrote down.

Nothing in the data room would have told you. Everything in the following eighteen months would have, if anyone had been reading it against the plan.

The uncomfortable symmetry is this: the investor and the incoming chief executive have the same problem, which is that they are downstream of a narrative produced by the people being evaluated. The chief executive at least gets to walk the corridors. The investor gets the pack, four times a year, and a phone call in which everyone is upbeat.

The money is the strategy; everything else is commentary

Here is the thing that makes the problem tractable, and it is the whole reason I stopped complaining about this and started building.

A company cannot hide what it does. It can present it, frame it, footnote it and bury it on slide thirty-one, but it cannot avoid doing it, and doing it leaves paper. Budgets are strategy with numbers attached. Hiring plans are strategy with names attached. Contract renewals, supplier terms, the capital-expenditure schedule, which teams grew and which quietly did not get replaced — these are the strategy the company is genuinely executing, as opposed to the one it approved. Written down, dated, and already in your possession.

The gap between those two documents is measurable. It is just nobody’s job to measure it.

Which is a very odd sentence to write in 2026, because reading several hundred documents and checking them against a list of claims is now precisely the sort of tedium a machine is good at. Not the judgement — the judgement is still yours, and should stay yours — but the finding. This sentence in the strategy. This line in the budget. They do not agree, and here are both, quoted, with page numbers.

I have been building exactly that, which is why this essay exists and why you should read the rest of it with appropriate suspicion of the person writing it.

One finding, in full

Abstraction is where this argument usually dies, so here is a single finding as it would actually appear, using the concentration limit from earlier.

The claim. From the strategy, page 14: “No single client will represent more than 20% of group revenue.” Typed as a hard constraint — not an objective, not an aspiration, a line that is either held or breached — and the evidence that would test it is named in the same breath: revenue by customer, the top-ten client schedule, the reporting pack.

The evidence. From the FY26 budget pack, page 31, in a table headed Revenue concentration — top five accounts: Meridian, 26.4%.

The pairing. Two documents, eighteen months apart, that no human has read side by side, because the first lives with the board secretary and the second with finance, and neither of them has any particular reason to open the other.

The binding. Both quotes are checked mechanically against the source bytes. If either sentence does not appear there verbatim, the finding dies before anyone sees it. This step catches nothing interesting and prevents the only failure that actually matters.

The innocent explanation, written before the finding is allowed to be called a breach. The constraint may carry a documented exception with a date on it. The definition of group revenue may have changed with an acquisition, making the two figures non-comparable. Meridian may be three legal entities the pack has consolidated for tidiness. Each gets checked, and at least two of them are the sort of thing an operations director will say out loud in the meeting, so it is considerably better to have answered them on the page first.

The counterfactual. What would change the verdict, stated plainly: a board minute approving an exception, or a revenue-definition note that makes the comparison invalid.

The dispute. The CFO sees it before the board does and can record an objection, which appears next to the finding and drops its severity until it is resolved. This is not a courtesy. It is the only reason anyone lets the thing into the building a second time.

What lands on the chief executive’s desk, then, is not “the business is out of compliance”. It is: here are two sentences from your own documents that do not agree, here is why they might innocently not agree, here is what would settle it, and here is what your CFO says about it. That is about forty minutes of a very good analyst’s attention. There are perhaps three hundred pairs like it, and no analyst.

The aspiration pile

The first thing that happens when you test a strategy against evidence is not the thing anyone expects.

You do not mostly find contradictions. You mostly find that large parts of the strategy cannot be tested at all. We will become the trusted partner of choice has no evidence that could confirm or refute it, this quarter or ever. It is not false. It is not a claim. It is a mood in the shape of a sentence, and there are usually eleven of them.

I have come to think this is the most useful finding a new chief executive can receive in week three, and it is certainly the cheapest. Sixty per cent of your inherited strategy is untestable prose; here is the forty per cent that says something; here are the four sentences that are load-bearing. That is not an insult to whoever wrote it — decks are written to secure agreement in a room, and vagueness is extremely good at securing agreement in a room. It is just that agreement and execution need different documents, and most companies only ever write the first one.

So the honest tool reports “your strategy cannot yet be tested” as a finding rather than swallowing it as a failure. It puts the mood-sentences in a pile marked aspiration, says so on the page, and gets on with the part that has evidence. There is something clarifying about seeing your inherited strategy sorted into two heaps, one of which is considerably taller than you were hoping.

A detour through public companies

A private company is an awkward place to test an idea like this. You cannot get in without being invited, and you cannot publish what you found once you have. So I built the same machinery for public ones, where three decades of filings sit on a government website for nothing, and pointed it at the short side — companies that might be about to have a bad year — because a negative thesis is the cleanest thing in the world to falsify. Either the restatement, the auditor change, the going-concern paragraph or the delisting arrives, or it does not. Being wrong is unambiguous and turns up on a schedule, which is a rare and valuable property.

The first real run went out on 29 August across fifty US retailers, six of them since delisted: 5,701 filings, 82,173 point-in-time observations, ten deterministic rules and not a single model anywhere near the scoring path.

