Decision trees aren't flowchart diagrams — they're branch instructions attached to each question. The AI reads them as natural language and applies them. No code, no if-then builders, no visual editors required.
Each question in the campaign has an optional branch value — what the AI should do when a contact gives a negative answer. These are translated into plain English and injected into the system prompt alongside the question set. The AI applies them as part of its judgment — no runtime decision tree is traversed by code.
The decision tree summary gives operators a plain-English view of the full qualification logic. The Campaign Tester lets you verify every branch before launch.
Alongside the question list in campaign configuration, a structured plain-text summary shows the full decision tree: each question, its positive path, and its branch outcome. No database fields to interpret.
How it works: Summary regenerates automatically when questions or branch values change — always current, always accurate.
The summary accurately represents both structured tokens (END → 'end conversation') and freeform prose instructions (shown verbatim as an Instruction line). What the operator reads is exactly what the AI receives.
How it works: Tested to confirm freeform branch instructions use Instruction prefix in summary — no loss of meaning in translation.
The Campaign Tester includes branch shortcuts — buttons that let you simulate a negative answer to a specific question without manually walking through the full conversation from Q1. Test every branch path in minutes.
How it works: All test turns are flagged is_test=True — branch testing never contaminates learning engine analysis.
Every change to the question set or branch values creates a new campaign version. You can compare what the AI received in any prior conversation to the current configuration — and revert to any prior version if a change made performance worse.
How it works: from_proposal_id is recorded on versions created by learning proposals — trace any version back to the evidence that justified it.
We'll configure your question set, write the branch instructions, and test every path with you.