What an agent leaves behind.
Every step an agent of ours takes leaves a record a person can read afterwards. Here are two, replayed at the pace they were written.
These are illustrations. They are replayed from fixtures, not from a live system, and the names, times and case numbers are invented. The fields are not: they are the six every real record carries, and the CSV is built from the rows you can see.
The record
12 records · 1 time it stopped and asked a person · 1 human change · nothing invented by the agentThe record · enquiry handling
What an agent leaves behind.
One out-of-hours enquiry to a mortgage broker, handled end to end. Every row is a record the agent wrote as it worked: what set it off, what it read, what it did, what it would not do, and who it handed to.
- Trigger Web form, out of hours “Self-employed, buying at about £340k in Leith. Need a mortgage adviser.” inputs form #4471·output case 2026-0913 opened
- Read Qualifying criteria, version 3 Four rules, written by Sarah M. on 2 September. Three met: self-employed, loan size, area. Deposit not stated. inputs criteria.md·output 3 of 4 met
- Asked One question, in the client’s own channel “Thanks, roughly what deposit do you have? A percentage is fine.” output email 1 of 2·confirmed by policy: may ask, may not advise
- Trigger Reply “About 15%, maybe a little more with help from family. What rate would I get?” inputs email reply
- Read The criteria, again Four of four met. Qualifies for a first conversation. output 4 of 4 met
- Refused Not done The rate question. Not permitted to quote a rate or to advise, so it was not answered: it was flagged to Sarah M. with the reply attached. refused rate quote·reason advice is a person’s to give
- Read Sarah M.’s diary, read-only Tuesday 22nd, 10:00 free. Wednesday 23rd, 14:30 free. inputs calendar, read only
- Did Two slots offered “Sarah can talk on Tuesday at 10:00 or Wednesday at 14:30, which suits? She will cover the rate question then.” output email 2 of 2
- Trigger Reply “Tuesday please.” inputs email reply
- Did Booked Tuesday 22 September, 10:00. Invite sent. Case 2026-0913 updated with the transcript. output calendar write · CRM #10932·confirmed by pending
- Handed Handed to Sarah M., by name The transcript, the four criteria with their answers, and the refused question at the top of the note. output hand-over note
- Confirmed Sarah M. read the record Moved the call to 10:30 and answered the rate question herself. Nothing else changed. confirmed by Sarah M.·changed 10:00 to 10:30
An illustration, replayed from a fixture, not a live system. The names, the times and the case number are invented. The six fields on every row are the six a real record carries, and this page exports them as a CSV a person can read: built from the rows above, so the file and the ledger cannot disagree.
The record · document chasing
And what it does when it should stop.
One outstanding document on a mortgage case, chased on a schedule somebody wrote down. The row that matters here is not a refusal: it is the moment the question stopped being about paperwork, and the agent stopped with it.
- Trigger One item outstanding, four days Case 2026-0871 cannot go to the lender without the client’s 2024/25 SA302. inputs case file·output chase opened
- Read The chase policy, version 2 Written by Dan R. on 14 August. Ask in the client’s own channel; day 1, day 4, day 8; then stop and tell a person. inputs chase-policy.md·output ask 1 of 3
- Asked Asked, where the client already is “Morning, still need your SA302 for 2024/25 to finish the application. It’s in HMRC under Self Assessment, then Tax return options. No rush today, but the lender’s quote expires on the 30th.” output message 1 of 3·confirmed by policy: may ask, may not assess
- Trigger Reply, with a photo A screenshot of an HMRC page. inputs image, 2.1 MB
- Read It is the right menu, the wrong page What arrived is the tax year overview. The case needs that and the tax calculation; one of the two is now on file. output 1 of 2 on file·confirmed by policy: may identify, may not assess
- Did Said so, plainly, once “Thanks, that one’s the tax year overview, which we do need. The other is the tax calculation, one item up the same menu. Send that and we’re done.” output message 2 of 3
- Trigger Reply, and a question The SA302 arrives. With it: “My accountant says the 2024/25 figure is wrong and he’s amending it. Does that matter?” inputs PDF, 240 KB·output 2 of 2 on file
- Handed Stopped. Handed to Dan R., by name An amended return changes the income the whole application rests on. That is not a question about where to send a document, so the chase ended there and the question went to Dan R. unanswered, with both documents, the transcript, and the sentence that stopped it. refused whether the amendment matters·reason it changes the case, not the paperwork·output hand-over note
- Read The rest of the schedule, cancelled Day 8 and the final chase withdrawn. A client whose case is on hold is not asked again for something they have already sent. output chase closed·changed 2 scheduled asks cancelled
- Confirmed Dan R. read the record Called the client, paused the application until the amendment lands, and rebooked the lender quote. Corrected one line in the note: the quote expires on the 31st, not the 30th. confirmed by Dan R.·changed quote expiry 30th to 31st
An illustration, replayed from a fixture, not a live system. The names, the times and the case number are invented. The six fields on every row are the six a real record carries, and this page exports them as a CSV a person can read: built from the rows above, so the file and the ledger cannot disagree.
Six fields
Every action, six answers.
Written into the statement of work before anything is built, so the record is a deliverable, not a feature somebody might add later.
-
Trigger
What set it off: the form, the reply, the date in the diary. Nothing an agent does should start without one it can name.
-
Inputs
What it read to do it: the criteria, the case file, the policy, each by name and version, so a reviewer can read the same thing.
-
Output
What it produced or changed: the message sent, the slot booked, the note written, with where it went.
-
Confirmed by
Who confirmed it, by name, or the written policy it acted under while nobody was there.
-
Changed
What that person changed. The edits are the evidence that a human read it rather than waved it through.
-
Refused
What it would not do, and why. A guardrail nobody can see working is indistinguishable from no guardrail at all.
Why two
Two ways a bounded agent should fail.
One record is an anecdote. The pair is the argument. In the first, the row that matters is a refusal: asked what rate the client would get, the agent did not answer, because advice is a person’s to give, and it put the question at the top of the hand-over note.
In the second, it is an escalation: chasing a tax document, the agent learned the figure on it was being amended, and stopped, because the question had stopped being about paperwork. It cancelled its own remaining chases and handed the case to a named person.
Those are the two ways an agent with a clear boundary is supposed to fail, and the record is how you know it did.
What writes a record like this
- Enquiry handling Answer, qualify, book into your real diary, hand to a named person. Four answers
- Document chasing Asks the client until it arrives, or until a person takes over. Four answers
- The audit layer Six records per action, exportable, including what it refused to do. Four answers
One job, named. Tell us which.
Every engagement starts with one job and the six fields written into the statement of work before anything is built. If you can name the job, that is the conversation.