Conversation summary
Condense relevant customer history with source references.
AI recommendations
See what matters and reach the next step faster
After the exact AI configuration has passed its release evaluation, OfferAgent can summarize a customer conversation, draft a reply, suggest a next action, qualify a lead or propose follow-up from a bounded evidence package. It shows sources, rationale, confidence and abstention, while the seller reviews the suggestion and the relevant product engine performs any real action.
AI should reduce interpretation, not hide it
A recommendation without evidence or boundaries is difficult to trust in customer-facing work.
How it works
- 01
Choose the task
Ask for a summary, reply draft, qualification, follow-up or next action.
- 02
Build bounded evidence
Collect only the account-scoped facts required for that task.
- 03
Generate with provenance
Return structured output, sources, rationale and confidence or abstain.
- 04
Review and act
Let a person edit or approve before another engine performs the action.
Product proof: evidence before recommendation
The AI panel separates source facts, generated output, confidence and human decision.
Prerequisite: Requires the AI add-on, an enabled provider and sufficient account-scoped evidence.
AI recommendation panelProduct flow
- 1Evidence assembled
- 2Recommendation explained
- 3Human review required
Open the signed-in product workspace
Sign-in and the correct plan are required. This link is not a public demo.
What this capability supports
Reply draft
Prepare editable wording without sending it.
Next action
Recommend a bounded seller step and explain why.
Qualification and follow-up
Structure a score or timing suggestion from declared evidence.
Prerequisites and clear boundaries
OfferAgent states what the capability does and what still requires configuration, permission or a human decision.
- AI does not send, price, accept or change customer records by itself.
- The layer abstains when required evidence is missing or contradictory.
- The runtime stays fail-closed until the exact provider, model, prompt, policy, recommendation type and locale pass their release evaluation.
- Recommendations depend on enabled AI access, provider availability and account policy.
Frequently asked questions
Can AI send the reply?
No. It creates an editable proposal; the seller controls sending.
Can AI change a quote price?
No. Pricing remains governed by deterministic profiles and the revision workflow.
Can we see why a recommendation was made?
Yes. The result includes evidence references, rationale and a confidence indication.
What happens when evidence is weak?
The layer can abstain and state what information is missing.