It found nothing. Top-decile contradiction pressure had 0.94 times the base event rate at a year — very slightly worse than picking at random, an outcome that requires a certain amount of talent. The deciles were not monotonic. The two standard academic baselines did no better.

I mention it for three reasons. A result you only publish when it flatters you is not a result. The failure was informative in a way the design never could have been: eight of the ten rules need revenue growth to fire, which means a company quietly shrinking to death barely registers — Bed Bath & Beyond went bankrupt while sitting near the bottom of the leaderboard, which is not a subtle miss. And this is exactly what the honesty machinery is for. Scores are frozen at publication and re-checked on the next run; if a past score has moved, the run fails. That rule is a nuisance roughly once a quarter, and it is the entire difference between a track record and a collection of fond memories.

Two ways to get this badly wrong

Since I am recommending a category of thing, I should name the two versions of it that deserve to fail.

The first is the surveillance tool. Point a machine at the operating evidence, generate a list of ways the management team is failing, and hand it to the board. Congratulations: you have built an instrument for firing people, the organisation has correctly identified it as such, and within one quarter every document that enters it has been written for it. Companies are extremely good at optimising for whatever is measured, especially when what is measured is their own loyalty.

The countermeasure has to be structural. Findings are candidates, never verdicts. Every one carries the innocent explanation that was considered and what would change the mind. The person being described sees it before the board does and can dispute it, on the record, with the dispute visible in the report rather than resolved in a corridor. If that sounds like it blunts the tool — yes, deliberately, and it is the only version anyone will let through the door twice.

The second is worse: AI that writes your strategy. This one sells beautifully for about eleven months. It is also an abdication dressed as efficiency, and the first board that follows it off a cliff will discover that “the system recommended it” is not a defence recognised anywhere in company law or in the eyes of anyone whose money you lost.

The line I hold is dull and non-negotiable. The machine generates candidates. A named human promotes them to conclusions. The record shows who did which and when. A strategy is a set of choices about what you refuse to do, made by people who will carry the consequences, and nothing about cheap inference changes that. What changes is that those people can now be arguing about the right four things by Wednesday instead of the wrong forty by March.

What good actually looks like

Strip away the product and the shape of the answer is straightforward.

Turn the strategy into claims that could be checked — objectives, constraints, priorities, assumptions — and be honest about the ones that cannot. Point them at the evidence the company already produces. Report the gaps as candidates with the counter-argument attached and the source quoted verbatim, so the conversation is about the finding rather than about whether the machine made it up. Show the same report to the judged before the judges. And do it again next quarter, because a single snapshot tells you where you are while the delta tells you where you are going, and the delta is the bit that matters.

For a new chief executive that is a baseline in week three instead of an instinct in month fourteen. For an investor it is the operational layer that neither financial nor commercial diligence covers, refreshed quarterly, sourced from documents the company generates anyway — rather than a phone call in which everyone is upbeat.

Neither of these is a machine deciding anything. It is a machine reading, and admitting what it read.

What I have actually built, since you asked

Having spent this long describing a category of tool, it would be coy not to name the ones I have built. Weight the argument accordingly.

Strategy as Software is the one this essay is really about — strategic assurance for an incoming chief executive, a board or an operating partner. Typed claims, citation binding, the blind second opinion, the aspiration pile, the dispute on the record, and a report that closes with the exact filenames and checksums it was built from. It runs on a laptop, because board packs are the most sensitive paper a company owns and “please upload your hiring plan to our cloud” is a sentence that ends conversations. It is in beta, about to meet three real engagements, with a written kill criterion attached: if three of them produce nothing the chief executive did not already know, the premise is wrong and it stops.

FAULTLINE is the detour above — the same machinery pointed at public filings, where the evidence is free, the base rate is honest, and being wrong arrives on a timetable.

SEOboss is the ancestor of both, and by some distance the least glamorous. It reads a website, finds where the site contradicts its own claims, drafts the fix, applies it, and checks on the next scan whether it actually worked. I built it because SEO became a commodity and the only defensible part left was proving the fix landed. Some months later, reading a company’s board pack, it occurred to me that I had already built the small version of this — for web pages — and that the hard part had never been the reading.

The order is not flattering. The important idea turned up last, disguised as a much duller one.

The hundred days you actually have

If you are walking into a chief executive’s office, or about to fund someone else walking into one, the question worth asking on the first Monday is not what is the strategy. Somebody will hand you forty-one slides and you will be no wiser.

The question is what would have to be true for this plan to be working, and what in this building would show me either way.

If the answer comes back quickly and points at real documents, you have inherited something rare, and you should say so out loud because whoever built it deserves to hear it. If it comes back as a description of a dashboard nobody looks at, you have found your first hundred days’ work.

And if the answer is that nothing in the building could possibly show you either way — that the strategy is forty-one slides of things that cannot be true or false — then you have not inherited a strategy at all. You have inherited a mood board with a budget, and the good news, such as it is, is that you found out on Monday rather than in month fourteen.

Do bring your own laptop, though. The fingerprint thing takes ages.