Does your business need AI — or a better process?
Before automating, find out where work gets stuck. A practical look at artificial intelligence, sales, management and supplier development.
The meeting ended. The work started over.
The proposal was nearly ready. Someone still had to check the delivery date with the supplier, find the spreadsheet with the current price and work out why the customer was being offered different terms from the ones agreed. Someone suggested using artificial intelligence to speed things up.
One question lingered: speed up what, exactly?
This could happen in a manufacturer, a distributor or a service business. The tools differ; the impasse feels familiar. When work means searching, asking and redoing, technology may help. It may also produce, faster, an answer nobody should have sent.
Before debating which AI tool to buy, it is worth asking a less comfortable question: do we know how the work actually gets done?
The trouble often sits between tasks.
A sales team can write proposals quickly and still lose business at the handover to operations. Purchasing can negotiate well and discover too late that the supplier cannot meet the specification. Each team does its part; the overall delivery remains fragile.
This is where process analysis earns its place. You do not need to start with an impressive flowchart. Take a recent delivery and trace it backwards: who approved it, what information was missing, where it waited and why someone had to redo the work.
A fast task inside a confused process is still part of a confused process.
Consider a commercial proposal: drafting might take twenty minutes, while confirming margin and lead time takes two days. Automating the text improves one step. Addressing the delay between teams may change the business outcome. These are different choices. It helps to know which one we are making.
Three situations. Three decisions.
Calling everything an “AI opportunity” hides important differences. Open the examples and consider what needs to be settled before a pilot.
Sales: the proposal is fast, but comes back wrong
A tool can prepare a first draft using approved information. Price, scope, validity and delivery dates still need a reliable source and an accountable approver. If every salesperson uses different terms, start by organising those rules. Persuasive writing does not fix a wrong margin.
Suppliers: the cheapest quote looks irresistible
Comparing responses and flagging missing fields can be useful support. The decision also needs to consider capacity, quality, lead time, dependency and the cost of failure. Visits, samples and operational evidence remain relevant when the risk warrants them. Supplier development means building the ability to deliver; a price table is only part of that conversation.
Consulting and management: the report looks impressive
Summarising interviews and organising hypotheses can free up time for analysis. A summary does not establish a cause. Check the hypothesis against records, listen to the people doing the work and look for evidence that challenges it. An intervention should address an observed problem, not the confidence with which it was described.
Choose a deliverable, not a promise.
The first test should be small enough to explain and relevant enough to deserve attention. “Improve productivity” is too broad. “Prepare a draft standard proposal without omitting scope or commercial terms” gives you something to compare.
- Observe the starting point. Record time, errors and rework for comparable deliveries.
- Decide what may go in. Use only information authorised for the tool and that particular use. Check how data is handled before sending customer or supplier information.
- Name the person accountable for the output. Someone must check the details that could compromise delivery. Plausible writing can contain fabricated information.
- Compare the whole job. Include preparation, checking and correction. Minutes saved on drafting may reappear in review.
- Let the test inform the decision. Expanding, adjusting or stopping are all legitimate outcomes. A pilot does not have to become a project simply because it began.
A hypothetical calculation helps: cutting a step from thirty to fifteen minutes across forty deliveries would save ten hours. If review adds eight minutes to each delivery, the saving falls to roughly four hours and forty minutes. That may still be useful. It is simply a different story from the one told in the initial excitement.
Ask the question before buying the answer.
The exercise below helps structure a discussion. It is not a technical assessment or a maturity score. Your answers stay on this page.
An opportunity can mean saying “not yet”.
Some businesses need to experiment more. Others need to stop trying five tools for the same problem. Telling them apart requires attention to the work, the people and the consequences of being wrong.
Sometimes the best next step is an AI pilot. At other times, it is agreeing an approval rule, fixing a handover or developing a supplier capable of supporting sales growth. None of these decisions becomes less valuable because it lacks a technology demonstration.
At the next meeting, replace “where can we use AI?” with a more precise question:
Which deliverable deserves to be better — and what stops us doing it well today?
The answer may open a project. It may prevent unnecessary spending. Either way, it has already begun to produce knowledge for the business.