Docs / Experiments
Importing experiments
Bring the experiments you already documented in Notion, Google Sheets or a CSV into Converise in one pass, so the registry and the memory start with your history instead of an empty list. Nothing changes in the tool you export from.
Before you start
- The import lives under Data → Import experiments in the sidebar. Members, admins and owners can import; viewers cannot.
- A good source has one row per experiment and a header row - or, for Notion, one page per experiment.
- One import takes one .csv file, or the .md pages of a Notion export, up to 4 MB and 2,000 experiments. Split a larger source into several imports.
- Columns with people's names or e-mail addresses are never imported. The person running the import becomes the owner of the imported experiments.
The four steps
- 1
Source - where the experiments come from
Choose Notion, Google Sheets or CSV file. The screen shows the three export steps for that tool; follow them and drop the file in. Converise reads the source and moves on in a few seconds.
- 2
Fields - which column is what
You see the first rows of your file and, below them, every Converise field with a column picker. Say which of your columns feeds each field. Converise pre-fills the ones it recognises from the column names; you confirm or change them.
The label next to a picker says where the choice came from: Recognised (a clear match), Suggested (a likely match, worth a look), From saved mapping (you accepted it in an earlier import of this source) or Set by you. A field left on Not imported stays empty, and a column no field uses is simply left out.
The Assignment unit - what was randomised - is the one required field that is not a column: you pick one unit for every experiment in the import. It starts on User.
- 3
Values - your words, Converise's words
Your source says "Done" or "Live"; Converise has its own fixed lists. Here each distinct value is matched once, and every record with that value follows:
- Business area and Audience - to the areas and audiences of your workspace. An admin can add a missing one straight from the list; it is created when you import, so a draft you abandon leaves your workspace as it was.
- Status and Decision - to the Converise statuses and to ship, reject, iterate or investigate. Only completed experiments are asked for a decision; a decision value on a running or planned one is kept as source text.
When the source leaves nothing to match, this step is ticked and skipped.
- 4
Preview - what will be imported
A summary of how many experiments were found, how many will be imported as they are, how many with a note to look at later, and how many will be skipped and why. Below it, the records themselves, filterable to the ones that need attention. The column choices stay editable under How we read your data. When the numbers look right, press Import.
Fields reference
The Fields step groups the fields by how much the import needs them. Required fields are needed to create a record at all. Recommended fields make the experiment usable and safe to reason about later. Optional fields carry everything else your source may hold.
| Field | Needed | What it is | Good to know |
|---|---|---|---|
| Name | Required | The name or title of the experiment. | A record without a name is not imported. |
| Business area | Required | The part of the business the experiment belongs to - product, pricing, lifecycle, ads and so on. | Matched to your workspace's areas in the Values step. A record with no area, or with an area you leave unmatched, is not imported. |
| Status | Required | Where the experiment is in its life: planned, running, completed and so on. | You translate your own status words in the Values step. A status left unmatched keeps its records out - nothing is assumed to be completed. |
| Hypothesis | Recommended | What the team believed would happen, and why. | A record without one is still imported, with a note to add it. |
| Variants | Recommended | The names of the control and the treatments. | Without readable variants the experiment is imported as needing review: it is in the registry and in memory, but variants have to be added before results can be attached. |
| Traffic split | Recommended | The share of traffic each variant was planned to get, in the order of the variants (50/50, 34/33/33, 1:1:1). | Used only when it matches the number of variants. Otherwise the variants are imported without planned shares. |
| Target audience | Recommended | Who was exposed: a segment, a market, a device group. | Matched to your workspace's audiences in the Values step. A value you leave unmatched, and every experiment when no column is picked, gets the All Users audience. |
| Primary KPI | Recommended | The name of the main metric the test was judged on. | Kept as text. The KPI itself is set up on the experiment when you attach results. |
| Assignment unit | Required | What was randomised between the variants: users, sessions, accounts, campaigns and so on. | Not read from a column: you choose one unit for the whole import, and it starts on User. It can be changed on any experiment later. |
| Actual start | Recommended | The day the experiment really started running. | Needs a column of dates. |
| Actual end | Recommended | The day the experiment really stopped. | Needs a column of dates. A record whose end is before its start is not imported. |
| Decision | Optional | What was decided after the test: ship, reject, iterate or investigate. | Goes to the experiment's decision log, marked as imported, with the decision date from the source. Only completed experiments are asked for one. |
| Result summary | Optional | What happened, in words. | Kept as text in the experiment's memory. Numbers in it stay text - they are never treated as measured results. |
| Record ID | Recommended | A stable, unique identifier of the experiment in the source, such as EXP-101. | Lets a later import recognise the same experiment and update it instead of creating a duplicate. Every row needs its own value. |
| Planned start | Optional | The day the experiment was planned to start. | Needs a column of dates. |
| Planned end | Optional | The day the experiment was planned to end. | Needs a column of dates. |
| Labels | Optional | Free tags: page, channel, squad, product line, test type. | Several tags in one cell are split on commas, semicolons or vertical bars. |
| Decision rationale | Optional | Why that decision was taken. | Stored with the decision. When it is empty, the log says the decision was carried over from the source. |
| Decision date | Optional | When the decision or the rollout happened. | Needs a column of dates. |
| Learnings | Optional | What the team learned - the insight worth reusing. | Kept as text in the experiment's memory. |
| Notes | Optional | Any other useful text: next steps, comments. | Kept as text in the experiment's memory. |
| Source link | Optional | A link back to the original page or record. | Needs a column of links. |
What happens to each record
| Record | What it means | Typical reason |
|---|---|---|
| Complete | Imported as a full experiment, ready for results to be attached. | Name, area, status and unit are known and the variants could be read. |
| Needs review | Imported into the registry and the memory, without variants, flagged "Missing variants". | The source has no variants column, or the variants could not be read. Add them on the experiment later to attach results. |
| Not importable | Skipped. Nothing is created for it. | No name, no business area or an area left unmatched, no status or a status left unmatched, no assignment unit decided, or an end date before the start date. |
A missing hypothesis, an unreadable date, a decision value left unmatched on a completed experiment or an audience that is not in your workspace (set to All Users) does not stop a record. It is imported and flagged, so you can fix it on the experiment afterwards.
After the import
The report counts what was Created, Updated and Skipped, and lists the records that Need your choice. The column and value choices you made are saved for that source, so the next import of it starts pre-filled.
Importing the same source again is safe:
- A record with a Record ID is recognised and updated instead of duplicated.
- Name, hypothesis, business area, labels and planned dates follow the source when only the source changed. If you edited the same field in Converise, your edit is kept and the report shows both values.
- Status, actual dates, variants and the assignment unit of an existing experiment are never changed by an import - the report tells you the source moved, and you change them on the experiment.
- A record matched only by its row position or name waits for your choice: skip it, update the existing experiment, or create a new one. So does a record whose experiment you deleted in Converise since: skip it or create it again. Nothing is overwritten until you choose.