Floemdata
AI

5 Common Failures with AI at Work

Five common mistakes in business use of AI tools, each with a fix: confidential data, unchecked output, leaving the math to the model, and more.

Floemdata · · 3 min read
A chain of five nodes with a gap and a warning triangle where the third step breaks

AI tools spread through teams without approval or rules. The five mistakes below are general observations; they do not describe any company or tool. Each comes with a cheap fix.

Five mistakes and their fixes

  • 1. Pasting confidential data casually. A customer list, salaries or a contract text are pasted without reading the tool's data terms (use for training, retention period). Fix: tie which kind of data may go into which tool to a written rule; mask personal data when it is not needed.
  • 2. Using the output without checking it. A model can write fluently and wrongly. Fix: add a step that compares critical output (numbers, quotes, dates) with the source.
  • 3. Leaving the calculation to the model. Totals, ratios and currency conversion are not a model's strength. Fix: have a spreadsheet or code do the calculation and let the model only comment on it.
  • 4. Locking in to one tool or provider. Model versions change every few months. Fix: test regularly with your own tasks and set things up so you can switch.
  • 5. Use with no owner and no record. Nobody knows who uses which tool for which job with which data. Fix: name an owner for every use and keep a record of critical uses.

Which first?

Our suggestion is to start with 1 and 3: a data rule is a one-page text and where the calculation happens is a design decision; both are cheap and lower risk quickly. In the design of Floemdata Analytics, too, the calculation is done in code, not by AI (a design principle; an AI assistant is on the roadmap, not verified).

Start small: one team, one tool, a one-page rule and a monthly review. Questions: info@floemdata.com.

← All guides