Where to start with AI in an SMB

In many companies, AI is already there, but in a scattered way: everyone tries a consumer tool on their own, with no rules on data and no measurement of the benefit. Moving to serious use does not require a grand plan. It takes a method to pick the right first use case, scope it, and prove its value. Here is the one I suggest.

Step 1: spot candidate tasks

Get the managers of two or three departments together for an hour and list the tasks that come up every day or every week. Good candidates often share these traits:

• lots of text or documents: emails, purchase orders, supplier invoices, meeting notes, customer requests;
• a share of reading and sorting: understanding, classifying, extracting, summarizing, answering from a known knowledge base;
• steady volume: dozens of occurrences a week;
• recoverable errors: someone can check before the result commits the company.

Common examples: processing orders received by email, summarizing and sorting customer requests, extracting data from invoices, searching internal documentation.

Step 2: score the tasks and pick one

Score each candidate task from 1 to 3 on five criteria. The one with the highest total becomes your pilot.

One use case, not three. Running several pilots in parallel spreads your teams’ attention thin and makes the results hard to read.

Step 3: set the data and usage rules

Before any development, three points need settling:

1. The data: what information will be sent to the model, and what stays in your systems. The sorting grid is in what data to share with AI.
2. Internal rules: a short policy stating which tools are approved and with which data. The French data protection authority recommends this in its Q&A on generative AI of July 18, 2024 (in French, checked October 27, 2026).
3. The regulatory framework: since August 2, 2026, the EU AI Act requires that people interacting with an AI system be informed of it (Article 50), and asks companies to take measures to support their staff’s AI literacy (Article 4, as amended by Regulation (EU) 2026/1744). Stricter obligations for “high-risk” systems listed in Annex III, such as recruitment, will apply from December 2, 2027 under that same regulation. For your specific situation, this page is not a substitute for legal advice.

Step 4: run a short, measured pilot

The pilot lasts 2 to 8 weeks. It includes:

• a test set of 30 to 100 real examples, with the correct answer defined by the people who do the task;
• numeric success criteria set in advance: for example, a minimum accuracy rate and a target processing time;
• a review screen where a person checks, corrects or rejects each AI suggestion;
• time measurement before and after, over the same period and the same volume.

I run these pilots as part of AI integration, starting at $2,300 (€2,000)excl. VAT, plus API usage fees billed by the provider. The cost breakdown is in how much AI integration costs.

Step 5: roll out, or stop

If the criteria are met, we move to production: final access rights, logging, user training, cost monitoring. The use case can then be built into a broader business tool; the estimator gives you a range for an app with built-in AI.

If the criteria are not met, we stop or change direction. A pilot showing that a task is not suited to AI cost little and saves you from a bigger investment.

When it is not the right time

I will be straightforward: some situations are not ready for an AI project yet.

• The data is not accessible. If the information is scattered across spreadsheets and personal inboxes, you first need a tool that centralizes it.
• The process is not stable. Automating a way of working that changes every month means automating confusion.
• Nobody can own it. Without a business owner available a few hours a week during the pilot, the project will not be evaluated properly.
• Off-the-shelf software already does it. Check the AI features of your current tools before funding custom work.

Where to start, in practice

1. A one-hour meeting to list candidate tasks.
2. The scoring grid to pick one use case.
3. A 30-minute call with me to check feasibility and the budget range.
4. Scoping over 1 to 2 weeks: test set, model choice, success criteria, fixed quote.

If you would rather structure how your teams use AI before any development, AI coaching ($170 (€150)excl. VAT per hour) or team training (from $1,700 (€1,500)excl. VAT) are good starting points.