Failure at Scale

We are seeing some very worrying AI-adjacent trends in the market. But we are also seeing projects achieve amazing things, both with AI in the solution architecture and with teams using AI tools to get more done in less time.

So look, no one wants to hear this, but it's true. Most AI initiatives are dead before they start. Forced AI rollouts, AI as a usage rather than an outcome metric, throwing phrases like agentic-first around without really understanding the limits of its complexity management, and slimming down tech teams in the hope that AI will pick up the slack.

Here's what we are seeing and what to do instead.

Who'd want to be a CEO in the AI age? Boards and investors are asking for AI stories. Product vendors are selling their golden visions of transformation. Consultancies (ahem) all pitching AI-first packages. And the poor CFO being asked to commit to budgets for tools, consultants, and org changes that nobody has properly evaluated in the hope of outcomes that nobody has properly defined.

We've seen this before, but never on this scale. It does not end well.

At Modu, we have a better way to build an AI story that will hold up under scrutiny.

Think of a Value Stairway. Three steps to heaven.

Step One

Pick one use case. Just one. You can add more later. A delivery process. An app feature. A support function. Fully define and instrument it. Fight about it. Debate it. Establish a baseline. What does the process cost today? How much time does it take?

Define some success criteria. What would a better alternative look like? Debate these. Fight about them. Make sure success means success and not usage or activity.

Define a change in the world you would like to see.

Test one AI intervention against that baseline. Just one. You can add more later. And do it quickly. A few weeks max. Anything longer and you'll forget what you are trying to achieve, and conversations will be more about means than ends.

Do not change anything else. Do not downsize your team. Do not announce anything on LinkedIn.

Step Two

The verdict. What happened? Be honest. Debate it. Do the numbers support the success criteria? It's ok if they don't. You've not wasted an opportunity. If things look good, move to step three. If they don't, either pick another use case or, better still, argue about another AI intervention to try.

Got no more ideas? Fine. What about a non-AI intervention? Yes. Heresy. But look at it this way: you've invested some time in debating and understanding a change you'd like to see. Are there really no improvements you can make?

The chances that the final intervention(s) include some form of AI are high. Just remember that AI encompasses many paradigms, algorithm families, architectures, training methods, and application subfields. You've got literally hundreds to choose from. It's always been much more than a superficially clever chatbot.

Step Three

Scale what works. Only those interventions that have proven their value, in your specific context, in your specific workflow, should be taken further.

This gives you a board-ready AI narrative built on evidence. What a story you will have to tell.

It also gives you cover when (not if) the broader market corrects.

Your decisions will have been defensible at every stage.

The AI story your board wants is not that you spent a fortune on the thing everyone is buying, but hardly anyone understands.

It becomes we know exactly what our AI investment is returning.

If you'd like to find out more about how we do this for our clients, drop us a line.