There is a lot of noise around AI, and plenty of promises that a small company has neither the time nor the budget to check. Our experience is more measured: AI is very useful for one specific kind of work, and quite disappointing for the rest.
What it does well is read text, understand its meaning, extract information and write a first draft. In other words, the part of administrative work that consists of reading, sorting, retyping and rephrasing. What it doesn't do is take responsibility for a decision, know the history of a customer relationship, or guarantee that an answer is correct. It can be wrong with complete confidence.
Hence the principle behind all our projects: AI prepares, a human validates before anything goes out. For freelancers, small businesses and SMEs, that is what makes AI automation both useful and safe. Here are seven practical use cases, with what the AI does and what stays human in each.
Seven AI use cases for SMEs
1. Sort incoming emails
What the AI does. It reads each email and its attachments, files it in the right category (quote, after-sales, billing, complaint), spots urgency and extracts useful details such as an order number or customer reference. It also flags what is missing so you can ask for it straight away.
What stays human. The decision on how to reply, and sending. Sensitive or ambiguous requests are passed to a person rather than handled automatically.
This is the most common entry point, and we wrote a full article on it: sorting customer emails with AI, without letting it reply on its own. It is at the heart of our inbound requests work.
2. Extract data from documents
What the AI does. Quotes, supplier invoices, purchase orders, contracts: it reads the document, even when the layout changes from one sender to the next, and pulls out the useful fields (sender, dates, lines, totals) in a usable format. No more retyping line by line.
What stays human. The check. Values are shown next to the original document, and discrepancies or uncertain fields are flagged so a person can confirm them before they go into your management tool.
This use case fits naturally into a quotes and invoicing workflow.
3. Draft replies
What the AI does. From your template answers, the case history and your tone, it writes a draft ready to review. For an acknowledgement, a request for a missing document or an answer to a frequent question, the time saving is real.
What stays human. Reading, adjusting, and clicking "send". A plausible answer isn't necessarily a correct one: an invented date or an unapproved promise costs more than the time saved.
This works hand in hand with triage: it is the same inbound requests chain, from receipt to a prepared reply.
4. Enrich and qualify leads
What the AI does. It gathers public information about a company, summarises what it does, spots signals useful for a first approach and proposes a ranking based on the criteria you have defined. The salesperson goes into the call with the context already in hand.
What stays human. Choosing the criteria, deciding who to contact, and the message itself. The score helps with prioritising, it is not a verdict: it should be open to discussion and correction.
This is one of the building blocks of our lead generation work.
5. Comment on reporting
What the AI does. It reads your figures once they are consolidated and writes a first commentary: what has moved, what looks unusual, what deserves a closer look. The dashboard arrives with an explanatory text instead of a bare grid of numbers.
What stays human. The interpretation. AI describes; it doesn't know the company's context. Management, or whoever follows the finances, confirms, qualifies and decides. The figures themselves come from the consolidation of your data, with discrepancy checks.
See our approach to financial reporting.
6. Turn meeting and call notes into CRM records
What the AI does. From a transcript or rough notes, it produces a structured summary: decisions made, actions to take, people involved, next deadline. It proposes attaching them to the right contact or deal in your CRM.
What stays human. Validating what gets recorded, especially when a note commits someone. Conversations containing sensitive data call for knowing exactly which service processes the text, and where.
This use case depends a lot on the tool you already use. When no standard solution fits, it is typically a job for a custom platform.
7. Generate documents from a case file
What the AI does. It assembles the information in a file to produce a quote, a report, a proposal or a presentation, in your company's format and tone. The document no longer has to be built from scratch, it only has to be reviewed.
What stays human. The review, plus amounts and commitments, which are checked before anything is sent. It is also up to a person to decide what goes in the document and what doesn't.
This is exactly what we built for Manon Sassy, an independent casting director in Paris: with a custom platform, she now prepares client presentations in 15 minutes instead of 4 to 5 hours. The details are in the case study. For quotes and invoicing, the logic is the same.
What AI shouldn't do on its own
The seven use cases above have one thing in common: at the end, a person looks before anything leaves the company. This rule isn't a precaution for its own sake, it protects three things.
- Sending to customers without validation. An email, quote or invoice sent without review commits your company. At first, nothing goes out without approval; later, you can possibly allow an acknowledgement or a purely factual answer, case by case.
- Commercial or legal decisions. Granting a discount, accepting a deadline, interpreting a clause: these decisions engage the company's responsibility and belong to a person, never to a language model.
- Unreviewed sensitive data. Customer emails, contracts and HR files contain personal information. You need to know which service processes the text, where, and with what guarantees, and keep third-party services to the minimum.
Keeping a human in the loop isn't a lack of ambition: it's what lets you start quickly, without waiting for a perfect tool, and widen the AI's autonomy later with your eyes open.
Where to start
There's no need to launch all seven projects at once. Pick the most repetitive task, the one where a mistake is easy to catch before it goes out, and run the tool alongside your current process before switching over. If you're torn between several candidates, our method for choosing which tasks to automate first will help you decide.
And if your volume is very low, or every case needs a fully bespoke treatment, AI isn't always the right answer: a simple rule or a better document template is sometimes enough.
Frequently asked questions
Repetitive tasks built around text: sorting incoming emails, extracting data from documents, drafting replies, enriching and qualifying prospects, commenting on reports, turning meeting notes into CRM records, and generating documents from a case file. In every case, AI prepares and a human validates.
No, and that isn't the goal. AI takes over data entry, sorting and first drafts, which frees up time for judgement, customer relationships and decisions. Anything that commits the company, such as a customer reply, a quote or a commercial decision, is still approved by a person.
The most repetitive and least risky task, where a mistake is easy to catch before anything goes out. Sorting incoming emails and extracting data from documents are often good starting points. Our method for choosing what to automate first helps you decide.
Related service · Inbound requests
Let's find where AI would save you time.
A free 30-minute call, no commitment: we go through your repetitive tasks, tell you which ones AI can prepare for you, and honestly which ones aren't worth the effort.