Every trading business has a back office that nobody quite has time for. Prices need reviewing every week. Product pages drift out of date. Somebody should know which products will run out before the next delivery lands. And what your suppliers have agreed is usually sitting in someone’s inbox.
We automated five of those jobs inside a real UK equipment supplier with a team of 20 to 30 people. The AI does the reading, checking and number crunching. A person approves every change. Here’s what we did, step by step, so you can try it yourself.
Start with one list: the products that matter
Before anything else, we built a monthly ranking of every stocked product by gross profit over the last 12 full months, in bands: Top 50, Top 100, Top 200 and Top 500. Spare parts, drop ship lines and anything not active are left out. A product only moves band after two months in a row on the new side of the line, so the list doesn’t flicker about.
The other jobs use it: the stock forecast reports on the Top 50, 200 and 500, and the supplier log uses it to decide which products are key. Before anyone relied on it, it had to match the list the business had signed off by hand. It matched line for line.
“Here is 12 months of completed sales by product with gross profit, plus our stock list. Leave out spare parts, drop ship lines, inactive products and anything with zero or negative gross profit. Rank the rest by gross profit in pounds and band them Top 50, Top 100, Top 200 and Top 500. Flag any products within £1 of each other at a band line, and tell me what you left out and why.”
A weekly price review you approve, not build
Every Wednesday the AI reviews every product’s sales speed, margin, stock and cost against written rules. Slow sellers come down 5%. Fast sellers and thin-margin lines go up 5%. Nothing moves more than 5% a week, nothing moves twice within six days, and clearance stock steps down each week it doesn’t sell, but never below a hard floor.
It then writes an import file for the ERP and emails the director a summary, including what the month’s blended margin will look like before and after the changes. He checks it and uploads it himself. The AI never touches live prices. The first two batches were approved and uploaded within a day of each other.
A Friday check of every product page
Product pages collect mistakes: warranty text that doesn’t match the real warranty, “free delivery” on products that aren’t, blank dimensions, gross weights where net weights belong. Every Friday the AI checks every stocked listing against the business’s own reference data. Each issue gets a priority and a confidence. Confident fixes are proposed for approval, and everything else becomes a question for a person.
The first full audit raised 12 questions, which were answered the same day. The approved fixes went live the next morning and every change was logged with the old value, so any of it can be rolled back. A recheck the day after found the fixes had held.
Fix purchase order dates before you forecast stock
A stock forecast is only as good as the dates on your purchase orders. So every other Friday the AI first compares the ERP’s order dates with the buyer’s own tracker, and lists what needs changing. A person changes the dates and replies “done”. Only then does it forecast how many key products will be in stock now, and in one, three and six months, and which will run out before the next order arrives.
The first run found 24 orders that needed re-dating. It also found that the open orders report everyone relied on was missing 37 live orders, because its filter only showed orders marked “Open”. That one check fixed a report the whole forecast depended on.
“Here are our open purchase order lines from the ERP and the buyer’s tracker. Read dates as UK format. First list any typos in the tracker, without guessing them. Then list every order where the ERP date differs from the tracker’s arrival date: order, supplier, current date, new date and units. Don’t change anything.”
One place for what suppliers have agreed
Each week the AI reads the sourcing agent’s updates and logs them into one supplier workbook: payment terms, quality issues, quotes and price cuts. It takes a dated backup first, never overwrites history, and logs unconfirmed terms without counting them. It then works out how much cash the agreed terms free up, clearly labelled as an annual estimate.
Your starter checklist
1. Find the one report or workbook that already pulls sales, stock, costs and orders out of your ERP.
2. Build your product ranking first and check it against the list you use today.
3. Write your pricing rules down, with a reason for each, before you automate anything.
4. Agree which source wins when product data disagrees: manual, supplier, ERP or website.
5. Check your open purchase order report includes orders awaiting approval and orders not yet sent.
6. Back up every workbook before the AI changes it, and log every change.
7. Make sure a person approves every price, listing and order date change.
8. Write a short handover for each job so anyone can run it.
One safety point: use a business account for your AI assistant and check its data settings before sharing customer, supplier or financial details. Claude, Microsoft Copilot and ChatGPT all offer business plans.
Build it yourself, or have us build it
If you’d like to build this yourself, our guide The Trading Back Office covers all five automations in detail: the rules and why they exist, every check, the problems we hit, 21 copy-ready prompts and the procedures to adapt.
If you’d rather we did it, our Trading Back Office Setup builds it inside your business, connected to your own ERP, shop and supplier data so nothing is keyed in. It’s £2,495 + VAT, fixed price, done in three weeks.
We’re not IT people. We’ve sat in the MD’s chair and built the systems behind it, and we’ll talk to you in plain business terms. Not sure where to start? Take our free Growing Pains Check to see where your business is under the most strain.
Leave a Reply