AI should be a conclusion, not a premise

Imagine a company decides it needs an AI-powered CRM. It sounds progressive. Perhaps AI-assisted development even makes building a bespoke system affordable. But what problem is the CRM actually solving? I encountered a version of this in a manufacturing business. The initial ambition was to build a CRM application to improve sales-order processing. Once we followed the work, the problem looked different. Information was moving through email. Spreadsheets were updated manually. Customer and order information was fragmented. Numerous spreadsheets sat on a shared network drive. The immediate problem was not the absence of an AI-built CRM.

It was how information moved through the business. So the interventions were much more ordinary. SharePoint and Microsoft Lists centralised information. Forms improved information capture. Workflows were revised. Power Automate handled predictable transfers. The business improved. Then something interesting happened. The bottleneck moved. Once the earlier information problems were reduced, bespoke quotation generation became the constraint. This time generative AI really did fit the work. Architectural drawings needed interpreting. Materials had to be researched. Information had to be brought together into a commercially useful proposal. AI helped produce an initial draft, while experienced people retained technical judgement and commercial authority.

This is what happens when we start with the problem rather than the technology. Sometimes AI appears immediately. Sometimes it appears later. Sometimes it never appears. All three outcomes are legitimate. The intervention space is wider than AI. We can stop, simplify, redesign, automate, assist or delegate. The useful question is therefore not: How much AI can we put into this process? It is: What does this work need? AI should be one possible conclusion of the diagnosis. It should not have to be the premise of the investigation.