An outcome is what happened after a decision: a first payment that never arrived, a loan that went 30 days past due, a call that confirmed fraud. Decisions on their own tell you what you decided. Outcomes tell you whether you were right. The Outcomes page is where you define each outcome once, publish it, and from then on use it in analyses, analytics recipes and policy mining.
- Where to find it
- Two kinds of outcome
- Defining an outcome
- Templates
- Publishing
- Uploading reported outcomes
- Using an outcome
- Troubleshooting
Where to find it
Open Outcomes from the tabs at the top of the data pages, next to Data intake and Portfolio. Outcomes read the data that arrives through data intake, so the analytics store must be set up first. If it is not, the page tells you that outcomes need the analytics store.
The page lists every outcome you have defined, with its title and name, the template it came from, what it is about, and its version. An outcome that has never been published shows Draft instead of a version.
Seeing the page needs read permission for batch and analytics features; creating, changing, publishing and uploading need write permission.
Two kinds of outcome
| Kind | How the result arrives | Example |
|---|---|---|
| Derived | Calculated from the data you already load, such as payments, positions and contact events. Nothing to upload. After publishing, the outcome is calculated automatically once a day. | First payment default, 30 days past due |
| Reported | Told to us by someone else, such as a fraud team or a card scheme. You upload the results as a file. | Confirmed fraud, chargeback |
Every derived outcome has an observation window. An outcome is counted only after its window closes, so recent decisions are not marked good before there has been time to go bad.
Defining an outcome
- Select New outcome.
- Pick a template from the buttons at the top. The template decides what the outcome measures and which settings it has. See Templates.
- Enter a Title. The identifier is derived from it. You can edit the identifier before the first save; afterwards it is fixed. Use lowercase letters, digits and underscores, starting with a letter, and keep it unique. This name is what you use in analyses.
- Optionally select a preset, such as Fraud, Credit or Default. A preset fills in the settings with typical values. You can still change any value afterwards.
- Adjust the settings and select Save.
Saving creates a draft. It does not affect any analysis until you publish it.
To change an outcome later, select its title in the list, change the settings and save. The template and identifier cannot be changed. If you need a different template, create a new outcome.
Templates
Payment outcomes
| Template | What counts as bad | Settings |
|---|---|---|
| Days past due | The account was at least the given number of days overdue at any point in the window | Days (how many days overdue) and Months (how long the window runs from the start of the account). Presets: Fraud (15 days, 3 months), Early fraud (5 days, 2 months), Credit (30 days, 6 months), Default (90 days, 12 months) |
| First payment default | The first payment due was not paid in full within the given days of its due date | Days and Payment roles (which payment events count as payments). Presets: Fraud (5 days), First payment (15 days), Late first payment (30 days) |
| Installment payment default | The chosen instalment was not paid in full within the given days of its due date | Installment (which one), Days and Payment roles. Presets: Second payment 10 days or 30 days, Third payment 10 days or 30 days, Fourth payment 30 days |
Contact and promise outcomes
| Template | What counts as bad | Settings |
|---|---|---|
| Contact not reached | No contact of the chosen kinds reached the customer within the window | Kind (path) and Kinds (which events count as contacts), Result (path) and Counts as reached (which results mean the customer was reached), Days. Presets: Welcome call (7 days), Early warning (3 days) |
| Contact not attempted | No contact was attempted within the window | Kind (path), Kinds and Days. Same presets |
| Promise kept | A customer's promise to pay was not met by the promised date | Grace days. Presets: Strict (0), With grace (3) |
Reported outcome
| Template | What counts as bad | Settings |
|---|---|---|
| Reported outcome | The value you report is one of the bad values | About (Call or Customer reference) and Bad values. Presets: Confirmed fraud, Chargeback, Yes or no |
Settings that ask for a path expect a JSONPath such as $.attrs.result. Settings that ask for a list take values separated by commas. If a value is not accepted, the page shows what it needs, for example a number, a list, or a path starting with $..
Publishing
Select Publish on a row to make the current settings live. Publishing:
- gives the outcome its first version, or the next version if it was published before. The row shows v1, v2 and so on.
- starts the daily calculation for a derived outcome, or opens the outcome for uploads if it is reported.
- keeps the earlier versions on record, so you can see what an analysis was measured against.
If nothing has changed since the last published version, publishing is refused with the message that nothing changed. Edit, then publish again.
Deleting an outcome stops the daily calculation of a derived outcome. Results already calculated stay in the analytics store.
Uploading reported outcomes
A reported outcome that has been published shows an Upload outcomes button on its row. Drafts and derived outcomes do not.
- Prepare a CSV or text file. Commas, semicolons and tabs are all accepted as separators.
- The first row must be a header with these columns:
| Column | Required | Meaning |
|---|---|---|
key | Yes | The identifier of the call or customer reference the result is about, depending on the About setting of the outcome |
value | Yes | The reported result, for example confirmed or chargeback. Numbers and true or false are read as such |
observed | No | When the result was observed. If it is empty, the time of the upload is used |
- Select Upload outcomes and choose the file.
Example:
key,value,observed
call-10482,confirmed,2026-10-14
call-10517,not_confirmed,2026-10-15
A value that matches one of the outcome's Bad values counts as 1 for that call. Any other value counts as 0.
Large files are sent in batches of 5,000 rows. When the upload ends, a message gives the totals as "recorded · refused". Rows that were already recorded earlier are not added twice, and count as recorded. A row is refused when, for example, its key is empty or its value cannot be read. A refused row does not stop the rest of the file.
If the file does not have both a key and a value column, nothing is sent and the page says so.
Using an outcome
Once published, an outcome is available to analytics recipes:
- In an analytics recipe,
$.outcome.<name>is 1 or 0 for whether the outcome occurred, and$.outcome_value.<name>is the reported value. $.__outcome_day,$.__outcome_weekand$.__outcome_monthgive the date the outcome relates to.
An analysis uses the published version, not the draft. If you change an outcome, publish it before you expect the change to show in an analysis.
Troubleshooting
| What you see | What it means |
|---|---|
| Outcomes need the analytics store | The analytics store is not set up for this account |
| The archive and the analytics store must be in the same database | The decision archive and the analytics store need to share a database before outcomes can be joined to decisions |
| Nothing changed since the last version | The outcome is already published as it is |
| Only a published reported outcome takes uploads | Publish the outcome first, and check it is a reported one |
| Upload between 1 and 5000 rows at a time | Applies to calls made directly to the API. The page splits large files for you |
| An outcome with this name exists | Choose another identifier |
| The file needs key and value columns | Fix the header row of the file |