Introduction Guide
Starting with Decisimo means bringing your policy. You upload the document or describe the decision, and the assistant builds a working decision from it for you to review and approve. You connect your own database and the outside data you already use. Once decisions are being recorded, those records and their outcomes become the material for the next revision.
Prepare your policy
Have the policy ready as a PDF or a spreadsheet, up to 50 MB, or describe it in your own words, or do both. A Word document is saved as a PDF first.
Decisimo reads the document and extracts its requirements. It reports how many pages it processed and how many requirements it found, so you can check what went in. Then it works out whether the policy describes an instant decision or a case that people work.
For an instant decision, the build gives you:
- A decision flow that represents the policy.
- Test data for trying the decision.
- An endpoint for sending a decision request.
- An endpoint binding.
- A release of the decision flow.
- A deployment.
If the policy describes work people do, the assistant asks whether you want to connect a database for cases or build an instant decision that refers the work to a person.
Change it with the assistant
Every decision flow has an “AI Assistant” button, and the chat opens beside a live preview of the flow. The same button sits in the case studio. Tell the assistant what you want to change, review its plan, and approve or skip each proposed change before it generates anything.
The preview marks additions green, modifications orange and removals red. The flow keeps a change only when you save it.
On a decision flow, the assistant can build or change:
- Rule sets, decision tables, decision trees, scorecards and functions.
- Data sources and flow steps.
- Routing and forks.
It can also remove steps and explain what the flow does. On a case design, you can ask it to change tasks and record types, set the decisions and conditions for completing a case, or configure the decision logic, AI model or external data a task uses. You approve and save each change there too, and you release the case design yourself.
You can drop a changed policy document into the chat on an existing decision flow. Voice input is also available in the flow assistant.
Connect your own database
Under “Data destinations” you connect your PostgreSQL database and choose which records it receives. Paste a connection string and select “Parse”, or enter the host, port, database, username, password and SSL mode, then select “Test connection”. For Supabase, use its Session or Transaction pooler connection string.
| Destination | What it holds |
|---|---|
| Decision archive | One row per decision: its input and output, the decision flow, the endpoint, and the times it was received and answered |
| Case board | The cases, tasks and events of your case work, kept in your database only |
| Outcome history | What happened to each decision, filled in by you |
| Data cache | Stored responses from external data sources, 30 days by default |
The decision archive is written after the response is sent. If that write fails, it is queued and retried, so a decision does not fail because of the archive write. Case board data lives only in your database. Decisimo creates its own tables and leaves your tables alone.
Connect the data you use
Outside data comes in through a REST API or SOAP service on the public internet. Sources can return JSON or XML and use HTTP Basic or OAuth2 authentication. Each source has a 15-second timeout.
Connect each provider as its own data source. The “AI Assistant” button on the data source screen proposes the configuration and checks the method, the authentication and the credentials before you approve it.
Enter a credential in the credential box. It goes to the vault, and the chat carries only a reference to it. If you paste a credential into the chat by mistake, Decisimo redacts it before it reaches the assistant.
Select “Run test” to call the source. A flow step can cache responses in your Data cache destination, which keeps them for 30 days by default. The flow assistant can configure a data call, and the case assistant can configure the external data a task uses.
Learn from your own decisions
Analysis reads the Decision archive in your database, and questions about outcomes read the Outcome history you fill in, so this comes after the flow has been deciding for a while. Backtest runs, mining runs and policy mining each have their own assistant chat.
Backtest
Run a release against your recorded production decisions and see which decisions would change, beside what actually happened.
Rule mining
Mine outcomes for candidate conditions. The results show each candidate’s bad rate and lift, and estimate how many extra bad cases it would catch. Every cell of an analysis result offers “Mine these rows”.
Policy mining
Policy mining can produce a scorecard with bins and points, and a cutoff curve with its stability. A recorded decision path can recover the policy behind recorded decisions, and you can build that recovered policy as a decision flow.
Impact analyzer
Compare two revisions of a rule set, decision table, decision tree, scorecard or function. You can run the comparison over an uploaded file or the Decision archive.
After a backtest and a mining run you propose a new revision, and the proposal queues its own backtest against the same records.
Frequently asked questions
Can I upload a Word document?
Save it as a PDF first, then upload the PDF.
Does the assistant change anything on its own?
You approve or skip each proposed change, then generate and save it. You release a case design yourself.
Do I need my own database for the first decision?
The first decision runs without one. The Decision archive, the case board and analysis live in your database, so connect one when you want those.
Where do I start?
Upload the policy you already use, or describe it in your own words. The assistant builds the first decision from it.
