Your weekly product review, minus the replay marathon

Workflow · Weekly product review

212 new recordings. Only a few need you.

On Monday, someone asks what stops people at checkout. The answer is in the recordings from last week. Nobody had time to watch them.

The job was never to watch sessions. The job was to know what went wrong, for how many people, and how it moved after the last release. Sequential finds what went wrong, counts how many, and measures how the rate moved after the release.

Your review, in four stops

You run nothing each week. Findings come from fixed rules over the events your site sends, such as clicks, page views, failed requests and errors. Sequential counts them exactly.

For Watch, about half an hour after a session ends, a judge reads a text summary of its events. It answers questions such as what the person came to do and where they struggled. It never watches the video. For Ask, the judge reads the same summaries against your sentence. In both, the answer is a likelihood, not a certainty. The judge does not watch video, so it can cover sessions that nobody had time to watch.

1. 01

   See what moved.

   The Dashboard builds itself from your project’s data, up to the last 30 days: traffic, errors, rage and dead clicks, top pages and more. A tile shows only when there is data for it. Traffic and some lists show the change, such as "+12 % vs the week before".

2. 02

   Watch only the moments that matter.

   Watch puts sessions in three groups (broke for the user, struggled, minor) and ranks each group by how much a session needs a look. Each row gives its reason, such as "Severe struggle — probably switched to support". "Watch from 2:14" opens the replay just before that moment.

3. 03

   See what keeps breaking.

   Findings counts errors, failed requests, clicks that do nothing, repeated angry clicks and dead ends, exactly: "12 of 300 sessions on /checkout · 5 users · last 7 days".

4. 04

   Find what no filter can.

   Ask in one sentence, such as "Clicked Send invite several times and nothing happened". You get the likely matches.

You still watch replays. You start at the moment that matters, in the sessions at the top of the list.

Example: Northlight Outfitters, our fictional demo shop

One Monday morning

Maya runs product. It is Monday, 09:10, and she has a coffee.

The top session in Watch reads "Severe struggle — probably switched to support". She presses Watch from 2:14. A customer types a postcode with a space. The form rejects it. The customer tries again and leaves.

Findings has it, counted: a failed request on the shipping page, 38 people. In plain words, people get stuck at the shipping address.

Her agent files WEB-231 in the team’s tracker, with the write-up from Sequential attached. The team ships release 2026.10.09. The release is recorded, and the finding shows it.

A week later, the finding shows: Measured after release: 41% → 12% of sessions on the shipping page. That is a measurement, not a verdict. Why it moved is for Maya to decide.

Her review took about as long as her coffee.

It tells you when it is not sure

• When it cannot tell what someone wanted or how the session ended, it says "too close to call". It does not guess.
• Point to a badge to see its score and the Worth watching probability.
• Your thumbs up or thumbs down grade the ranking. They do not retrain it.
• After 30 checks, the Calibration tab tells you if the Worth watching answer is right as often as it says, on your own sessions. It also says that people check only what the ranking showed them.
• The hosted preview keeps 7 days of data by default, and deletes checks after about 15 days. So on the preview, the 30 checks must come in that time.

You can check the ranking yourself, on your own sessions.

Or ask your agent for the review

Do you use Claude Code, Cursor, Codex or Claude Desktop? An owner or admin connects it from the Agents page. Then ask it for the weekly review. The agent pulls the same ranked sessions, counts and before-and-after rates. Then it offers to file what needs fixing.

Sequential finds, counts, records releases, measures and answers. Your agent does the work in your tools, on the schedule that you give it.

Good to know

• Sequential answers. Your agent acts. Sequential does not file tickets, post to your tools or send a weekly briefing. Your agent does the follow-up.
• Before and after is a measurement, not a cause. It shows how a rate moved after a release. It never says why.
• Week-on-week changes need two weeks of data. Until then, no figure shows. The hosted preview keeps 7 days of data by default, so there the figure may not show.
• Daily allowances. Judging covers sessions with at least 3 events, up to a daily allowance per project. Watch shows how many sessions it did not judge. Ask and the Insights tab have their own daily allowances. Opening Watch uses no allowance.
• Ask finds sessions. It does not count them.
• Calibration needs 30 checks. It gives no verdict before that.
• What goes to the judge. Event summaries and Ask questions go to the third-party service that judges sessions. Sequential leaves out the user id of the session. The Insights tab sends a fuller summary. Sequential does not scan questions, so keep names and emails out of them.

Questions

Review the week. Do not live it again.

Start with your own sessions. Next week, your review starts with a ranked list, not 212 recordings.