AI in a trading company

Why AI in trade starts with the orders, not the forecasts

AI is sold to trading companies from the analytics side: demand forecasts, price optimisation, recommendations. Those projects can be valuable, but they share a precondition we check before starting — a complete, tidy history of sales and stock. In practice that history tends to be patchy: stock-outs masquerading as no demand, unmarked promotions, the same product living under two codes. A forecast computed on such data is a number generator, not a decision tool.

Meanwhile, in the same company, a stream of work flows every day that AI can take over now: orders arriving by email and phone in a dozen formats, retyped by hand into the ERP (the business management system that handles orders, stock and accounting); quote requests waiting for pricing assembled from the price list, the customer’s terms and availability; supplier price-list updates entered line by line; purchase invoices checked against orders and goods receipts by hand. These are document- and text-based processes — the data exists, the rules can be written down, and a mistake is detectable and reversible.

That is why a good first AI project in a trading company looks unspectacular: it does not predict the future, it puts the present in order. In return, the first results appear within weeks — and the project builds the very data history the forecasts will one day need.

What we actually automate in trading companies

• Order handling. Reading orders arriving by email, as PDFs and spreadsheets — in layouts nobody has described before — matching the lines to product codes and preparing the ERP entry for a salesperson to approve. This is the most common first process: it recurs daily, and order confirmation adds a second checkpoint. In the pilot we measure both the accuracy of the entries and the time needed to check them.
• Quotes and price lists. Drafting a quote from the price list, the customer’s terms and availability — for a salesperson to approve — and loading supplier price-list updates with a list of changes to review, instead of entering the lines one by one.
• Enquiry handling. Recognising and routing questions about availability, order status or documents to the right person, and drafting the reply for approval. The whole process — from enquiry to answer — is described separately on our customer service automation page; in a trading company it works identically.
• Stock and warehouse documents. Reconciling warehouse documents with the system and a list of discrepancies to explain — instead of discovering them at stocktake. Plus completeness checks: what was issued but not invoiced, and the other way round.
• Supplier settlements. Checking purchase invoices against orders and goods receipts, with a list of price and quantity discrepancies to explain — and keeping track of retrospective rebates and bonuses that today are settled in a spreadsheet once a quarter. Ambiguous cases go to a person, not straight to posting.

The common denominator: a person approves the output, the system prepares it. Which processes qualify for this kind of automation at all — the five conditions a good candidate must meet — is set out on our AI automation page; in a trading company those conditions are checked in exactly the same way.

What we do not promise

• Demand forecasts without a data history. A forecasting model has to learn from a complete, well-ordered history of sales and stock. If stock-outs masquerade as missing demand and promotions are not flagged, an honest offer starts by putting the data in order — and says plainly that the forecasting result comes later.
• Pricing decisions with nobody accountable. The system drafts an offer from the price list, the customer’s terms and availability, but the decision on price and discount stays with the salesperson who knows the customer and the margins.
• Unattended operation from day one. ERP entries, offers and replies to enquiries start in a mode where a person approves them, and oversight is loosened on pilot data. The full set of conditions under which we advise against a project is collected on what we don’t do.

Price lists, margins and customer data

Trade documents are this industry’s most sensitive data: purchase prices, margins, individual customer terms — information whose leak hurts more than many an outage — and the contact details of the people ordering are personal data. Before any document reaches a model, you have to settle which data leaves the company at all, where it goes and on what basis — from the choice of provider and processing region to limiting the data to the minimum the step needs. How we approach this, and what to ask any supplier, is set out on our security and GDPR page.

How we work

As in every project: we pick one process and measure one indicator before we start, for example minutes from an order arriving to its ERP entry. Then we run the smallest possible pilot, with a person approving every result. Before switching over, the system runs alongside the current way of working for a while. The salespeople’s corrections are recorded and used when we periodically adjust the system — that is what improves the proposals over time. The full course is set out in our implementation methodology, and the full scope of the service, from process audit through to upkeep, on our AI implementation page.

What it costs

Three things drive the cost: the number of systems to connect, the state of the data, and how reliable the results must be, that is, how rarely they may be wrong. The size of the product range matters less than whether the orders and documents live in one system or in the mailboxes and spreadsheets beside it. All the components, upkeep included, are broken down in our note on what an AI implementation costs.

Frequently asked questions

Where should a trading company start with AI?

With the processes that recur daily and have their data in documents: orders retyped into the system by hand, quote requests waiting for pricing, supplier price lists entered line by line, purchase invoices checked against orders. There the data already exists, a mistake is reversible, and results show in weeks. Demand forecasting and price optimisation first require a complete sales history — which is why they are rarely a good first project.

Will AI forecast our demand and set our prices?

A forecast is only as good as the data it is built on — and in trading companies the sales history tends to be patchy: stock-outs masquerading as no demand, unmarked promotions, the same product living under two codes. The honest order of work is to put the sales and stock data right first, then forecast. Pricing decisions are a separate matter: the system can prepare the analysis and a proposal, but the decision on price and margin stays with a person — it is too expensive a place for a model’s quiet mistake.

Will this work with our ERP and B2B shop?

The audit at the start of a project settles that. We check whether the API (the interface other software uses to connect to the system) or file import gives access to the data and actions the chosen process needs. On that basis we confirm whether the integration is feasible, what it covers and what it costs — and whether an existing system feature would be enough. The trade and warehouse systems in use vary widely here, and that difference affects the cost — and so whether the project pays off. More important than the system itself is how orders and documents reach it: email, phone, the B2B shop, electronic data interchange (EDI) — because that stretch of manual retyping is what we automate first.

What does an AI implementation cost in a trading company?

Cost is driven by the number of systems to connect — ERP, shop, warehouse, email — the state of the data, and how reliable the results must be, that is, how rarely they may be wrong. The size of the product range matters less than whether orders arrive through one channel into one system or through five into two. Upkeep is a separate line: per-operation fees and periodic retesting after model version changes. We break the components down in our note on what an AI implementation costs.

What about the confidentiality of price lists and customer data?

Purchase prices, margins and per-customer terms are a trading company’s most sensitive data, and the contact details of the people ordering are personal data. Before any document reaches a model, you have to settle which data leaves the company at all, where it goes and on what basis — from the choice of provider and processing region to limiting the data to the minimum the step needs. The principles, and the questions worth asking any supplier, are set out on our security and GDPR page.