Why the big idea usually stalls
The ambitious projects — an AI that runs your sales, a chatbot that replaces your support team — tend to drag on. They touch too many systems, nobody can say exactly what “done” looks like, and the first demo is impressive right up until a real customer asks something slightly unusual.
Small, boring jobs are different. They happen hundreds of times a week, they follow a pattern, and it’s obvious whether the result is right. That’s exactly the kind of work today’s AI is good at.
Where AI pays off today
These are the projects we see working again and again:
Notice what they have in common: a person still checks the output, at least at first. The AI does the first 80% and someone experienced does the last 20%.
Where it still struggles
AI is a poor fit when one wrong answer is expensive and hard to spot — final legal advice, medical decisions, anything that moves money without a human signing off. It also struggles when the information it needs lives in people’s heads rather than in documents. If your process isn’t written down anywhere, write it down first; you’ll learn a lot doing it.
A good first AI project should be a little bit boring. If it’s exciting, it’s probably too big.
A four-week pilot you can actually run
- WEEK 1Pick one task and measure it. How many times a week does it happen? How long does it take? What does a good result look like?
- WEEK 2Build the simplest version. Often an off-the-shelf model plus a small script or integration is enough. No custom training.
- WEEK 3Run it alongside your team. People keep doing the work as normal and compare their answer with the AI’s. Note every miss.
- WEEK 4Decide with numbers. Time saved, accuracy, and how the team felt about it. Scale it, fix it, or drop it.
If the numbers look good after four weeks, widen it. If they don’t, you’ve spent a month — not a year — finding out.
A word on your data
Before you paste customer data into any AI tool, check where it goes and whether it’s used for training. Business plans from the major providers generally let you switch that off, and under India’s DPDP Act you’re responsible for how personal data is handled — even by your vendors. When in doubt, strip out names, phone numbers and IDs before anything leaves your systems.