Commercial Proposal Draft

The automation prepares a text draft of a commercial proposal based on deal data in the CRM and a set of templates from File storage. Instead of the manager starting from a blank page or copying a previous proposal, the AI agent assembles the document structure, inserts relevant blocks, and returns the file for final review. Average preparation time drops from 2 hours to 15 minutes.

What the automation does

1. Receives a signal about a new deal in the 'Proposal Preparation' status from the CRM.
2. Retrieves the client card: company name, contact person, industry, request, budget estimate, communication history.
3. Determines the proposal type by product or service and pulls the corresponding template from File storage.
4. Assembles the document structure: introduction, description of the client's task, proposal, scope of work, timelines, price, payment terms, contacts.
5. Generates text in the corporate tone of voice, accounting for the client's specifics and previous touchpoints.
6. Saves the draft in the deal card in the CRM and in the client folder in File storage.
7. Notifies the manager with a link to the ready draft for review and sending.

The result is a text document that requires edits covering 10-20% of the content: price clarification, adding specific examples from practice, a final tone check. The manager remains the author of the proposal but does not spend time on the routine assembly of the structure and copying data from the CRM.

What the automation does not do

• Does not send the proposal to the client without manual confirmation from the manager. The final send always goes through a human.
• Does not negotiate prices or discounts. Pricing decisions remain with the manager or sales lead; the automation only takes figures from the CRM or a pre-prepared price list.
• Does not replace the initial discovery call. If client data in the CRM is absent or fragmentary, the draft will be superficial and will require manual rewriting.

The automation is built as a low-code pipeline. The CRM serves as the data source and final storage, File storage — as the template library and draft archive, the AI agent on the AI model — as the text generator. The integration layer connects the components via webhook and API.

Technical flow

The chain triggers when the deal status changes in the CRM. The orchestrator (workflow engine or Zapier) picks up the payload with the deal and passes it to the AI agent's prompt. The agent queries File storage to retrieve the required template, assembles the final prompt with all variables, and returns structured text. The result is written back to the CRM as an attachment and to File storage as a file in the client's folder.

Solution components

Implementation steps

1. Audit of the current proposal preparation process: which templates are used, how many structure variants exist, which fields are filled in manually.
2. Building the template library in File storage with a unified structure and variables ({клиент}, {услуга}, {цена}, {срок}).
3. Setting up a webhook or trigger in the CRM for the 'Proposal preparation' event.
4. Developing the prompt for the AI agent: role, tone, output structure, rules for handling missing data.
5. Configuring the pipeline in the low-code orchestrator: CRM → data normalization → AI agent → writing to CRM and File storage.
6. Testing on 10-15 real historical deals, comparing with managers' drafts.
7. Rolling out to one manager or a pilot team, collecting feedback for two weeks.
8. Tuning the prompt and template library based on pilot results, full rollout to the department.

How the AI agent works with data

The AI model receives a JSON with the deal and the selected template as input. The prompt instructs the agent to use only the provided facts, not to fabricate figures and cases, and to mark empty fields with the {требует уточнения} marker if data is missing. Tone is set through examples of the company's previous proposals. For sensitive fields (price, timelines) the agent does not generate values itself, but inserts exactly what came from the CRM.

Output format

The draft is saved in two places: as an attached file to the deal card in the CRM, and as a document in the client's folder in File storage. The manager gets quick access while working with the deal; the sales manager sees the history of all proposal versions in one place.

The baseline set of conditions under which automation delivers the stated effect of reducing proposal preparation time.

Data and access

• CRM with filled deal cards: client, service, contact person, request, budget reference. Empty fields directly reduce the quality of the draft.
• File storage with a proposal template library in a unified structure. Minimum 3–5 options for the main service or product types.
• API access or webhook in CRM for reading deals and writing attachments.
• Access to the AI provider API (Anthropic for the AI model).

Process readiness

• The sales department has a formalized proposal preparation process: there is an understanding of which template applies in which cases.
• The fields that the manager fills in CRM before moving to the proposal stage are defined (minimum: product, volume, base price).
• An automation owner on the client side is agreed upon: who is responsible for updating templates and prompt edits after launch.

Team

• Sales lead or manager — process owner and requester.
• 1–2 managers for the pilot and feedback on draft quality.
• CRM technical owner or admin with rights to configure webhooks and API tokens.

Implementation timeline

A typical project takes 2–4 weeks: one week for audit and template collection, one week for pipeline and prompt assembly, 1–2 weeks for the pilot and tuning. The timeline increases if the template library is not yet structured or CRM requires preliminary data cleanup.