Client story · Field & Ledger · 2026
First value replaced the tour
Field & Ledger scored a finished tour as activation. A time-to-event model showed the tour ending at minute six, and a kept result often never arriving. Putting a sample result first cut time to value 41%.
−41%time to first value
- Capture
- Model
- Insight
- Interface
- Test
- Hold
Summary
Field & Ledger helps operations teams check a file against their own rules. Onboarding ended when the tour ended. The board deck called that activation. Customers wrote to ask whether the check had run. Samir Qureshi, head of product, knew those could not both be success.
Tour completion, file upload, and the first accepted result were stored as separate events with their own clocks. A time-to-event model estimated minutes from account creation to a result the team kept, not to the end of the tour. Tour complete sat near minute six. A kept result, when it happened, was much later, and often it did not happen. The setup heatmap never reaches the result. Heat stays on the tour steps. The sample is cold.
The first screen now runs one sample and lists the exceptions. The tour waits behind that result. New workspaces, five weeks: time to a kept result fell 41%. The share of new teams who reached that result in session one rose 24%. Eleven people were timed in the room before the split. The clocks and the room agreed.
Two clocks that could not both be success
Activation had been defined as the tour’s last step. It is a clean event. It fires. It makes a chart go up. It is also unrelated to whether the product checked a file. The research question was which event predicted a team that stayed, and how long each event actually took. If the tour was fast and the result was rare, the interface was celebrating the wrong finish line.
Samir’s condition made the design obvious once the clocks were separate. Put their file on the first screen. If the sample is wrong they will know immediately, which is what the product is for. A tour of empty states was teaching the shape of the tool to people who still did not know if it worked on their data.
Three events, three clocks
We refused a single “onboarding complete” flag. Tour completion, file upload, and the first result the team kept were separate timestamps. Kept meant the team did not discard the sample and did not immediately re-upload a different file in confusion. A result that is thrown away is not value. It is a preview.
Timed interviews, eleven people, sat on top of the logs. Each person was asked to reach a result they trusted, and the clock in the room was compared with the clock in the product. Where they diverged, the product clock was wrong. Several tours marked complete while the person was still asking whether anything had been checked.
Empty files and second uploads
Empty sample files were excluded. An empty file produces a tidy zero-exception screen that looks like success and means the check never saw a row. Counting those as time-to-value would have rewarded the tour for handing people a blank. Duplicate exception rows from a re-upload were collapsed to the first kept result. The second upload is often the same file after a rename.
Internal workspaces and implementation-partner demos were removed. Partners finish the tour in a scripted order and upload a golden file. They are the best customers in the dashboard and they are not the cohort. Workspaces that never passed email confirmation were censored, not labeled as slow. They had not started.
Time to a result the team kept
The model was a time-to-event estimate of minutes from account creation to a kept result. Tour completion was a covariate, not the outcome. That single choice changes the product. With tour completion as the outcome, the tour looks like a six-minute success. With a kept result as the outcome, the tour looks like a delay that sometimes substitutes for the work.
Teams who uploaded early reached a result. Teams who finished the tour did not, unless the upload happened to be inside it. The heatmap of setup explains the behavior. Attention stays on the tour checklist. The place a sample result would appear is cold, because most sessions never got there. Leah’s line was the model’s line. The clocks disagree. The heatmap never reaches the result.
Heat on the steps, cold on the result
The setup heatmap is a picture of a checklist winning. The tour steps across the top are hot. The result region is cold. A product team can look at that picture and see “people love the tour.” The time-to-event model forbids that reading. Love, here, is time spent before value.
A list, before any tour
The blueprint is a list. ledger-sample.csv, three exceptions, and one row per check. Exception rows carry the tension: missing site code, duplicate vendor, amount over the rule. Clear rows stay quiet. They are context, not the decision. There are no tour steps on this grid. Before, upload was a later step with its own celebration. After, the first screen runs one sample and shows a result. The tour can remain behind it.
- Quiet
- Low
- Warm
- Hot
- Tension
List blueprint for Field and Ledger. Five checks from a sample file. Tension sits on the exception rows.
Eleven clocks, then the split
The room came first, on purpose. Eleven timed interviews are not a launch. They are a way to find out if the outcome label is even the right event before thousands of workspaces are split. The room and the model agreed on the label. Only then did the test start. Leah’s commitment was to time it, and to leave the tour available after the result existed.
The tour was not the goal
New workspaces were split for five weeks. Time to a kept result was primary. The share who reached that result in session one was the hold. Tour completion was measured and was deliberately not the goal. A variant that slowed the tour and sped the result would have won, and that is what we wanted permission to do.
Time to a kept result fell 41%. The share of new teams who reached it in session one rose 24%. Changing the outcome label changed the interface. The model was only as good as the event it was pointed at.
| Control | Tour, then upload |
|---|---|
| Variant | File and result first |
| Sample | New workspaces, 5 weeks |
| Primary | −41% time to a kept result |
| Held | +24% reached that result in session one |