The statistic we do not have

We cannot tell you what decision decay costs your company. Not approximately, not with a range, not with a confident number and an asterisk. Nobody can, because nobody has measured it, and that includes us.

This is an awkward thing to publish on a website whose entire argument rests on the problem being expensive. It is also the most useful thing we can tell a buyer who is going to check.

Four good numbers, all of them adjacent

The research we cite is real, and every figure links to its source. What none of it does is measure the specific failure we are building against: a decision changes, and the teams already acting on the old version are never told.

PMI measured project communication. Of every $1 billion spent on projects, $135 million is at risk, and $75 million of that traces to ineffective communications (PMI, 2013). That is communication inside project delivery. Some of it is our problem. Most of it is briefs that were unclear in the first place, which is a different failure with a different fix.

McKinsey measured decision-making efficiency. Sixty-one percent of executives say at least half their decision-making time is wasted, which for a Fortune 500 company works out to roughly 530,000 days of managers’ time a year (McKinsey). That counts the meetings it takes to reach a decision. It does not count what happens to the decision afterwards.

The workslop study measured AI output quality at handoff: 40% of employees received work in the past month that looked finished and was not, costing about one hour fifty-six minutes each time (HBR, September 2025). Closer to us, because roughly half of it moves between colleagues. Still not the same thing. Workslop is bad work that looks good. Decision decay is good work built on a premise that expired.

MIT Project NANDA measured return on GenAI pilots and found about 95% of organisations got none (The GenAI Divide). Their diagnosis, that the systems do not retain what they learn, describes the hole we are trying to fill. Their measurement is of pilot ROI, which is a different quantity entirely.

Four studies that establish the neighbourhood is expensive. None that prices the house.

Why nobody has measured it

Not because researchers have not thought of it. Because the thing is invisible by construction, and measuring it means measuring an absence.

To count instances of decision decay you would need to know, for every decision a company changed, which teams were acting on the previous version and whether each of them found out. The second half of that is not recorded anywhere. There is no log of who was not told. The only trace it leaves is the escalation twelve weeks later, which gets filed under the customer complaint rather than under the decision that caused it.

Ask a company how often this happens and you get an anecdote, because anecdotes are what the data looks like from the inside. Everyone has a story. Nobody has a rate.

What we are doing instead

Fifteen structured interviews, with the people on the receiving end rather than the people who make the decisions. Heads of Sales, Customer Success leads, senior PMs. The ones who find out late.

The questions are deliberately narrow, because broad ones produce agreement rather than data. When did you last discover a change from a customer instead of from a colleague. How long had it been true before you learned it. What had you already committed to in that window. What would it have cost to find out on day one instead of day forty.

That produces a first-party figure with a real methodology and an obvious limitation: fifteen interviews is a sample, not a census, and the people who agree to talk to a startup about a problem are more likely to have the problem. We will publish the number with the sample size and the selection bias attached, and we will not round it into something more impressive.

What would tell us we are wrong

If the interviews come back and the median gap between a decision changing and the downstream team learning about it is under a week, we do not have a business. A week is a sprint. Nobody buys software to close a gap that closes itself.

If the cost of each instance turns out to be an afternoon of rework rather than a damaged customer relationship, we are selling a convenience rather than a risk control, and the whole positioning is wrong.

Both are live possibilities. We think they are unlikely, based on the interviews we have already done and on the shape of the escalations people describe, but thinking is not knowing and we will say which it was.

Until then, treat everything else on this site as an argument about a mechanism rather than a claim about a magnitude. The mechanism we are confident in. The magnitude is the thing we are going out to measure, and we would rather tell you that now than have you find the seam yourself.