The AI Friday Call: How to Run a Weekly Marketing Review

What the research actually found

The figure often quoted for this idea comes from a study by Giada Di Stefano, Francesca Gino, Gary Pisano and Bradley Staats, published as a Harvard Business School working paper titled "Learning by Thinking". The quoted number comes from one field experiment, as published summaries of the paper describe it: new employees at an outsourcing company in India were assigned to groups during training, and one group spent time at the end of each day writing about the lessons they had learned. The reported result is that this group scored 22.8% higher on the final training test than a control group that spent the time on more practice. A third group wrote the same reflections and then also explained their notes to a fellow trainee. If the exact figure matters to a decision of yours, read the paper itself rather than a summary, including this one.

The authors' broader argument, drawn from several studies, is that once someone has some experience of a task, deliberately articulating and codifying that experience can be worth more than accumulating more of it.

Three things the study does not show:

• • It is not a study of verbal reflection. The reported 22.8% result is for written reflection.
• • It is not a study of marketing teams or productivity. It measured trainees' scores on a final test.
• • It is not a promise for your team. A result in one setting is evidence for a principle, not a forecast.

So we do not claim a Friday call will make anyone 23% more productive. The defensible claim is narrower and still useful: stopping regularly to state what you learned, in a form that is recorded, helps learning stick. A spoken review is a practical way to do that for a busy team, and AI transcription makes the recording nearly free.

Why spoken, and why AI

Most small teams will not write a reflection every day. They will talk for 20 minutes once a week. Speaking is faster than writing, it surfaces disagreements, and it lets the person closest to the numbers explain them. The weakness of talk is that it evaporates. AI transcription and summarisation fix that: the conversation becomes a written, searchable log without anyone taking notes.

AfM applies the same idea to its own work: its operating rules require every meeting to be transcribed and attributed to the client and project it concerns, so what was decided is on record rather than in someone's memory.

The 20-minute agenda

Keep the same five questions every week, in the same order. Consistency is what makes weeks comparable.

1. • What shipped? (3 minutes) Campaigns launched, pages changed, emails sent. Facts only.
2. • What moved? (5 minutes) The outcome number and the two or three indicators that feed it, compared with last week and the forecast.
3. • What did we learn? (5 minutes) One or two sentences each. What surprised us? Which assumption turned out wrong?
4. • What is stuck? (3 minutes) Anything waiting on a decision, access or approval.
5. • What is the one change for next week? (4 minutes) One decision, one owner.

If the call regularly runs over, the agenda is being used for discussion that belongs elsewhere. Park it and book it separately.

Turning the transcript into a log

After the call, give the transcript to an AI assistant with a fixed instruction, for example:

"From this transcript, produce a log entry with five headings: shipped, moved, learned, stuck, next change. Use only what was said. Quote figures exactly. If something was unclear or disagreed on, say so under the relevant heading. Do not add recommendations."

The last line matters. The log records what the team said and decided; it is not the place for the AI's own ideas. Someone should read the entry the same day and correct anything wrong.

Weekly log template

Heading	What goes in it	Illustrative example
Shipped	What went live, with dates	New pricing page headline live Tuesday
Moved	Outcome and indicators against last week	Enquiries 11, up from 8; page views flat
Learned	What surprised us, which assumption failed	Most new enquiries mention the headline objection
Stuck	Waiting on decision, access or approval	Waiting on budget approval for the retargeting test
Next change	One decision and its owner	Apply the same headline approach to the services page (owner: Sam)

Recording people properly

A transcribed call records what people say, which can include personal data. Tell everyone on the call that it is transcribed and what the transcript is used for, and keep it only as long as it is useful. The Information Commissioner's Office describes transparency as being clear, open and honest with people from the start about who you are and why and how you process their data. Check the settings of your transcription tool, particularly whether it stores recordings or uses them to train models, before you use it for calls that include clients or customers.

How to tell if it is working

After four weeks, read the four log entries in a row. A working review shows three things: the "next change" items actually happened, the "learned" section contains things that changed later decisions, and the "stuck" section gets shorter. If the logs read as status reports with no decisions, tighten the agenda rather than abandoning the habit.

The review is also the natural place to check the forecast described in predictable growth with AI, and to read the results of tests run using the method in scientific AI marketing testing.

FAQ

Why Friday rather than Monday?

Friday reviews look back on a complete week while it is fresh, and the one change is ready to start on Monday. Any fixed day works if it is held every week.

Can one person run a Friday call on their own?

Yes. A founder can record a solo voice note answering the five questions and have AI turn it into the log. Having to say the answers aloud still forces clarity.

Did the study find that explaining to a colleague helped more?

As the study is summarised, it included a group that explained their notes to a peer, but the widely quoted 22.8% figure is reported for the written reflection group. We have not relied on the sharing group's result here.

How long should transcripts be kept?

Keep the written log, which is the useful record, and delete raw recordings and transcripts once the log has been checked, unless you have a clear reason and basis to keep them longer.