A Mini Tool for Your Business in One Afternoon With AI: How Far to Go
By Anis Hammouche·October 26, 2026
With AI, you can build a small tool useful for your business in one afternoon, for example a quote tracker or a document generator based on a template. It works if the need fits in one sentence, the data is simple and an error is cheap. Beyond that, it breaks: sensitive data, multiple integrations, maintenance and security require skills that one afternoon does not give you.
What an afternoon really allows
An afternoon is enough for a tool that does one thing on data you control. Two possible examples: a spreadsheet that reads a list of quotes and flags those with no reply for fifteen days, or a form that fills a quote template with the information you enter. These are examples of what is achievable, not results measured on a real case.
The work happens in three stages. You describe the need precisely to the AI, which proposes a first version. You test it on real cases, noting what does not work. You correct, then decide whether to keep the tool or throw it away. The third stage often takes longer than the first: that is where the limits show up.
The four-question test
Before you start, answer these four questions. They help you sort projects before spending an afternoon on one.
- Does the need fit in one sentence? If you need three paragraphs to describe the tool, it is too big for an afternoon.
- Is the data simple? A file with a few columns, without sensitive personal data, is easy to handle. Client records, medical files or salaries call for real thought about data protection.
- Is an error cheap? A tool that calculates reminders is risky if it sends emails automatically to the wrong recipient. A tool that gives you a list to check is much less so.
- Can you test with real cases? If you cannot check the result on a dozen real cases, you do not know whether the tool works.
The table below applies this test to five common types of tool.
| Type of tool | Doable in an afternoon | Where it breaks |
|---|---|---|
| Quote or prospect tracking sheet | Yes | Several people edit at the same time, complex calculation rules |
| Quote or document generator from a template | Yes | Mandatory mentions forgotten, wrong VAT calculation if the rule changes |
| Automatic email reminders | A prototype, yes | Sent to the wrong contact, personal data, no human review |
| Tool connected to your accounts, CRM and website | No | Authentication, connector updates, long-term maintenance |
| Application with user accounts and payments | No | Security, data protection, user support |
Why the feeling of saving time can mislead you
AI tools give a feeling of speed, which does not guarantee a real gain. The METR study, published in July 2025, illustrates this gap. Sixteen experienced developers worked on 246 real issues in open source projects they already knew. When AI was allowed, they took 19% longer. Afterwards, they still estimated that AI had saved them about 20% of their time.
The authors state the study's limits. Sixteen people is a small group. The tools tested were those of early 2025, and METR has since published an update with new data. The result does not show that AI will never save time. The practical lesson still holds: measure your own time on two or three tasks, with and without the tool, and do not rely on your gut feeling alone.
What to check before using the tool
A homemade tool is still a work tool: check it as you would bought software. At least four points.
- Where the data goes. The CNIL recommends avoiding, as far as possible, entering personal data into an AI tool, and asking the provider how it complies with the GDPR. Read the provider's policy on its official page before you connect your client file. See the CNIL's questions and answers.
- Backup. If the tool disappears or the file is corrupted, what do you lose? If the answer is nothing, you can carry on. Otherwise, back up before you test.
- Access. Who can change the tool, and who can see the data? A shared link without restrictions can expose your data to anyone.
- Maintenance. Who fixes it when it breaks? If the answer is "nobody", the tool becomes a debt.
When to stop and bring in someone else
A project that takes too long to reach a satisfying result is costly, even when it is technically feasible. I myself abandoned two AI projects, one about poker and one about the stock market. It took too long to get a result I was really satisfied with, and my priorities changed.
A simple rule: if a test fails twice without a clear cause, stop tinkering and decide. Either you simplify the need, or you take an existing solution, or you bring in a developer for the technical part. The cheapest choice is not always to do everything yourself.
Frequently asked questions
Do you need to know how to code to build a mini tool? For a simple prototype, no. For a reliable tool that handles real data, you need to understand what the produced code does, otherwise you cannot fix it or judge it.
How much does the tooling cost? It depends on the tools you choose. Check prices on the providers' official pages, noting the date you checked, because they change often.
Can you sell the tool to other companies? Not without reviewing the licences of the components you use, the data you process and your liability in case of an error. That is a separate step, not to be confused with internal testing.
How do you know a project is too big? If the need does not fit in one sentence, if the data is sensitive, or if the tool has to run without you, the project goes beyond an afternoon.
Stuck on a specific case? Write to me through the contact page.
