All posts

Research / Revenue operations

Automate Quote Preparation Without Automating Commercial Authority

Build an AI quote-preparation workflow with approved product facts, pricing rules, missing-information checks, commercial approval, CRM controls, and delivery handover.

By Waypoint ExponentialPublished Revised
Loose teal blocks form an assembly with a missing slot and a terracotta cube, while a brass gate separates it from a finished block on a cream plinth

An AI quote assistant can collect requirements, find approved product information, and prepare a draft for review. Keep calculations in the agreed pricing system, route exceptions to named owners, and bind approval to the exact quote revision you send. Complete the workflow with a delivery handover so a faster proposal doesn't create an unclear operating commitment.

Choose which quotes enter the workflow

Start with a quote family whose commercial rules you can explain. A standard renewal or a defined service package may fit. An unfamiliar installation requiring a site assessment needs a different path. Define the eligible customers, products, and transaction types before connecting the assistant to the quote queue.

Write down the output and its recipient. An internal requirements summary, a priced draft, and a customer-ready offer each carry different expectations. State which the first release produces, where staff review it, and who can move it to the next state.

A recent SalesOps discussion about AI quoting asks what happens when a customer requests an exceptional discount, combines bundles, or changes a billing schedule. Those questions identify cases to investigate in your own process. A forum account doesn't establish how often your company encounters them or which tool handles them correctly.

Inspect recent completed quotes alongside deals that stalled or needed correction. Follow the time spent finding information, preparing the offer, and waiting for decisions. If the largest delay comes from an unowned capacity decision, faster drafting alone won't resolve it.

Define the workflow states in terms staff recognise: missing information, draft, awaiting approval, approved, issued, accepted, and handed over. Record who can make each transition. Give an incomplete request a visible owner and a next question rather than letting it disappear into a generic pending queue.

Check what the installed CRM and quoting tools already support. Configure, price, quote software, usually called CPQ, may already handle the catalogue and approval rules. Your AI work can focus on interpreting the enquiry and assembling evidence around that process. Assess the actual licence and configuration before adding another system.

Build the draft from approved facts and open questions

Create a source map for every field that matters. Product identity comes from the approved catalogue. Customer eligibility comes from the appropriate account record. A delivery date needs the authorised capacity or availability source. Assign an owner to each source and define when the workflow must refresh it.

Use an old quote to find context, but verify its product, terms, and price against the current approved sources. A previous exception may apply to a different customer entity or period. Keep the accepted historical document available without turning its contents into today's policy.

Separate customer statements, confirmed company facts, and suggestions. A customer asking for delivery next week has stated a preference. The assistant mustn't store that preference as a promised date. Show the source message and the person who can confirm the commitment.

Consider an illustrative inspection-services firm. A customer asks for the same equipment inspection as last year, adds another site, and requests an earlier completion date. The assistant can retrieve the earlier scope and current service definitions. The salesperson still needs the new site's equipment list, while the delivery manager needs to confirm access and capacity.

The internal draft can show confirmed lines and a missing-information checklist. It must distinguish a deliberately excluded item from an unanswered question. If the product choice depends on that answer, keep the affected line provisional and stop it from reaching the issued offer until the owner resolves it.

Keep retrieval within the user's permitted account and document scope. A similar proposal for another customer may contain confidential requirements or negotiated terms. Approved reusable wording belongs in a maintained template library with a named owner and version.

Treat incoming documents as evidence to interpret. Instructions inside an attachment mustn't change the assistant's access, pricing rules, or approval path. Test that boundary with a request containing conflicting instructions, and make unsupported claims visible to the reviewer.

Calculate prices through controlled rules

Let the pricing system or an explicitly maintained calculation service produce quote amounts. Supply validated product identifiers, quantities, units, customer agreement references, and the relevant currency. Validate the returned totals and keep a reference to the calculation used for that revision.

The model can extract a requested quantity or explain a price difference. It shouldn't invent a missing rate, derive a discount entitlement from persuasive wording, or calculate the authoritative total in prose. If a required rule or input is absent, route that gap to its owner.

