From email to quotation: where AI can help your business
The request is in an email, the dimensions in an attachment, and the material is missing. Here is how AI can prepare a brief for sales, flag gaps and leave prices and delivery commitments for approval.
A customer sends an email with two attachments and asks for “price and delivery, please.” A salesperson opens the drawing, searches earlier messages for the quantity and checks whether the quote should include a surface finish. Before calculating a price, they have already spent time assembling the requirements. This preparation is a useful place to try AI.
Complete the request before pricing it
The proposed output is a working brief for sales: what the customer wants, where each detail came from and what still needs clarification. The system should distinguish the customer's request from the company's confirmed commitment. Requested delivery by Friday does not establish available production capacity.
Start with one recurring type of request, such as brackets manufactured to a drawing. Define the information your estimator needs. The process owner should also identify the situations in which preparation stops and a person takes over.
A fictional request: 240 brackets and an unclear deadline
This example is fictional. An email requests 240 brackets to the attached revision C drawing, with delivery “by Friday.” The second attachment is an older quotation. Neither the material nor the surface finish is clearly specified.
- Quantity: 240 pieces, source: the current email.
- Manufacturing reference: the attached drawing marked revision C; the responsible employee confirms its use.
- Deadline: the customer requests “by Friday”; the exact date and feasibility need confirmation.
- Material and finish: clear information is missing; prepare a follow-up question.
- Older quotation: historical context, not an automatically valid price or specification.
This brief helps sales decide what to clarify. It should not look like a completed quotation. If an attachment is unreadable, the output must say it could not be processed rather than fill the gap with a plausible value.
What to check in attachments
Define an approved source and a checking method for each field. A filename does not prove which revision is current. If the email and drawing conflict, show both values and the discrepancy. Preserve units and decimal separators in dimensions; a missing value must not become zero.
Begin with readable emails and a familiar PDF format. CAD processing, engineering assessments of manufacturability and automated calculations require separate design work. Successfully extracting document text does not establish those capabilities.
Microsoft's October article on document processing recommends linking extracted values to the original document and using confidence scores to inform review, not guarantee correctness.
Make pricing and sending separate controlled steps
AI can draft a question about material or summarise the requirements. An approved calculation process or responsible person must supply the price. Delivery dates, quotation validity and exceptions also need checking. A historical price must not carry over to a new job without review.
Approval should be a visible system state. Microsoft Power Automate supports an action that waits for an approver's response. Approval can therefore be a process step rather than just a prompt instruction.
In the proposed pilot, keep every customer message as a draft. Treat attachment contents as input data: a sentence in a PDF must not authorise the system to send a quote or disclose other jobs. Restrict access to the people and records needed for that request.
Measure the work sales actually does
Prepare a sample of routine and difficult requests: missing material, two revisions, an unreadable scan, different units and a conflict between the email and attachment. A person should establish the expected result beforehand. Compare preparation time, including corrections, with the current process.
Track field accuracy, detected discrepancies and errors missed by the review. Faster text generation with the wrong quantity is not success. Requests the system cannot handle must reach a named employee with their original attachments.
Where to start in your company
Write down one request type, the required fields and the person responsible for approving the quotation. Then consider connections to email, storage and estimating tools. Informio helps companies design AI implementations; this workflow is a proposal for a custom solution, not a ready-made email or ERP feature automatically included in the AI chat.
Arrange a consultation with Informio. Using an anonymised request, we can examine which steps AI could prepare and which should remain with your team.