Adding AI to your business software: where it helps and where it does not
AI is very good at some business tasks and risky at others. A practical guide to choosing your first AI feature.
· Sandeep Mittal
Almost every client conversation now includes the question "Can we add AI?" Usually the answer is yes, but the value depends entirely on choosing the right job for it.
Where AI works well
- Answering questions from your own material: an assistant that reads your course content, policies or product documents and answers in plain language, with links to the source.
- Reading documents: pulling names, amounts, dates and line items out of invoices, forms and emails, then filling them into your system for a person to check.
- Summarising: a short summary of a long lead history, support ticket or student record before a call.
- Drafting: first drafts of replies, product descriptions or quiz questions that a person edits and approves.
- Sorting: sending enquiries to the right team, tagging feedback or flagging urgent messages.
Where to be careful
- Final decisions: approvals, grades, prices or anything with legal weight should stay with people. AI can prepare; people decide.
- Exact calculations: tax, fees and totals belong in normal code, which is exact every time.
- Sensitive data: send only what a task needs, choose providers whose business terms do not train on your data, and keep a log of what was processed.
A good first project
Pick a task your team repeats every day, measure how long it takes now, and build a small pilot on real data. If the pilot saves time and your team trusts the results, roll it out. If it does not, you have spent weeks, not months, finding out.