Take a bird's-eye view first
Step back from the individual task and look at how the whole thing works. Where are people losing time? Which systems don't talk to each other? Where is information being copied, chased or re-entered? What's already working well, and what's creating friction nobody has got round to fixing?
That wider view usually makes it much easier to see whether the answer is AI, better software, an integration, a process change, or some combination of them.
Quite often the biggest improvement has nothing to do with AI at all. It might be connecting two systems that currently rely on someone copying information between them. It might be replacing a spreadsheet that has quietly become business-critical, modernising ageing technology, or restructuring how customer information is stored.
None of that is especially exciting, but it creates the foundation for more ambitious improvements later.
AI is one capability inside a system
Once that foundation is in place, AI can be introduced where it does something conventional software cannot. It might read an incoming enquiry and route it automatically, extract information from a document, summarise a customer history, identify something that needs attention or suggest the next sensible action.
The important bit is that it sits inside a wider process that already knows what to do with the result. If it extracts information from a PDF, where does that information go? If it identifies a high-priority enquiry, who gets notified? If it's uncertain, does somebody review it?
That's the difference between adding AI and building a useful system.
It's also why some of the most useful applications barely look like AI. There's no separate chatbot and no interface announcing that something is AI-powered. The intelligence sits inside the workflow, helping the system understand, prioritise, translate or recommend. To the person using it, it simply feels like better software.
There are times when introducing AI earlier makes sense, and usually it's acting as a connector or a translator: helping an existing process go further or move faster, rather than replacing the foundations underneath it. Adding an AI layer too early can patch over a poor process for a while, but it's rarely the right long-term answer.
The order that works
Get the process right. Get the information where it needs to be. Connect the systems that need to talk to each other. Then add AI where it makes the whole thing meaningfully better.
AI can absolutely supercharge what a business is able to do. It isn't a shortcut for understanding the problem properly.
By Josh Gosselin, Founder, DEXM