The most expensive AI mistakes are not bad tools. They are good tools bought before anyone defined the decision the tool was supposed to serve.
The pattern is familiar by now. Pressure builds to do something with AI. Vendors arrive with demos that look like the future. A pilot is launched because a pilot feels like progress. A year later there is a subscription, a slide about innovation, and no measurable change in how the business actually operates or decides.
Since this note was written, the evidence has only hardened. McKinsey’s 2026 operating-model research finds top performers are twice as likely to redesign workflows before selecting AI tools, yet roughly 79 percent of organisations skip that step entirely. Transformations led primarily as technology programmes fail at rates above 80 percent, and only 21 percent of companies have fundamentally redesigned their operating model around AI[O].
Tools are the last decision, not the first
Choosing an AI tool is the most visible AI decision, so organisations reach for it first. It is also the least consequential one, because a tool can be replaced. What cannot be easily replaced is a wrong assumption about where AI creates value in your specific business, baked into contracts, workflows and expectations.
The decisions that actually matter come earlier. Which problems are worth applying AI to at all. Whether your data and workflows can support it. Who is accountable when an AI-supported process produces a wrong answer. What adoption pace your organisation can hold without exhausting people. None of these are vendor questions. All of them shape which vendor, if any, makes sense.
Readiness is not a technology audit
AI readiness is usually read as infrastructure: do we have the data, the systems, the security. That is a third of the picture. The other two thirds are organisational. Does leadership share a realistic view of what AI can and cannot do here? Is there use-case discipline: a way of ranking opportunities by effort versus value instead of by enthusiasm? Is ownership for AI decisions explicit, or does it float between IT, innovation and whoever attended the last conference?
Organisations that skip these questions do not avoid them. They answer them by accident, through whichever tool they happen to buy.
A sensible sequence
- Start with the decision, not the demo. Name the business problem in one sentence that contains no product name.
- Assess readiness honestly across organisation, workflows, data and governance. The gaps are usually organisational, not technical.
- Rank use cases by effort versus value. Most AI portfolios contain two or three cases worth doing now and a long tail worth ignoring.
- Only then talk to vendors, with your criteria written down before the first demo. A demo is a persuasion environment. Your criteria should not be formed inside one.
| The usual order | The sensible order |
|---|---|
| 1. See demos | 1. Name the business problem |
| 2. Pick a tool | 2. Assess readiness honestly |
| 3. Find use cases for it | 3. Rank use cases by effort versus value |
| 4. Discover the gaps | 4. Then, and only then, talk to vendors |
The practical takeaway
If you are facing an AI decision, write down what would have to be true for it to pay off, before you see another demo. If nobody can write that sentence, the next step is not a tool. It is a readiness conversation.
The market is proving the point the hard way
Two newer numbers extend the argument. Gartner predicts more than 40 percent of agentic AI projects will be canceled by end of 2027, mostly early experiments driven by hype and misapplied before anyone defined the workflow they serve[GA]. And Deloitte finds 85 percent of companies expecting to customise agents to their business[DE], an admission that the tool alone was never the product. Both point the same way: the sequence is the strategy. Use case, workflow, then tool.
What I see in the field
Vendor selection feels like progress because it produces meetings, demos and a signature. Workflow redesign feels slow because it produces questions about how you actually work. I have watched organisations choose the comfortable path twice and pay for it twice. The sequence is not negotiable: use case, workflow, then tool.
Readiness and use-case discipline before vendor selection, that is the entire logic of our audit. We tell you where AI moves your results, design the workflow, and only then talk about tools. Vendor-neutral, every time.
Personalised answer within 48 hours.
Sources
- McKinsey & Company, The Operating Model Advantage: Why AI Winners Are Rewiring Their Organizations, July 2026. mckinsey.com
- Gartner, Press Release: Over 40 Percent of Agentic AI Projects Will Be Canceled by End of 2027, June 2025. gartner.com
- Deloitte, The State of AI in the Enterprise 2026 (survey of 3,235 leaders). deloitte.com
The AI Act Just Moved. Your Deadline Did Not.