Most conversations about AI start with a tool. Someone has seen a demo, a competitor has announced something, or a vendor has made a promising offer. The question becomes “should we use this?”, when the better question is “where does our time actually go?”
Start with a week of work, not a list of tools
Pick one team, such as finance, sales support or customer service, and map what they do in a normal week. Not the job description: the actual tasks. How many emails, how many copy-and-paste steps, how many reports built by hand, how many times the same question gets answered.
Most teams are surprised by the result. A large share of the week tends to go on moving information from one place to another.
Sort the tasks into three groups
- Rules-based and repetitive: invoices, data entry, status updates. These are strong automation candidates and often don’t need AI at all.
- Judgement with patterns: answering customer questions, drafting proposals, summarising documents. This is where AI tools tend to help most, with a person reviewing the output.
- Judgement that really matters: pricing decisions, difficult client conversations, hiring. Keep these human, and give people better information to make them.
Estimate the payback before choosing anything
For each candidate, estimate the hours saved per month, the cost of errors today, and what it will take to set up and maintain. A modest automation that saves a team ten hours a week, every week, is worth more than an impressive pilot that nobody opens after the first month.
Plan for adoption, not just installation
Tools fail more often because people don’t change how they work than because the technology doesn’t perform. Include training, update the process documentation, and name one person who owns the result. Check usage after thirty days and again after ninety.
If you can’t describe the task, how long it takes today and how you’ll know it improved, you’re not ready to buy the tool.
Once you can, choosing the tool becomes much easier, and usually much cheaper.
