An SME rarely needs to “do AI”. It needs to process requests faster, reduce administrative work, make a decision more reliable or offer a new service. Artificial intelligence matters when it improves that outcome in a measurable way.
The role of an AI consultant is therefore not to begin with a catalogue of models. It is to find a problem that is costly, frequent and understood well enough to justify intervention.
Start with a process, not a tool
A request such as “we want an internal ChatGPT assistant” describes an assumed solution. It does not explain who will use it, for which decision, with which data or what happens when an answer is incomplete.
A better first exploration follows a real process: preparing a quote, qualifying an enquiry, analysing a case file, checking an order or finding a contractual obligation. Take one recent example and follow it from trigger to final outcome.
For each stage, four questions reveal most of the opportunity:
- how often does it happen;
- how much active and waiting time does it consume;
- which errors, rework or lost opportunities does it create;
- which data and approvals are required to decide.
This prevents a team from automating the visible task while the actual constraint remains upstream.
Separate simplification, automation and AI
Not every point of friction needs a language model. A clearer rule, a form or a connection between two systems may deliver a better result with less risk and maintenance.
Conventional automation works when inputs and rules are stable. AI becomes relevant when the work involves language, varied documents, classification or a decision that retains an element of judgement.
Many strong solutions combine all three levels: simplify the flow, automate deterministic steps, then use AI for the genuinely ambiguous parts.
Estimate the economic potential
An initial estimate should not promise an artificially precise return. It should still make the decision comparable.
A useful range can be built from monthly volume, active time per case, rework cost, current delay and the value of the outcome. For commercial activity, include missed requests, response time and the capacity to serve more customers.
Potential value is not limited to headcount savings. It may mean:
- handling more volume without increasing effort at the same rate;
- reducing mistakes that cause credits, disputes or repeat work;
- shortening the delay between request and response;
- enabling a service that would be too expensive to deliver manually;
- retaining knowledge currently held by only a few people.
Choose a controlled first scope
The best first project is not necessarily the most spectacular. It should create useful learning with limited risk.
A sound scope has a business owner, representative examples, an observable outcome and a fallback procedure. It does not try to replace every decision in its first version. It prepares a case, recommends an action or automates a bounded part of the workflow, with human review where the stakes require it.
What a diagnostic engagement should deliver
Before development begins, management should be able to answer five questions:
- Which economic problem are we addressing?
- Why is AI useful here?
- Which data and integrations are required?
- How will quality and adoption be measured?
- What is the smallest product that can test the hypothesis?
AI7’s product and automation diagnostic organises this work around one priority process. It produces a process map, ranked options, an MVP scope, indicative architecture and estimate.
The purpose is not to sell development automatically. It is to support a sound decision: build, simplify another way or avoid investing for now.
If you have identified a slow, costly or difficult-to-scale process, describe how it works today. The first discussion will determine whether the subject genuinely warrants an AI project.