Automate the preparation around a quote: gathering requirements, finding the relevant service items, calculating from approved rates, and assembling a draft. Give a named person responsibility for checking scope, approving the price, and deciding when the quote can go to the customer.
For a small business, that can be a useful first AI project. The work has a clear beginning and a result someone already knows how to review. The boundary matters: a tidy document is only useful if its quantities, assumptions, and prices are dependable.
Here is a practical way to design that workflow, using a hypothetical commercial-cleaning business. The example is an illustration, not a Codi client story or a claim about measured savings.
Start with one service and a clear trigger
Suppose an enquiry arrives asking for weekly cleaning at a small office. Today, an administrator reads the email, checks the service list, asks about missing details, copies rates into a template, and sends the draft to the owner.
Keep the first version limited to that standard service. Exclude specialist cleaning and jobs that require a site inspection until the team has a dependable process for them.
The trigger is a new enquiry marked ready for quote preparation. Before a draft can proceed to pricing review, it needs the site, the requested service, the quantity or pricing basis, and the proposed frequency. Your business may need different fields. Agree on them with the person who quotes the work.
If you are still comparing possible projects, start with how to choose your company’s first AI workflow.
Separate what the customer said from what the system assumes
AI can help turn an email into a consistent set of fields. Each extracted detail should point back to the sentence or attachment it came from. Missing information should stay visibly missing.
For example, “weekly cleaning for our office” does not establish the floor area, the tasks included, or how long a visit will take. The workflow can prepare clarification questions, but it should not silently fill those gaps from a similar customer’s job.
Use three simple labels in the draft:
- Confirmed: stated by the customer or verified by a staff member.
- Missing: needed before the quote can proceed.
- Proposed: an assumption a person must explicitly accept or replace.
These labels make review more useful than a single confidence score. The reviewer can see exactly which decisions remain open.
Structured output can make extracted fields consistent, but it does not establish that their contents are correct. OpenAI’s documentation explicitly notes that structured outputs can still contain mistakes. Source checks and business validation remain part of the job.
Treat text inside customer emails and attachments as information to interpret. A request to “skip approval” or “use last year’s discount” must not change the workflow’s permissions or pricing rules.
Calculate from an approved price source
Use the current approved rate table, with an identifiable version and effective date. A conventional calculation should produce the line totals and subtotal. AI can help suggest which service item fits the enquiry, while a person checks that selection and the quantity behind it.
In this hypothetical example, a staff member has confirmed a three-hour weekly visit after reviewing the scope. The approved illustrative rate is $60 per hour:
| Item | Quantity | Rate | Amount |
|---|---|---|---|
| Standard office cleaning | 3 hours per visit | $60 per hour | $180 per visit |
The arithmetic is 3 × $60 = $180 per visit. That is an illustrative service subtotal. It excludes any taxes, travel charges, materials, or other additions that the business would need to establish before issuing a real quote. Weekly service should not silently become a fixed monthly price based on an assumed four visits.
If no approved service item matches, stop for manual scoping. If the rate table is unavailable, leave the price unresolved. Do not recover a price from an old email or ask the model to invent a reasonable market rate.
Discounts, unusual payment terms, and exceptions to the service scope belong in an approval path with a named decision-maker.
Give exceptions a place to go
Define the next action for an incomplete or unusual enquiry before testing the workflow. Otherwise, the reviewer becomes responsible for finding every hidden problem in a polished quote.
| What the workflow finds | Next action |
|---|---|
| Required scope information is missing | Hold pricing review and prepare clarification questions for staff |
| Two attachments give different quantities | Show both sources and ask a person to resolve the conflict |
| A service has no approved rate | Route to manual scoping and pricing |
| The customer asks for a discount | Show the request separately for the authorized pricing reviewer |
| The same enquiry arrives twice | Link it to the existing quote record for review |
| An approved quote is edited | Return the new revision for approval before release |
None of these paths needs to send a customer message automatically. The first pilot can keep clarification and delivery in the existing staff process.
Use the First AI Project Kit to map your own version. It includes a playbook and editable worksheet for recording the trigger, inputs, reviewer, and exception paths.
Make approval control the next action
The pricing reviewer needs a small review packet, not just an “Approve” button. Put the original enquiry beside the proposed quote and show:
- The exact quote revision being reviewed.
- The service scope, quantities, and unresolved assumptions.
- The rate-table version and calculated line items.
- Any requested discount or departure from standard terms.
- The intended customer and delivery destination.
In the cleaning example, the owner should be able to check why the draft uses three hours, where the $60 rate came from, and what the $180 excludes without searching through several systems.
Human approval can be built into a workflow. Microsoft documents an approval action that waits for an approver’s response before continuing. The next step still needs explicit conditions: a rejection returns the quote for correction, and a missing response leaves it pending. Waiting for a response alone does not enforce your pricing policy.
Record who approved which revision and when. Approval should apply to that revision only. If someone changes the scope, rate, quantity, or recipient afterward, require a fresh review. Keep delivery as a deliberate staff action during the first pilot, and record when the approved quote is actually sent.
Test the complete job before expanding it
Run the workflow against a small set of historical enquiries your team is permitted to use. Include complete requests, missing information, conflicting attachments, discounts, duplicate submissions, and changes after approval. Keep test messages and quotes out of customer channels.
Compare each result with a staff-prepared reference. Check whether the workflow preserved the source facts, chose the right service, used approved rates, calculated correctly, and routed exceptions to the right person.
Measure hands-on preparation, review, and correction time together. Also track the share of enquiries that need manual handling and any errors that reach the delivery stage. A faster first draft can still make the overall job slower if it creates more checking work.
Agree on the acceptance criteria before judging the pilot. Any quote released without the required approval, or priced from an unauthorized source, is a reason to stop and investigate. Keep the existing manual process available while the team learns which cases the workflow can handle.
Scope the first build around a reviewable result
You have a useful starting brief when you can name the service, required inputs, approved rate source, pricing reviewer, exception paths, and delivery step. Add a baseline for the whole job and a small set of examples that demonstrate what should happen when information is incomplete.
Some businesses will need only a form, a rate lookup, and a quote template. AI becomes useful where varied enquiries need interpretation or a draft needs assembling from several sources. Choose the simplest approach that meets the brief.
If your team spends more time untangling enquiries than making pricing decisions, talk with Codi about the workflow. Start with one service and a review packet your team can trust, then use the pilot results to decide what deserves automation next.