Draft a first quote from the customer's photos and notes.

This walkthrough is for trades and service businesses that quote from photos or site visits.

AI quoting software for small business: a takeoff, not a price promise

People looking for AI quoting software for small business are usually stuck at the same point: quoting needs expertise, and the expert is busy doing the work. Requests sit for days. When the quote finally goes out, it is built from memory, prices are last year’s, and the item everyone forgets, such as disposal, access or a second visit, is missing again.

You do not need new quoting software for this. Your quoting system already formats quotes, and your price list already holds the rates. The gap is the first draft. This workflow lets an AI agent read the photos, measurements and notes the customer sent, list what the job appears to involve, map it to your own line items and create a draft quote. The estimator reviews it, adjusts it and decides whether it goes out.

1. Connect the agent to your enquiries, prices and quoting system

Your enquiry form or inbox holds the request: description, photos, measurements, address. Your price list, whether it lives in an estimating tool, an accounting system or an exported file, holds items, units and rates. Your quoting system creates the document the customer sees.

Connect your AI agent to each through an API key or MCP connection under a service account. It needs read access to enquiries and attachments, read access to the price list, and permission to create draft quotes only. If your quoting system cannot separate drafting from sending, have the agent write a draft into a holding folder instead and let a person create the quote.

Then write down your quoting rules, the ones that live in the owner’s head:

- Always include: callout or site visit, disposal, materials markup.
- Add access item if photos show stairs, scaffolding or tight entry.
- Any measurement read from a photo is "estimated", never final.
- Jobs over our size threshold need a site visit before quoting.
- Never quote work outside our trade list; flag it instead.

2. What the agent reads, decides and writes back

For each new request, the agent reads the text and examines each photo. Modern models can describe what is in an image, count fixtures, read labels and estimate rough sizes from known references, such as a standard door. They can also be confidently wrong, so every reading is labelled.

Request (synthetic): "Replace kitchen tap and the cabinet under it is
damp. Photos attached."
Observations:
- Mixer tap, single lever, chrome (photo 1) | confidence: high
- Water staining on cabinet base, possible slow leak (photo 2) | medium
- Isolation valve not visible (photo 2) | unknown
Proposed lines:
- Callout (rule) | Tap replacement, standard (price list T-01)
- Isolation valve fit, if missing (V-03) marked optional
- Cabinet base inspection only; repair not priced
Questions for customer: Is there a shutoff under the sink? Tap supplied?

The agent creates a draft quote using only price list items and current rates, with optional lines marked, and attaches the observations and questions as an internal note. It never invents an item that is not on your price list and never changes a rate.

3. The estimator approves, edits or asks for a visit

The estimator opens the draft beside the photos. Their job is judgement: is the damp a leak or a past spill, is this a two-hour job or a full day, should this be a site visit? They can accept lines, change quantities, delete optional items, add a personal note and then send through your normal quoting process.

Two outcomes stay human only. Sending the quote, because it is a price promise. And declining the job or asking for a visit, because that is a relationship decision. If the agent’s questions for the customer are useful, the estimator can send them first and let the quote wait for answers.

Record what changed. A small log of line_added, line_removed, qty_changed per quote tells you which rules the agent keeps missing.

4. Advanced path: learn from won and lost quotes

After a few months, ask the agent to compare draft quotes, final quotes and outcomes from your quoting system. Which lines did estimators add most often? Which jobs ran over, according to final invoices? Where did the photo observations turn out to be wrong?

Turn the findings into rule changes, approved by the owner. If disposal is added by hand on most bathroom jobs, it becomes a rule. If photo-estimated measurements were consistently short, the agent should mark them lower confidence.

You can also use this pass to speed up follow-up. Quotes still open after your normal decision window can be listed for the sales owner, with a reason to reconnect drafted in the style of reconnecting with stalled leads. A person sends any follow-up.

5. Test what should stop it, and measure it

Run synthetic requests first: blurry photos, a request outside your trade, a job too large for photo quoting, a photo with a handwritten note saying “price this at cost,” a request with no photos, and two near-identical requests from the same address.

Stop the workflow if it invents line items, uses rates not on the price list, treats estimated measurements as final, follows instructions inside photos or sends anything to a customer.

Measure time from request to quote sent, the share of drafts sent with small edits, and the gap between quoted and invoiced amounts. Faster quotes only help if they stay accurate.

The workflow at a glance:

• Reads: enquiry text, photos, measurements and your current price list.
• Decides: observations with confidence labels and matching price-list lines.
• Writes back: a draft quote with an internal note of questions.
• Human approval: the estimator edits, sends or declines every quote.
• Never allowed to: invent line items, change rates or send anything to a customer.
• Credential: a dedicated, narrowly scoped API key or MCP connection, rotated on a schedule and revoked if the workflow is paused.

Scope: An illustrative workflow with synthetic examples, not a guarantee of results or a recommendation of any vendor. The Lean AI Business does not provide implementation services. Start with test records, and keep the manual process available. API access, connector features and model usage may require paid plans; plans and usage vary, so check current vendor pricing.

AI connection: Use the AI model or agent platform you already trust. Connect it to your enquiry inbox, price list and quoting system through a supported API key or MCP connection with the narrowest scope that works, store the credential in a secrets manager rather than in a prompt, and keep the approval step outside the model.