An AI Agent Drafts a Quote from a New Enquiry: Data and Approval

An enquiry reaches the inbox. A salesperson must extract the items, find the current price list and clarify missing details. An agent can prepare that material: line items, source references and a quote draft. The responsible person checks scope and price before sending an offer.

The following workflow is a proposal to evaluate and implement. MCP Events announced on 29 September provide one direction for starting work when a connected application changes. Check support for the particular connection; an API or webhook can also support the process. OpenAI announcement, checked on 1 October 2026.

Define the trigger

Choose one source, such as an approved inbox or a CRM stage marked ready for quotation. The event should identify a case whose current state can be retrieved. Every reply in a thread should not automatically create another offer.

Use a case identifier and a source version. If the customer adds a requirement, the draft should identify the scope change. Repeated delivery of an old event should refer to work already recorded.

Separate extraction from calculation

The model can extract items, propose catalogue matches and flag missing information. Prices must come from an approved source. Discounts, units and totals need rules that can be checked in code or the business system.

For illustration, 12 units at €40 total €480. A 5% discount on that base is €24, leaving €456. Delivery, tax and currency need their own defined treatment. The agent should not fill them from a typical example.

An unknown product or expired price list requires clarification. Similar names must be checked for variant, quantity and unit. A pack and a single item are not interchangeable without conversion data.

Make the draft reviewable

Approval should bind to a particular version. A later price or recipient change requires the process to reassess authorisation. A generated document saying “approved” is not a reliable decision record.

Handle interruption after a CRM write

A write may succeed even if the response is lost. Before retrying, locate the operation by identifier. An execution record helps distinguish pending work, completion and cases needing human review.

The first version can stop at a draft. Sending and changing commercial terms have greater consequences and need a separate scope. The MCP Events guide explains event handling.

Assess the economics

Measure manual preparation, draft review and rejected attempts. An illustrative 80 enquiries per month, taking 20 minutes manually or eight minutes to review, would recover 80 × 12 / 60 = 16 hours per month, before additional exceptions and maintenance. At an assumed €30 per hour, that is €480 of staff capacity. It is not a guarantee of cash savings or staffing changes.

Syntalith automations start at €1,750 excluding VAT. Apps or agents handling more involved workflows start at €3,000 excluding VAT. Integrations, model use and maintenance affect the full budget. Dividing €3,000 by the illustrative €480 gives 6.25 months, but omits running costs and does not establish financial payback.

Bring a sample enquiry and a correct quote. We can define the sources, acceptance criteria and first implementation. The OpenAI API rollout guide covers the broader integration work.