TL;DR: Most operational data is not measured. It is inferred by somebody filling in a mandatory field with the most plausible option available, and then aggregated into a percentage that reads like evidence. A consultancy logged 62% of losses as price and cut fees by AUD $544,000; of eleven clients actually asked, two said price. A CRM showed 214 active candidates when consultants had spoken to 47. A spreadsheet said "in progress" for 76 days. In each case the record was filled in diligently by a competent person who had no way to check, and the business then managed on it.
The Field Was Mandatory
A proposal is declined. The email from procurement thanks the firm for its time and wishes them well. It gives no reason, because they never do.
Somebody opens the CRM, marks the opportunity Closed Lost, and reaches a mandatory field with six options. Price. Timing. Capability. Incumbent. No Decision. Other.
Nothing in the email supports any of them. So they think about what they know, remember the client asked twice about fee structure, and select Price. It is the most plausible thing on the list, and Other is what you pick when you have not thought about it.
Eighty-two times a year.
Twelve months later a slide reads: 62% of losses on price. Nobody built that number dishonestly. It came out of a mandatory field, filled in by a diligent person, counted by a spreadsheet doing exactly what it was asked.
The firm cut its fees 8%. AUD $544,000 over two quarters. The win rate did not move.
When eleven of those clients were eventually asked, two mentioned price. The rest talked about a delivery team they had never met, or the absence of any evidence the firm had worked in their sector before.
Recorded Is Not Observed
This is not a CRM problem, or a consulting problem. It is what happens whenever a system requires an answer that nobody is in a position to give.
Consulting. Six dropdown options, eighty-two losses, no debriefs on seventy-one of them. The reason field recorded what one person inferred from a polite email.
Recruitment. A CRM showing 214 active candidates. Asked how many they had actually spoken to in the past week, the consultants said 47. The other 167 were, in the kindest reading, resting. The pipeline did not look unhealthy. It looked full. These are not the same thing.
Onboarding. Claire in Bristol tracked 34 milestones per hire. One compliance module read "in progress" for 76 days, because in progress is what the field said when it was last touched and nothing had touched it since.
Property. Dave in Chicago held warranty dates for 340 units in a spreadsheet. The spreadsheet was correct. Nothing compared it to the invoice, so the business paid AUD equivalent thousands for a repair that was covered.
Insurance. Kathy in Atlanta reconciled eleven carriers and verified 68% of policy lines. The other 32% were recorded as reconciled, because that is what the process produced.
Facilities. Sarah in Sydney had a maintenance schedule listing 420 events. The schedule was accurate about what was due and silent about what had happened.
The shared structure: a system that captures a value, a person who supplies it without a way to verify it, and a report that presents the result with no trace of how it got there.
Why the Guess Survives Contact With the Board
Three reasons, and none of them is that anybody was careless.
The form does not offer honesty. No dropdown includes "I do not know". Given a mandatory field and six options, a reasonable person selects the most defensible one. The system has asked a question the person cannot answer and refused to accept the only true response. Every subsequent number inherits that.
Nothing distinguishes a measurement from an inference. Once "Price" sits in the field, it is indistinguishable from a reason a client actually gave. A quantity surveyor's site inspection and a guess made from a polite email occupy the same column, render in the same colour, and sum into the same total. The provenance is lost at the moment of entry, which is the only moment anybody knew it.
Aggregation launders it. One guess is obviously a guess. Eighty-two guesses become 62%, and a percentage in a board pack has the texture of measurement. The larger the sample, the more authoritative it looks, when in fact a larger sample of inferences is simply more inferences. This is the step where the error stops being recoverable, because by then nobody is looking at individual records.
Worse, the process passes inspection. There is a mandatory field, a review cadence, and an annual analysis. An auditor would find a documented practice and conclude the firm understands why it loses. It does have a practice. What it lacks is a source.
This is why "improve CRM hygiene" fails as a remedy. The data was entered on time, in the right field, by the right person. Hygiene was never the problem. Nobody had asked the client.
Closing the Gap Between the Record and the Fact
Across the Blueprint series, every agent that fixed a bad field did the same three things.
It asked the source. The client, the carrier statement, the equipment record, the candidate. Not a proxy, and not the memory of somebody who was in the room. The consultancy's debrief rate went from 13% of losses to 61% for no reason other than the request being drafted automatically and sent within a fortnight rather than remembered after two months.
It kept two fields where the business had one. What was stated and what the evidence supports, recorded separately and never merged. A client saying price while the proposal named no sector experience is a fact about both, and collapsing it is how 62% happened.
It made "unknown" a valid answer. When nobody responds, the honest record is unverified, and unverified is excluded from the analysis rather than defaulted into the largest category. Any dashboard built this way shows the unverified proportion next to everything else, so a percentage carries its own confidence with it.
Kathy's agent reads all 2,840 commission lines instead of sampling 68%. Dave's compares the warranty date to the invoice before anyone approves it. Sarah's records what was completed rather than what was scheduled.
The mechanism is always the same: the checking happens without a person having to remember to check, so the human sees the discrepancy rather than the summary.
Running costs across these designs: AUD $40 to $400 a month. Set against half a million given away on a hypothesis nobody tested.
Your Mandatory Field
Find the number you manage on that nobody measured. Loss reasons. Pipeline status. Project percentage complete. Lead source. Customer satisfaction inferred from the absence of complaints.
Then ask three questions.
Who filled this in, and what were they looking at when they did? If the answer is a person reasoning from context, it is an inference.
What would "I do not know" look like in this system? If there is nowhere to put it, you are collecting confident answers to questions nobody can answer.
What did we do because of this number? That is the real cost, and it is usually larger than the effort of collecting the data properly. A consultancy gave away AUD $544,000 rather than make eleven phone calls.
The record was not wrong because anybody lied. It was wrong because the business asked a question it had no way of answering, insisted on a response, and then believed the total.
Every Blueprint in the archive closes a gap like this one. Architectures, build guides, cost breakdowns, failure modes. Free to read. Free to build from.
56 agent designs. Your version of the mandatory dropdown is probably already in there.
Subscribe to AdAI News. New blueprint every week.
by SP, CEO - Connect on LinkedIn
for the AdAI Ed. Team
