What a consulting meeting owes the written record, and why minutes do not cover it

Search for how to document a client meeting and the answer is a template, with minor variations: date, attendees, objectives, discussion points, decisions, action items, next steps. The template is not wrong. It is answering a different question, which is how a project team keeps itself coordinated between meetings. Scope disputes live in a separate corner of the internet, filed under change control, and the two bodies of advice almost never mention each other. That gap is the whole subject here.

The reason they belong together is specific to professional services. A software team's meeting decides what somebody will go and build; the build is the deliverable. An advisory meeting frequently is the deliverable, or shapes it so directly that the distinction stops mattering. When the output of an hour is a judgment, or a decision the client could not reach on their own, nothing else gets produced that could serve as evidence of what was agreed.

Each side leaves remembering its own half of the conversation

Neither party is being dishonest when the accounts diverge. People recall what they said themselves more reliably than what was said to them, and after any real delay both versions have thinned out enormously. What a written record protects against is duller than bad faith. It protects against two honest accounts that barely overlap.

The numbers are worse than most people expect. A 2018 review in Policy Insights from the Behavioral and Brain Sciences, gathering the laboratory work on memory for conversation, puts accurate recall of a conversation after delays ranging from several minutes to several weeks at zero to twenty percent of the idea units that were actually spoken. In one of the studies it reviews, 128 participants recalled a seven-minute conversation after a five-minute gap; the average was ten percent of idea units, the best participant managed forty percent, and the worst recalled nothing at all. That is five minutes, not three weeks.

The asymmetry is the part that explains client disputes. Another study in the same review tested recall three to four days after a conversation and found participants recovering six percent of their own statements and three percent of what the other person said. Of the statements they did attribute to their conversational partner, only 56 percent were accurate. Double that for a professional in a domain they know well and the shape stays the same: you leave with a decent record of your own undertakings and a poor one of the client's.

Both sides are in that position simultaneously. Each remembers having been clear. Each remembers the other side having been vague. Writing it down does not fix either memory, and the research is blunt about this: taking notes increases how much gets reported later without improving how accurate the report is. What a written record does is hold a version that stops decaying, made at a moment when the conversation was still intact. The meeting where your account and theirs first diverged is sitting on a recording.

In consulting a commitment made out loud is a change to the engagement

Treating a verbal commitment as a to-do item is the mistake that costs money. In a sales conversation, an unrecorded promise costs you the next step. In an engagement, the same sentence may have changed what you owe and what the work was costed against, and neither of those appears on a task list.

The test is whether the sentence would need a signature if it had arrived by email. Most do not, which is exactly why they pass unexamined.

Each row is a line that would never appear in a to-do list, because none of them is a task of yours. They are conditions and assumptions, and between them they decide what the engagement means. The distinction between an agreed next step and a warm one is the same problem one layer up, and the way a next step either does or does not commit somebody is worth reading alongside this. In an engagement the stakes are higher, because what is at risk is not the pipeline but the deliverable the client is paying for.

Minutes answer what the meeting was about, which is the wrong question

Minutes are a record of proceedings. An engagement record is a record of obligations. The two look similar on the page and answer different questions, and a consulting practice that keeps only the first has written down the least useful half of the hour.

A default worth starting from: the recap goes out the same working day, before the next conversation overwrites the last one, and it is written for the client to confirm rather than for your own file. Confirmation is what converts a record into a shared one. A note sitting in your drive is evidence of what you believed, and a note the client replied to is evidence of what both of you agreed, which is a materially different object when somebody disputes it. The same split runs through how consulting engagements are run on the record.

Length is not the variable people think it is. The failure of a long recap is not that the client stops reading, although they do. It is that obligations get buried among topics, and a buried obligation is functionally unrecorded.

A record ordered by time survives a dispute better than one ordered by topic

How the record is organized changes what it preserves. A timeline format preserves the words and the order; a topic format preserves the gist. Since a dispute is almost always about exact wording and sequence, the organizing choice is not cosmetic.

