Running and repeating an analysis

This page explains how to start an analysis, run it again on a schedule, and turn findings into owned work. Each ready-made analysis follows a playbook that decides a specific risk, credit or portfolio question. Use the analysis overview to choose the right question and understand its result.

Start an analysis

Go to Analyses → New analysis. Choose an analysis from one of these groups:

  • Portfolio and losses
  • Decisions
  • Scores and models
  • Watching for risk

The dialog tells you what is missing when an analysis cannot run and links to the page that fixes it. The form asks for the inputs that analysis needs. Fields the available data already answers are preselected. Optional inputs are under More settings.

You can also start an analysis where its context is already known:

  • The Analyses step of a Data intake preselects that intake. See analysis data for how data is prepared.
  • The endpoint page offers Shadow review and Champion and challenger for the deployment you are viewing.
  • Policy mining offers Scorecard from decisions.

What happens during a run

Starting an analysis queues a run for a background worker. You can leave the page while it runs. The status moves through Waiting, Running and Done. The run page polls for changes and updates itself.

A watch-out can stop a run. The status is Blocked, and the page names the watch-out that stopped it. Failed means that an error occurred. Results stay with the run, so opening it later shows the result produced at that time.

Analyses run on your node and database in the background, with one job running at a time in each queue. A typical run takes seconds to a minute on a book of tens of thousands of records. Large archives take longer.

For the full data requirements and the options for starting, repeating, emailing and opening cases, see analysis data and running analyses.

Read findings between runs

The result page shows the analysis answer. See the analysis overview for how to read its hypotheses, watch-outs, findings and evidence.

Between runs, the page compares the current run with the previous run of the same analysis on the same data. A finding is either a hypothesis that was rejected or a watch-out that warned or blocked. Its state tells you what changed:

  • New means it appeared since the previous run.
  • Still open means it was present before and remains.
  • Cleared means it was present before and is now gone.

The first run marks everything it finds as new. A run with nothing new says so. This keeps a repeat useful for change detection instead of making you review the same findings each time.

A hypothesis is a statement the analysis tests. Its verdict is held, rejected or unclear, with the number behind that verdict. A watch-out is a known trap the analysis checks for. It stays quiet, warns or blocks the run. A result marked Directional has a sample too small for its number to be firm.

Repeat an analysis

On a result, use Repeat and choose Once, Every day, Every week, Every month or Don't repeat. Repeating analyses appear in Repeating analyses on the Analyses page with their cadence. Each one also has a tile on Home showing its headline numbers, its last run, and any Overdue or failed flag.

Use Pause to stop a repeat and Resume to start it again. Pausing keeps its history.

Dates in repeat settings are rolling. A setting such as “the last 30 days” or “yesterday” is stored relative to the run day. Each run works out that window again, so it examines its own period instead of repeating the first run's dates.

A repeat waits when its previous run is still running. If it reads a feed, the feed refreshes first. The schedule shows Refreshing the data, and the analysis starts after the new records arrive. You can also schedule a feed refresh without an analysis.

Overdue means the repeat should have started more than a few minutes ago and has not. This usually means the background worker is not running. Contact support.

The Digest analysis gathers open findings from every active repeating watch. It shows what is new since each watch's previous run, what remains open, and which watches did not run. Use it for a weekly review instead of opening each watch separately.

Email findings

Email findings to accepts one or more users. When to email offers On new findings and Every run.

On new findings is the default. It sends when something new appears and also when a run is blocked. Every run sends after every completed run. The email includes new, still-open and cleared findings with their numbers.

Each recipient receives the email in that person's language. Supported languages are English, Spanish, French, Portuguese, Indonesian, Czech and Chinese.

Open cases from findings

Use Open case on a finding to create a case in a case design you choose. The case carries which analysis produced the finding, which check found it, which group it concerns, and the number behind it. See case design for how cases are structured.

Opening the same finding again returns the existing case and shows Case already open. A first opening shows Case opened.

A repeating analysis can open cases automatically. Set Open cases in to a case design. Each new finding then opens one case, subject to the cap shown under Follow-ups it started. This gives the finding an owner, a task and a history so someone can act on it.

What can start an analysis

An analysis can start in four ways:

  1. A person starts it once from the portal.
  2. A daily, weekly or monthly schedule starts it with rolling dates.
  3. The management API starts a run or creates a schedule. These are the same two operations the portal uses, and the user needs write access to the dashboard area.
  4. No other trigger starts it automatically.

There is no event trigger. A new release, deployment, webhook or arriving data does not start an analysis by itself. A scheduled feed refresh is the closest available behavior, and a new feed delivery is picked up on the next scheduled run. The AI Decision Assistant cannot start an analysis yet. See the AI register for its current scope.