Usable with a quick edit9 of 10 checks passed
The memo is clear, evidence-based, and actionable; a PM could add a condition for changing priorities with a quick edit without re-checking evidence or redoing reasoning.
Output
To: Onboarding squad Subject: Fix access and sync before optimising chart creation
Our biggest measured loss is before data connection: 59% of signups, roughly 7,300 people, never connect a source. We ask PMs and marketers to complete a database setup that many cannot do themselves, then make a lengthy sync depend on keeping a page open.
Another 8 percentage points disappear between connection and first chart: about 20% of those who connect. Teammate invitations fall to 9% of signups, but we should establish whether collaboration is necessary for value before treating that as our next bottleneck.
First-chart users have much higher week-2 retention, 62% versus 11%. That makes chart creation a promising activation milestone, but does not prove that pushing everyone through a chart will cause better retention.
Fix in this order
1. Give users a route forward without database credentials. Offer a clearly labelled sample-data experience and make CSV upload a prominent alternative.1 Add a setup hand-off that lets users ask a colleague to connect the source, then resume when it is ready. Replace the generic setup guide with short, source-specific instructions.
This addresses the largest measured loss and the strongest support signal: missing credentials account for 41% of onboarding tickets. That is a share of tickets, not signups, so we cannot yet quantify how much drop-off it explains. Sample data should demonstrate value while preserving a clear path to the user’s own data.
2. Make syncing durable and recoverable. Run syncs in the background so closing the page does not cancel them. Show meaningful status, explain failures and allow retries without repeating setup. Notify users when their data is ready.
A 20-minute foreground wait is a fragile requirement. Sync failures generate 18% of onboarding tickets and can waste the effort users have already invested.
3. Turn connected data into a useful first chart. Help users identify relevant tables instead of presenting every table equally. Use their chosen template to produce an editable chart from compatible data, with guided choices where mapping is needed. Do not finish template selection with an empty canvas.
This targets the roughly one in five connected users who never reach a chart. Measure whether they produce something useful, not merely whether a chart renders.
4. Remove avoidable preliminary work. Defer role and team-size questions unless their answers immediately improve setup. Review whether email verification must block exploration, retaining verification where required for account security.
These steps add friction before value, but we lack evidence that they drive the main loss.
Measure alongside delivery: instrument each screen, connection method, error and return visit. Test changes against first charts using customers’ own data, week-2 retention and trial-to-paid conversion. Track sample-data charts separately. Revisit invitations once we understand whether users need collaboration to succeed. To: Onboarding squad Subject: Fix access and sync before optimising chart creation
Our biggest measured loss is before data connection: 59% of signups, roughly 7,300 people, never connect a source. We ask PMs and marketers to complete a database setup that many cannot do themselves, then make a lengthy sync depend on keeping a page open.
Another 8 percentage points disappear between connection and first chart: about 20% of those who connect. Teammate invitations fall to 9% of signups, but we should establish whether collaboration is necessary for value before treating that as our next bottleneck.
First-chart users have much higher week-2 retention, 62% versus 11%. That makes chart creation a promising activation milestone, but does not prove that pushing everyone through a chart will cause better retention.
Fix in this order
1. Give users a route forward without database credentials. Offer a clearly labelled sample-data experience and make CSV upload a prominent alternative. Add a setup hand-off that lets users ask a colleague to connect the source, then resume when it is ready. Replace the generic setup guide with short, source-specific instructions.
This addresses the largest measured loss and the strongest support signal: missing credentials account for 41% of onboarding tickets. That is a share of tickets, not signups, so we cannot yet quantify how much drop-off it explains. Sample data should demonstrate value while preserving a clear path to the user’s own data.
2. Make syncing durable and recoverable. Run syncs in the background so closing the page does not cancel them. Show meaningful status, explain failures and allow retries without repeating setup. Notify users when their data is ready.
A 20-minute foreground wait is a fragile requirement. Sync failures generate 18% of onboarding tickets and can waste the effort users have already invested.
3. Turn connected data into a useful first chart. Help users identify relevant tables instead of presenting every table equally. Use their chosen template to produce an editable chart from compatible data, with guided choices where mapping is needed. Do not finish template selection with an empty canvas.
This targets the roughly one in five connected users who never reach a chart. Measure whether they produce something useful, not merely whether a chart renders.
4. Remove avoidable preliminary work. Defer role and team-size questions unless their answers immediately improve setup. Review whether email verification must block exploration, retaining verification where required for account security.
These steps add friction before value, but we lack evidence that they drive the main loss.
Measure alongside delivery: instrument each screen, connection method, error and return visit. Test changes against first charts using customers’ own data, week-2 retention and trial-to-paid conversion. Track sample-data charts separately. Revisit invitations once we understand whether users need collaboration to succeed.
What a PM had to fix
1Decision deferredMake the callQuick edit
What we’d changeRank the four parts of fix 1 so the squad knows what ships first, and make the calls fix 4 leaves open.
Grades and run details
Decision model 95 · LLM judge 91
Decision model checks
- passUses the supplied evidence correctly86%
- passAddresses the actual decision66%
- passRespects explicit constraints25%
- passIdentifies material uncertainty88%
- partialAvoids unsupported claims53%
- passProduces the required deliverable64%
- passPrioritises by impact98%
- passActivation defined by what predicts retention94%
- passGets the base of every number right74%
- passGets people to value sooner91%
Run
- Run
- #1
- Time to output
- 40 s
- Submitted
- 25 Sept 2026