This has been measured directly. In a set of experiments published in Law and Human Behavior in 2019, 44 participants heard a 272-second recorded conversation and reported on it after a 15-minute filler delay, half of them using a timeline laid out in sequence and half writing a free account. Overall quantity of correct detail did not differ. Correct verbatim detail did: the timeline group reported more than twice as much of it, at no cost to accuracy, and made fewer sequencing errors. The follow-up experiments found the same format also produced more correct attributions of who said what.

Verbatim and sequence are precisely the two things an engagement argument turns on. Not what the meeting covered, but whether the phrase about the second region came before or after the conversation about the basis of the fee, and which of the four people in the room said it. A recap grouped under headings has already thrown both away. A record carrying timestamps has not, and the practical form is unglamorous: topics with the minute they started, each commitment pointing back at the moment it was made. Which minute would you open first?

The question that was parked is the one that returns as a change request

A question raised and left open behaves like a loss reason nobody wrote down: it does not disappear, it just stops being visible until the cost arrives.

The dangerous item is not the disputed one. It is the question somebody raised, the room acknowledged, and nobody answered, which everyone then leaves believing was settled because it was not argued about. It costs nothing at the time and surfaces as a change request two months later.

A rule of thumb that works: a question raised in two separate meetings and answered in neither is a scope risk, regardless of how minor it sounds. Twice means somebody keeps needing it, and never answered means nobody has taken it on. Both facts are visible only if the record carries unanswered questions as a standing category rather than dropping them because they produced no action.

The three that recur across engagements are worth naming. Who signs off on the client side, which budget a later phase comes from, and what happens if a dependency the client owns arrives late. Each one gets asked casually and answered with a generality, and everybody leaves treating it as settled. Tracking which questions keep coming back across an entire client base is the same technique as reading what clients repeatedly raise without running a survey, applied to one engagement instead of a market.

What we got wrong building the record

Two mistakes of ours, both in the part of the product that turns a conversation into a record, and both of a kind that looks like nothing until somebody relies on it.

The first was a filter panel. Three screens show what appears to be the same set of controls: search, host, participant, duration, analysis status, language, period. On the summaries screen, three of them are drawn and never reach the query behind it: duration, analysis status and language. A consultant looking for the meeting where a particular commitment was made sets a filter, gets a short list, and concludes the conversation is not there. The list was real; it was simply filtered by something other than what the screen said. A control that is rendered but not connected is worse than an absent one, because an absent control sends you to look somewhere else while a dead one tells you the answer is no. We describe the behavior honestly now, which is the smaller half of the fix.

The second is in how a commitment finds an owner. Each row of the agreements table carries the name as it was said in the room, and that string gets normalized and matched against the organization's directory of name variants. Match, and the commitment is attached to a person and shows up in their list. No match, and the row keeps the spoken name and belongs to nobody. On screen the two outcomes are identical: a task, with a name next to it, looking complete. We had treated the directory of name variants as administrative setup, something to fill in during onboarding and forget about. That was wrong. It is the step that converts a sentence into an obligation somebody owns, and until it is populated the record holds commitments that no one will ever be reminded of. A row with a name on it is not the same as a row with an owner, and nothing in the interface was saying so.

What to do next

Open the last engagement where the scope moved and the client remembered it differently. Find the meeting where the change first came up. Then answer three questions in order: was there a written record of that meeting, does it contain the sentence, and did the client ever reply to it.

Most of the time the first answer is yes, the second is no, and the third never came up because the recap was filed rather than sent for confirmation. That sequence tells you which of the three habits to fix first, and it is almost never the one people assume. For teams already recording calls, the related question of how much a generated note can be trusted on its own decides how much checking the habit needs around it.

The engagement you are mid-way through is already recorded. Send us the scoping call and the two reviews that followed it, and we will mark every commitment made out loud, which of them reached a written record, and which questions came up twice without an answer. The scoping call and the two reviews after it.