Reviewed 21 October 2026
Knowing ten tools is not the same as having a customer. A commercially useful skill improves a task that someone already needs done, with a result they can assess. Start with one repeatable workflow, a small test and a written scope. A subscription, certificate or impressive demonstration does not guarantee demand or income.
Find a problem with a visible cost
Choose an industry you understand: property enquiries, travel requests or product information, for example. Interview five business owners about a recurring task, its volume, time cost and expensive mistakes. Identify who can approve the result. Do not propose an AI system before understanding how the work currently moves between people.
Three services suitable for a limited pilot
Consider organising approved customer questions into a documented knowledge base, extracting fields from similar documents with human checking, or preparing operational reports from authorised data. Define the input, output and exceptions. A beginner should not sell autonomous medical, legal, credit or employment decisions as a quick productivity experiment.
Learn in a useful sequence
Begin with spreadsheets and data cleaning, clear instructions and output verification. Add automation and programming when the workflow genuinely requires them. Keep an error log: invented information, missing fields, wrong calculations and unsupported answers. Your ability to diagnose and correct failures is more useful to a customer than a list of tools.
Build an acceptance test
Create 30 non-confidential test cases, including incomplete and misleading inputs. Compare outputs against human-reviewed answers and record correction time. Accepting 27 out of 30 gives 90% on that sample; it is not a general accuracy guarantee or automatic approval for a sensitive process. Agree acceptance thresholds and an exception route with the client.
Price the work, not just the subscription
An illustrative calculation: eight hours at an internal rate of €20, €30 in tools and two review hours at the same rate produce €230 in direct costs. Add overheads, applicable taxes and margin. These are teaching assumptions, not market prices. Separate setup, ongoing support and variable usage, and cap documents and revisions.
Treat data handling as part of the service
The CNIL’s guidance highlights the risks of supplying confidential or personal information to public tools without suitable assessment. Use authorised test data, examine provider terms and access controls, and agree retention and deletion. Explain who reviews outputs and who remains responsible for approving their use.
A four-week route to a testable offer
Week one: customer conversations and one chosen workflow. Week two: prototype and test set. Week three: a one-page offer stating delivery, exclusions and price. Week four: a limited paid pilot and a before-and-after review. If checking errors removes the promised saving, redesign the service. France Num offers resources on business adoption.
A first message that earns attention
Try: “Your team appears to repeat this task. I can test a limited way to organise it, with human review and a report on time and errors. If we agree the scope and permitted data, I will send a fixed pilot proposal.” Offer a relevant outcome, not a catalogue of software. Never demonstrate your skill using another client’s confidential files.
Frequently asked questions
Do I need to be a programmer?
Not for every first pilot. Technical help may be necessary for integrations, sensitive data or a process beyond the tool’s limits.
Will a month of learning guarantee income?
No. Learning time does not establish demand, quality or the ability to win and serve a client.
What makes convincing evidence of skill?
A documented example showing the problem, input, output, errors and limitations without revealing client secrets.
A question to discuss
Which recurring task in your work would justify a small, measurable pilot?
