A practical approval workflow for faster quotes, clearer scope and protected margins
To automate proposals without losing pricing control, separate preparation from authority. Let the workflow collect verified details, select an approved template, calculate from a controlled price source and assemble a draft. Keep discounts, margin exceptions, unusual scope, contract terms and the final customer commitment behind named human approval.
That distinction turns proposal automation into a dependable business process rather than a faster way to send mistakes. The aim is not to let a language model invent a persuasive price. It is to remove repetitive copying, surface missing information and give the right person a complete, traceable draft to review.
For the upstream meeting stage, see how to confirm decisions, owners and follow-up before they enter a proposal or delivery workflow.
Why proposal automation is useful now
Proposal writing sits naturally within current business AI use. The Department for Science, Innovation and Technology reported that 85% of businesses already using AI used natural-language processing or text generation. The 2026 UK Business Data Survey also found businesses commonly using AI to summarise internal information or draft reports and correspondence. A proposal is therefore an obvious candidate—but it combines prose with price, scope, customer data and contractual promises, so the control design matters more than the writing speed.
The useful automation boundary is simple: AI may help organise and draft; approved systems should calculate; authorised people decide. Current government guidance on generative AI calls for traceability, clear human responsibility and meaningful review where impact or risk is higher. The Competition and Markets Authority also makes clear that a business remains responsible when it uses AI to engage with customers.
Nine stages in a controlled proposal workflow
| Stage | Workflow support | Control retained by people | Evidence to keep |
|---|---|---|---|
| 1. Capture the request | Creates a proposal record from an enquiry and brings together the customer, opportunity, requested outcome and deadline. | Confirms the request is genuine, in scope and appropriate for the business to pursue. | Source message, owner, received time and consent or lawful basis where relevant. |
| 2. Check service fit | Matches the request to approved services and flags missing facts or conflicts with normal delivery rules. | Decides whether to bid, decline, refer or ask a qualifying question. | Fit decision, missing inputs and reason for any exception. |
| 3. Retrieve approved sources | Uses the current service catalogue, template, case evidence, standard assumptions, exclusions and terms. | Owns each source, its review date and who may change it. | Source version, effective date and permissions. |
| 4. Build the price | Runs a defined calculation using quantities, rates, units, taxes and mandatory charges supplied by approved systems. | Owns the formula, rate source, rounding policy and any price floor. | Inputs, formula version, calculation result and timestamp. |
| 5. Test exceptions | Checks discounts, margin thresholds, unusual quantities, rush work, special terms and missing delivery costs. | An authorised person approves, rejects or changes every triggered exception. | Rule triggered, approver, decision, reason and previous value. |
| 6. Assemble the draft | Creates the proposal structure, selects approved language and inserts verified customer, scope and pricing fields. | Checks that the narrative matches what can actually be delivered. | Template version, included sections, source links and draft version. |
| 7. Review the commitment | Presents a concise difference view: price, scope, assumptions, exclusions, terms and every exception. | Approves the complete commercial promise, not just the wording. | Reviewer, approval time, changes and final version. |
| 8. Send and follow up | Sends the approved version through the correct channel and prepares measured follow-up reminders. | Controls timing, tone, negotiation and sensitive customer situations. | Exact sent file, recipient, delivery status and follow-up history. |
| 9. Accept and hand over | Records acceptance and creates delivery, billing and onboarding actions from the approved proposal. | Confirms valid acceptance, scope changes and the operational start decision. | Accepted version, acceptance evidence, change record and handoff owner. |
Use three control levels, not one approval button
A single “approve proposal” step hides too much. Better proposal automation classifies decisions by authority:
- Allowed automatically: copy verified customer fields, choose the current template, apply standard rates, calculate totals, include standard assumptions and prepare a draft.
- Must escalate: discount requests, price below a floor, non-standard scope, accelerated delivery, unusual payment terms, manual price overrides, missing inputs or conflicting customer information.
- Never decide alone: whether to make an unapproved promise, accept unbounded liability, conceal a mandatory charge, invent evidence, waive a contractual protection or send a high-impact exception without authority.
The reviewer needs enough context to make a real decision: the customer request, approved source, calculated result, proposed exception, effect on margin and the exact wording that will be sent. A person clicking approve without that context is not meaningful control.
Build a proposal source pack before adding AI
Most proposal problems begin upstream. If rates live in several spreadsheets, service descriptions disagree and nobody owns the terms, automation will reproduce the confusion more consistently. Create one controlled source pack with:
- service names, inclusions, exclusions and delivery assumptions;
- current rates, units, quantity breaks, taxes and mandatory charges;
- margin floors, discount limits and named approval authority;
- approved proof such as case examples, accreditations and capability statements;
- standard payment, cancellation, expiry and change-control terms;
- brand language, proposal structure and accessibility requirements;
- an owner, version, effective date and scheduled review for every source.
Only provide the customer information needed for the proposal. ICO guidance describes data minimisation as using information that is adequate, relevant and limited to the purpose. A proposal tool rarely needs an entire mailbox, unrestricted drive access or every historic customer record.
