AI data analysis that finally makes your company’s figures talk

There are four levels, from the simplest to the most advanced^[5]:

Most companies already master the first level. AI makes the other three accessible, without needing a team of data scientists.

It all depends on your data and what you want to do with it. Here are the main families of tools^[6]:

Several tools let you get started without spending anything:

• Power BI Desktop, Microsoft’s software, is free for creating interactive reports on your computer^[7]. Sharing them with your colleagues, however, requires a licence.
• Powerdrill offers a free plan with a number of daily credits^[6].
• The free versions of ChatGPT, Claude or Gemini can analyse a file and produce a chart from it, with usage limits.

Keep one rule in mind: in their free versions, these assistants may reuse what you give them to train their models^[8]. So keep them for public or anonymised data.

Yes, provided you know how they were obtained. An AI assistant does not work like a spreadsheet: it writes its answer word by word. If it gives a figure without actually calculating it, that figure may be plausible… and wrong, especially on a large file it has not read in full.

Fortunately, recent assistants can run calculations when you give them a file, but this is not always visible. The right reflex: ask it to show the formula, code or query it used, and check one or two totals yourself. In the solutions we develop, every answer shows its calculation and the data used, so you can check it.

Predictive AI uses your history to estimate what is likely to happen. A few telling examples in business:

• forecasting sales over the coming months, to adjust your purchasing and your teams,
• anticipating stock needs, and avoiding both stock-outs and overstocking,
• spotting customers who are about to leave, from the drop in their orders or their exchanges,
• estimating your cash flow, taking into account your customers’ usual payment delays.

One limit to be aware of: a forecast assumes that the future looks a little like the past. It will not see a crisis or the sudden loss of a major customer coming. It will, however, alert you very early when reality starts to diverge from it.

Yes. It can run regressions, statistical tests, customer segmentations or time series, writing the necessary code itself. What used to require Python or R skills becomes accessible in a few sentences.

Interpretation, however, remains your job. Correlation is not causation, and AI knows neither the context of your figures nor the promotion launched last month. Always ask it to detail its method and assumptions.

No. That is precisely the point of AI: a manager, a sales rep or a financial controller can query the data themselves, in plain language. You do, however, need someone who knows your figures well and becomes their point of reference: they know where they come from, validate the definitions, such as what counts as an “active customer”, and spot when something is off.

No, but it changes their job. It takes over data preparation and repetitive queries, which take up a large part of their days. Analysts then spend more time on what has the most value: understanding the business, choosing the right indicators, interpreting the results and convincing decision-makers.

Two texts mainly govern these uses:

• The GDPR, as soon as your data concerns people, whether customers or employees. The CNIL recommends starting from a specific need, limiting the personal data given to AI, checking whether the provider can reuse it, and not letting it make important decisions on its own^[2].
• The European AI regulation, which in particular requires companies to train their teams in the controlled use of AI^[3][4]. Most sales or stock analyses do not fall under “high-risk” uses.

The answer depends less on the AI itself than on the state of your data. If your figures are already neatly organised in a single piece of software, preparation will be quick. If they are scattered across five Excel files that don’t all say the same thing, you will first need to put them in some order. Three factors affect the budget:

• the number of sources to connect, from your CRM to your accounting, not forgetting your shared files,
• the quality of the data to be reworked, with its duplicates, different formats or incomplete histories,
• what you expect from the solution, from a commented dashboard to an assistant that answers your questions, or even forecasts.

To compare transparently, also think about licences. Dashboard software is paid for every month, for every user.

At Kreante, every project is priced at a fixed rate on a written scope, and you own the code.

The market brings together dashboard software vendors, conversational analysis tools and agencies that design custom solutions^[6][9].

This list is not exhaustive.