AI for Sales Quotations: Catch Wrong Prices Before You Send
A customer asks for a repeat order. Your salesperson copies the previous quotation, changes the quantity and sends it out.
But the supplier cost has changed. The previous discount was a one-off approval. And the customer now wants delivery to Johor rather than collection from your Klang Valley warehouse.
Consider an illustrative order for 200 units: copying an outdated price of RM48 instead of the approved RM53 reduces the quoted value by RM1,000, before any delivery charge.
A useful AI quotation workflow should help prevent that mistake, not simply produce a nicer PDF faster.
Start with an approved pricing source
Before connecting an AI agent, decide which record has authority over the selling price.
For your business, that might be a customer price list in the accounting system, a contract schedule or a controlled Excel workbook. It should not be whichever quotation the agent finds most recently in someone's email.
Define the order of precedence explicitly. For example:
- Use an active customer contract price when the item, quantity and validity conditions match.
- Otherwise, use the approved customer price list.
- Otherwise, use the standard selling price.
- Send missing, conflicting or expired prices to a named reviewer.
Check what your existing software already provides. SQL Account's documentation describes customer-specific price enquiries, price references for different units of measurement and sales price history. AutoCount's quotation documentation describes automatic selling-price selection according to its pricing settings.
These are useful foundations, not proof that your records are clean or that an AI connection is already available. Confirm your edition, configuration and integration access before agreeing on scope.
AI can interpret a customer's request. It should not invent the commercial terms needed to fulfil it.
Design the workflow around exceptions
Consider an illustrative Penang distributor receiving an emailed request for “30 boxes of the usual filter, same price as last time”.
The proposed agent should first identify the customer account and retrieve the relevant order history. It can suggest a product match, but it should stop if “the usual filter” could mean two different stock codes.
Next, it should check:
- The exact item, pack size and unit of measurement.
- The applicable price and its effective period.
- Quantity discounts and whether they can be combined.
- Delivery location, charges and requested date.
- Payment terms and quotation validity.
Keep calculations in deterministic software rules. Use AI to interpret wording and explain exceptions, not to perform unchecked price arithmetic or decide tax treatment.
A stock balance should not automatically become a delivery promise. Require an agreed availability check that accounts for allocations, warehouse location and fulfilment constraints. If that information is unavailable, mark delivery as subject to confirmation.
Once the checks pass, the workflow can prepare a draft quotation in the existing system or an approved template. Writing into AutoCount, SQL Account or another platform depends on the access supported by that installation; it is not guaranteed by having the software name on an integration list.
Keep discount authority with people
Write down who may approve a lower price, extended payment terms or free delivery. Avoid asking the agent to infer authority from job titles or previous messages.
An illustrative approval policy could allow standard-price drafts to proceed to salesperson review, while sending below-floor prices to the sales manager. A missing cost should trigger a review rather than a guessed margin.
The approval request should show the requested price, approved reference price, relevant cost basis and reason for the exception. Specify whether your margin check includes freight and other direct costs.
AutoCount's published quotation guide describes approval rights controlling whether quotations can transfer to other documents. That is not the same as controlling whether a PDF can be emailed to a customer. Check both permissions in your proposed workflow.
Require approval against a specific quotation version. Any subsequent change to price, quantity or terms should invalidate that approval. Keep the original request, pricing source, changes and approver together for later checking.
Pilot one product range before expanding
Start with repeat products, stable units of measurement and a small set of customers with clear pricing. Leave bespoke engineering estimates, negotiated bundles and uncertain supplier costs outside the first pilot.
The preparation work is practical: remove duplicate customer records, standardise pack sizes, retire old price lists and nominate someone to maintain pricing. Train sales staff to resolve exceptions in the workflow rather than approve privately and leave no record.
Test historical requests in a controlled environment, then run live drafts alongside your existing process without automatic sending. Include deliberately difficult cases: expired contracts, ambiguous items, changed delivery addresses and revised quantities.
Measure preparation time, first-review accuracy, pricing corrections and approval turnaround. Compare the same kinds of quotations before and after the pilot. Faster drafting is not a saving if managers spend longer correcting it.
If requests are already structured and the only task is applying a price list, ordinary software rules or a quotation template may be enough. AI earns its place when interpreting messy requests and coordinating checks removes real work.
Solver AI's approach is to validate the workflow first, test it in a controlled environment and then roll out. For quotations, the aim should be fewer commercial mistakes with less preparation work, not autonomous discounting.