Key points
- Before any AI project, ask what problem it should solve. Projects most often disappoint because they began with a tool: someone saw a demonstration, bought a licence, then hunted for a use.
- Successful projects tend to start in reverse: with one task that eats time, costs money or hurts quality, described plainly enough to tell whether it improved.
- Describe your chosen task briefly: who handles it, how often, how much time it eats and where it breaks down. Other ideas can wait; a first project should aim at one target, so everyone agrees what success means.
- Pick the one number that ought to move, such as hours a week, and measure it now. Without this starting figure, the baseline, nobody can judge afterwards if the project succeeded. A week of rough tallies on paper will do, if you record and keep them.
- Some apparent AI problems are really about how work is organised. A clearer template, a spreadsheet formula or plain automation, meaning fixed-rule software with no AI involved, may fix them quicker and cheaper. Good advisers point this out, so if nobody ever offers a simpler option, be wary.
- Much business software now has AI built in. Check what you already pay for: a first step may cost nothing extra, only time to configure and learn it.
- Where a decision touches money, wellbeing or someone's legal standing, a person should keep making it, with AI limited to drafting, sorting or suggesting. AI can be wrong while sounding sure, so plan to catch its mistakes.
- Begin with one task and one number, and keep a person on the decisions that matter.
Transcript
Before any AI project, ask what problem it should solve. Projects most often disappoint because they began with a tool: someone saw a demonstration, bought a licence, then hunted for a use. Successful projects tend to start in reverse: with one task that eats time, costs money or hurts quality, described plainly enough to tell whether it improved.
Describe your chosen task briefly: who handles it, how often, how much time it eats and where it breaks down. Other ideas can wait; a first project should aim at one target, so everyone agrees what success means. Pick the one number that ought to move, such as hours a week, and measure it now.
Without this starting figure, the baseline, nobody can judge afterwards if the project succeeded. A week of rough tallies on paper will do, if you record and keep them. Some apparent AI problems are really about how work is organised. A clearer template, a spreadsheet formula or plain automation, meaning fixed-rule software with no AI involved, may fix them quicker and cheaper.
Good advisers point this out, so if nobody ever offers a simpler option, be wary. Much business software now has AI built in. Check what you already pay for: a first step may cost nothing extra, only time to configure and learn it. Where a decision touches money, wellbeing or someone's legal standing, a person should keep making it, with AI limited to drafting, sorting or suggesting.
AI can be wrong while sounding sure, so plan to catch its mistakes. Begin with one task and one number, and keep a person on the decisions that matter.