Microsoft's current Dynamics 365 Sales documentation describes pricing from catalogue entries, line items, configured rules, discounts, and taxes. It also distinguishes draft pricing, which can refresh when a quote opens or changes, from active and closed pricing. That makes the quote state relevant to the amount a reviewer actually approved.

Apply that lesson to your own stack: record the quote revision and pricing inputs, then recheck the relevant sources at the defined approval and issue points. If a source change alters the offer, produce a new revision and send it through the affected approval path.

Make units and periods explicit. A per-site service and a per-item service can share similar descriptions while producing different totals. Recurring charges also need an agreed period and start condition. Ask finance or the commercial owner to maintain the calculation and rounding rules rather than embedding informal assumptions in a prompt.

Keep customer-specific overrides visible. Record the reason, authorised decision, scope, and expiry where applicable. A salesperson's request for a concession can enter the review pack without changing the approved price or teaching the assistant to apply the concession to later deals.

Give each approver a specific decision

Build an approval map around actual authority. The product specialist confirms technical fit when needed. Delivery accepts the proposed work and timing. The commercial owner approves the offer within their authority, with finance or other specialist review for the exceptions your policy identifies.

Route the decision and its evidence together. Show the customer entity, quote revision, relevant source records, changed terms, and the exact exception. An approver should be able to see what they are accepting without reconstructing the deal from a long email chain.

Salesforce's Advanced Approvals training describes rules with named approvers and conditions, including conditions based on quote records and their related lines. Use your business requirements to define those rules. A product's ability to route an approval doesn't determine who has commercial authority in your company.

Avoid sending every quote to every manager. Define the standard path and identify the conditions that require an additional decision. Test whether an exception affects another approver's decision before deciding which approvals a changed revision needs to repeat.

Bind each approval to the relevant revision and decision scope. If someone changes the product, delivery date, or payment arrangement after approval, the workflow must check whether that change invalidates a recorded decision. A generic approved flag on the opportunity is insufficient evidence for a different offer.

Keep sending behind the agreed release check. The check should confirm the recipient, approved revision, unresolved requirements, and required decisions. For an assisted first release, an authorised person issues the offer. Any later automated sending needs its own scope, controls, and acceptance evidence.

Define delegation and absence cover so the queue can operate when an approver is unavailable. The guide to placing AI review at the decision explains how review capacity and accountable ownership affect this handoff.

Write to the CRM without losing ownership or history

Agree a small set of permitted updates. The assistant may create a draft quote against an existing opportunity or propose a requirements update. Those permissions don't automatically include changing the account owner, closing the deal, issuing an order, or sending a customer message.

Use stable identifiers for the customer, opportunity, and quote. Resolve duplicate or ambiguous matches before writing. A matching company name alone doesn't establish the right subsidiary, buying entity, or active deal.

Show proposed changes alongside the current values during the first release. Keep an inferred requirement distinct from a confirmed field, with a source and confirmation owner. Staff need a practical way to correct a suggestion without removing the original evidence.

Check whether another user changed the relevant record after the assistant read it. Reject or reconcile a conflicting update rather than overwriting newer information. Have the integration owner implement that check using the controls the installed system supports.

Give each attempted write a stable operation reference and record its outcome. If a request times out, reconcile the actual CRM state before retrying. An uncertain response mustn't create a second quote or repeat a downstream action.

Keep approvals, document generation, and customer issue as distinct recorded outcomes. If the CRM saves the quote but document generation fails, show staff what succeeded and what needs recovery. The next operator should continue from the known state.

Maintain an audit trail that links the source request, generated draft, reviewer changes, final decision, and issued document. Set access and retention for the records according to your company's policy. The trail needs enough context to explain a commitment without copying every customer document into an unrestricted log.

Hand the accepted commitment to delivery

Specify how you confirm customer acceptance and which document revision it covers. Match the customer's accepted version to the issued offer before creating the next operating record. If they change a requirement during acceptance, route the change to the relevant owner.