Protect pricing for different commercial models
| Commercial model | Approved calculation | Common escalation |
|---|---|---|
| Fixed fee or package | Current package price plus selected options and mandatory charges. | Custom deliverable, missing dependency, rush date or discount below authority. |
| Hourly or day rate | Approved role rate multiplied by agreed effort, with expenses and tax handled explicitly. | Uncertain effort, blended rate, capped fee, travel or work outside normal assumptions. |
| Retainer or recurring service | Approved recurring charge, included allowance, term and clearly stated setup cost. | Minimum term change, rollover, overage, cancellation or bespoke service level. |
| Usage or quantity based | Versioned unit rate and quantity bands with a visible method where the total cannot yet be calculated. | Forecast uncertainty, minimum commitment, exceptional volume or non-standard unit. |
For proposals to consumers, current CMA guidance says prices should be clear, complete and accurate, including mandatory charges, and contract terms should be fair and transparent. Business-to-business arrangements can involve different law and context, but clear scope, price and terms still reduce disputes and make the handoff safer. Obtain appropriate legal or regulatory advice for your market rather than treating a general workflow as legal advice.
Prevent wrong-source and prompt-injection risks
An enquiry, uploaded brief or third-party document is untrusted input. The National Cyber Security Centre warns that language models can be influenced by prompt injection and recommends designing systems around the worst case of what the model is permitted to do. A customer attachment should not be able to change a rate, select a hidden term, access another customer’s data or trigger an irreversible send.
Keep calculation and authority outside the generative step. Restrict the proposal tool to the minimum approved sources; label external content as data rather than instructions; validate fields; prevent the model from editing price rules; require human review before sensitive actions; and log the exact source, draft and approval. If the remaining risk is unacceptable, keep that proposal type manual.
Run a 20-proposal pilot
- Choose one frequent, low-variation proposal type with a clear owner and current template.
- Review 20 recent proposals and list every input, correction, exception, approval and handoff.
- Build the source pack, price calculation and exception rules before connecting draft generation.
- Run in draft-only mode. Compare each generated draft with the human-approved final version.
- Test missing fields, old prices, discounts, unusual scope, malicious instructions and access boundaries.
- Keep a named reviewer and record why each exception was approved or rejected.
- Expand only after the error pattern is understood and the control rules work consistently.
Measure quality and commercial control together
Do not judge the pilot only by how quickly a PDF appears. Track request-to-first-draft time, percentage of drafts with all required inputs, correction rate, unapproved price attempts, exceptions per proposal, approval waiting time, sent-to-accepted time, post-sale scope changes, billing corrections and gross-margin variance against the approved proposal. Review complaints and lost deals qualitatively; not every useful signal is a conversion percentage.
A good result is a faster, calmer process with fewer avoidable revisions and a cleaner delivery handoff—without weakening price, margin or customer trust. The team should spend less time copying and chasing, while retaining the judgement that protects the business.
Frequently asked questions
Can AI automate a complete proposal?
AI can assemble a controlled first draft from approved services, customer facts, templates and pricing inputs. A named person should still approve unusual scope, discounts, margin exceptions, legal or regulatory wording, payment terms and the final promise before the proposal is sent.
How do you stop an AI proposal from using the wrong price?
Keep prices in a version-controlled source rather than in prompt text, require the workflow to record the source and effective date, block expired or incomplete inputs, recalculate totals deterministically and route discounts or margin exceptions to an authorised approver.
Should proposal automation send quotes automatically?
Begin in draft-only mode. Automatic sending is only sensible for a tightly bounded, tested proposal type where all required inputs are present, no exception is triggered and the business is comfortable with the remaining risk. Sensitive or unusual proposals should retain human approval.
What information should an automated proposal use?
Use the minimum approved information needed: verified customer details, service catalogue, scope rules, current prices, mandatory charges, assumptions, exclusions, terms, brand language and approval authority. Do not expose unrelated inbox history, confidential files or old price sheets.
What is the best first proposal to automate?
Choose a frequent, low-variation proposal with known inputs, a current template, clear ownership and enough recent examples to test. Avoid the largest or most bespoke deal first. Run a contained pilot, compare every draft with the approved result and expand only after the controls work reliably.
Sources and further reading
- Department for Science, Innovation and Technology: AI Adoption Research
- Department for Science, Innovation and Technology: UK Business Data Survey 2026
- UK Government: Generative AI framework
- Competition and Markets Authority: Providing clear and accurate information about prices
- Competition and Markets Authority: Unfair contract terms
- Competition and Markets Authority: Complying with consumer law when using AI agents
- Information Commissioner's Office: Data minimisation
- National Cyber Security Centre: Prompt injection is not SQL injection
The practical next step
Take 20 recent proposals and mark where somebody copied information, searched for a rate, corrected scope, waited for approval or repaired the delivery handoff. Choose one common proposal type, identify its source pack and name the people who own price, scope and terms. Then review Motif proposal and quote automation, use our first-process scorecard or plan a controlled proposal pilot with Ostina.
