Wedlake Consulting
Menu
Book a free intro →

Wedlake Consulting · Remote or in Manchester

← Notes

Let AI compare supplier quotes? Give it a test first

Use a set of old quotes to find out whether AI can prepare an accurate comparison and save your buyer time after checking.

Published
Reading
4 min
Topic
Automation

A buyer opens a folder of supplier quotes and builds a comparison for the job. One quote prices individual items. Another prices packs and puts the pack size in a footnote. A third offers a substitute with a different lead time. The buyer has to work out what each supplier is actually offering before the prices mean much.

I’d test whether AI can prepare that comparison before spending money on a tool. Pick a buying job your team already does, give the tool the same documents and see what the buyer has to correct. I would want an answer to two questions. Can the buyer check the comparison against the quotes, and does the whole job take less time?

Pick the quotes that make your buyer pause

Ask the buyer to walk through a completed purchase. Watch where they look for the quantity, delivery charge and lead time. Ask which bits they had to confirm with a supplier and which details changed the decision. Use that walkthrough to specify what the draft comparison must show.

I’d ask for the supplier’s item reference, quoted quantity and unit, price basis, delivery terms and any substitute. Give the buyer a source reference for each important value, such as the filename and page or the email where it appears. If a charge or lead time is missing, the draft should say it needs confirming.

Make a small set of past jobs with documents you are permitted to use in the tool. Remove details the test does not need. Include an ordinary comparison and awkward cases your colleagues recognise, such as a revised quote with the earlier version still attached. Keep some jobs aside for a later check so they stay unfamiliar while you adjust the instructions.

Decide what would make you send the draft back

Before running the test, ask the buyer to write down what they expect for each awkward case. If the supplier quotes by the box and gives the quantity per box, the draft should preserve both. If the quantity per box is missing, the draft should leave the unit cost open and identify the question to send back.

I would make a wrong price basis a reason to stop and investigate. The same goes for taking an old price from a superseded quote or silently treating a substitute as the requested item. A rough layout is easier to fix. Record the difference so the team spends its time on errors that could change a purchase.

Have the buyer use the source references when checking the draft. If a reference lands on the wrong page or points to a value the supplier never gave, record that as part of the result. Where you can work out a unit price from confirmed inputs, have a spreadsheet formula or the purchasing system do the arithmetic and show the inputs beside it.

Count the time it takes to get a usable comparison

Ask the buyer to time an ordinary comparison from opening the documents to having something they would use to make a decision. Time the assisted version to the same point. Include collecting the files, feeding them into the tool, checking references, correcting the draft and finishing any work by hand. Keep the two timings against comparable jobs.

In an AI Opportunity Review, I’d map the job with the buyer and use the timings to decide where it belongs on the roadmap. If it makes the case for a build, keep the sample quotes and expected answers so the finished tool can be held to the same standard.

  1. ManualMap the workThe buyer shows which details take time to check.
  2. ManualBuild and testThe buyer checks a draft comparison against the quotes.
  3. ManualReport hours savedCompare the time to finish the job, including review.
How I'd apply my review, build and reporting method to a supplier quote trial.

If review eats the saving, find the part that still helped. You may have a case for extracting item references and quantities while leaving the price comparison as it is. Test that smaller job and see whether it earns its place. Use the time it actually releases to judge what a build would be worth.

Keep the examples for the next change

Before putting a useful trial into everyday use, run it on the jobs you held back. Have the buyer check them to the same standard. Keep the test documents, expected answers and review notes somewhere the team can find them. Rerun the checks when the instructions change or a supplier starts sending a different form of quote.

If you have a folder like this and want to work out what is worth building, bring an example of an ordinary job and one that took longer to untangle to an AI Opportunity Review. Those jobs give us a way to agree what the tool needs to get right and how we’ll measure the time it saves.

Written by

Owain Wedlake, technology and AI consultant in Manchester. Questions or a different view? Reply by email.

Next step

Start with a conversation.

A free intro session. Bring one process that eats your team's time, and leave with a straight answer on what's worth changing.

Free · No obligation · Remote or in Manchester