Prepare the handover from that accepted revision. Include the agreed scope, exclusions, customer dependencies, delivery conditions, and responsible staff. Link the source documents so delivery can inspect the commitment without searching the salesperson's inbox.

Give the receiving team an acceptance step. In the inspection-services example, the delivery coordinator checks that the work record includes both sites' agreed scope and the confirmed scheduling conditions. An unresolved access question returns to a named owner before scheduling proceeds.

Track a returned handover as part of the quoting process. If staff repeatedly find missing equipment details after a sale, add that field to the relevant intake or review stage. Don't treat the resulting delivery rework as a separate department's problem.

Define the change path after handover. A new site or revised scope needs a linked decision and updated operating record. Keep the original accepted commitment available so teams can distinguish a new request from work they already agreed to perform.

Our distributor order-to-cash guide follows the wider chain through fulfilment and payment. This quote workflow should have a clear receiving owner at its own boundary, even when a later phase handles the rest of that chain.

Test the difficult quotes before release

Build an acceptance set from authorised historical cases and designed exceptions. Remove unnecessary personal information and keep access appropriate to the source records. Have commercial and delivery owners define the expected result before reviewing generated drafts.

Include standard quotes, incomplete enquiries, and requests that the workflow must stop or escalate. Test the saved record and the final document as well as the assistant's visible response. A correct summary doesn't prove that the CRM contains the same quantity or approval state.

Use acceptance cases that exercise decisions and handoffs.
CaseExpected behaviourEvidence to inspect
The customer omits a required site detail.Keep the affected scope provisional and assign the question.Inspect the draft status and missing-information owner.
An old offer contains an expired concession.Use the current authorised pricing source.Check the calculation reference and exception record.
Someone changes a material term after approval.Recheck the affected decisions before issue.Match approval records to the final revision.
A CRM request times out after saving.Reconcile the saved result before a retry.Confirm that only the intended quote exists.
A colleague updates the same requirements field.Stop or reconcile the conflicting write.Check that the newer information survives.
Delivery can't accept an incomplete handover.Return the specific question to its owner.Inspect the receiving queue and unresolved commitment.

Also test access boundaries and an attachment that attempts to redirect the assistant. Confirm that a restricted account's records stay unavailable and that document text can't grant new permissions. Keep approval and issue controls outside the model's discretion.

Run the first cases alongside the existing process, then enable bounded draft creation with staff review. Expand permissions only after the responsible owners accept the evidence. Give operators a fallback route when a source, calculation, or integration is unavailable.

Measure the path to an accepted handover

Define the starting event and endpoint before comparing performance. Track enquiry-to-issued-quote time, then the time and effort needed for an accepted delivery handover. Separate customer waiting time from internal preparation and approval delays.

Compare quotes of similar complexity. Track missing requirements, pricing corrections, repeated approvals, and returned handovers alongside staff effort. Include the time people spend checking and repairing the assistant's output.

Count completed eligible quotes rather than generated drafts as the denominator for operating cost. Include model calls, integration support, software, and review. An increase in drafts can raise activity while the same number of customers still waits for approval.

Keep commercial outcomes separate from directly observed workflow gains. A shorter preparation cycle may help sellers respond sooner, but conversion and revenue also depend on demand, price, and the offer. Don't attribute every movement after launch to the assistant.

For an SME, the next release may cover another standard quote family after the first receiving team accepts the process. A larger company or PE portfolio needs local catalogue, authority, and handover checks before copying the workflow to another business unit.

Maintain a release record when the catalogue, pricing policy, or approval map changes. Re-run the affected acceptance cases and name the owner who authorises the new configuration. This keeps the workflow aligned with the decisions people actually make.

The revenue operations service guide describes how we scope quoting and customer handoffs. A practical first brief names the quote family, source owners, permitted CRM writes, approval path, and delivery acceptance. Use those commitments to judge the release and decide what to extend.

Put the work into practice

AI implementation, agents and production deployment

We help SMEs and scale-ups put AI into a specific business workflow. We define the problem, prepare the data, build the software, and help your team operate it in production.