Hi Reader

AI's an ingredient, not a meal. Most AI initiatives don't fail because the technology doesn't work. They fail because there's an element missing and no one noticed until it was too late.

My big three are Possibility, Proof, and Politics. Get two right and you still fail in predictable, expensive ways. I'll show you the three failure modes in a minute. First, the big three themselves.

POSSIBILITY - Thinking Big Enough

The tension: Most organisations approach AI from below — “what can we automate?” They digitise the status quo and call it transformation. The real competitive advantage comes from approaching the limit from above: assume everything can be done by an agent, then work backwards. Not “how do we use AI to do what we do faster” but “would we still do this if we were designing the company from scratch?”

It’s been raining AI for years.

Your 2024 AI strategy was an umbrella.

The 2026 reality needs a boat.

When you get this wrong: You build a faster horse. You operate the mess for less. Every rule of thumb in your business stays a rule of thumb, when AI could now actually calculate the right answer.

When you get this right: You ladder up not just efficiency savings, but better decisions and genuine reimagination. A leadership team maps a core process end to end, and starts crossing out steps; steps that only exist because of a constraint AI has removed. Half the process disappears. The decision you’ve been making on gut feel for twenty years, because the proper analysis was never worth the time? You can now actually calculate the right answer. You stop asking “how can AI help?” and start asking “what does AI make possible that wasn’t possible before?”

The commercial payoff: The organisations that only think in efficiency leave the real value on the table. Every rule of thumb in your business is a decision you stopped trying to make properly. AI changes that maths.

Three questions:

1. Is this aimed at a problem the business already feels, or a solution someone else is selling you?
2. If we stopped this AI initiative tomorrow, what business outcome would we lose?
3. Are you automating how you work today, or reimagining what becomes possible tomorrow?

PROOF - Testing on real work, not slides

The tension: AI is not magic. It’s software and statistical models. It responds to good briefing, good data, the right tool for the right job. But most AI business cases are built in fantasy land: theoretical ROI, vendor demos on clean data, pilots designed to succeed at nothing important. The knowing-doing gap is the actual problem: every leader in your market has been to the conferences and seen the demos. They know AI matters. They can’t convert any of it into P&L impact.

When you get this wrong: You scale something that was never proven. Or you pilot forever and never scale. Quick wins feel good but teach you nothing about the hard part. Your pilot “succeeded” and nothing changed, because it was never connected to anything the C-Suite cares about. Meanwhile, you’re spending more on the business case than it would cost to just build the thing.

When you get this right: You test on your problems, your data, your people. One domain expert told me he built in days what IT said would take months i.e. not a prototype, a production tool. A commercial team is now running an AI document review on a 200-page bundle and caught six missing documents and three inconsistencies that humans missed. You know exactly which metric moved and by how much. The business case writes itself because the evidence is already there. You pick the hard problem i.e. the one with more stakeholders and more politics, because that’s the one that ladders up to something that matters.

The commercial payoff: The most expensive AI failure isn’t building the wrong thing. It’s spending six months deciding what to build. When building is nearly free and fast, the expensive failure is the delay itself.

Three questions:

1. Have you tested this on real work, with real people, using real data. Or is the business case still a slideware exercise?
2. Can you name the specific metric that moved during the pilot, and by how much?
3. Are we sprinkling AI on top of a broken process? Is there a better (non-AI) way of solving this?

POLITICS - Owning outcomes in real organisations

The tension: This is where most AI initiatives die. Teams pilot something brilliant, then discover nobody senior enough cares to make the decision about what happens next. People complain about stakeholder resistance as though it’s an obstacle to the real work. It IS the real work. It’s not that they don’t get it, it’s that you haven’t explained it in a way that talks about something they really care about.

When you get this wrong… immediately: The pilot finishes and sits in a drawer. The team doesn’t know who can decide what happens next. At three months: The people who built it start leaving or disengaging. The vendor starts calling. The builders in the business, the ones who are already faster than your governance, route around the system entirely. You get shadow AI with no framework. After a year: The organisation has spent the money, lost the momentum, and the next AI proposal gets met with “we tried that, it didn’t work,” when actually it was never given the chance to work. Meanwhile, the only thing protecting you from an AI-native competitor is friction in your customers’ organisation. And that moat is eroding. Funnily enough, the second most common enquiry my business gets is from start-ups who have done a successful pilot with a blue chip company but now find themselves stuck and are burning runway trying to get a decision on what happens next.

When you get this right: You assign ownership before the pilot starts, not after it succeeds. The person who can say yes is already in the room, not being briefed after the fact. The team only works on problems that business leadership already cares about, so they never get slowed down asking for permission. The person who’s been quietly using AI to skip the tedious work? Instead of punishing them, you recruit them. Governance isn’t the brake. It’s what lets you drive faster. And the end state is a business leader describing objectives surpassed, with AI as a contributing factor not an AI team presenting a dashboard nobody asked for.

The commercial payoff: Your management team is conflicted: they’re proposing the AI investments they’re being asked to evaluate. Your vendors are conflicted: they’re selling the solutions your team is considering. The board needs an independent perspective that starts with business objectives, not someone else’s product roadmap.

Three questions:

1. Has management named a single accountable owner and defined what success looks like?
2. What existing investment or priority is being displaced to fund this?
3. Who has the authority to say no? And have they been asked?

THE THREE FAILURE MODES

Two out of three produces a predictable failure mode.

Possibility + Proof, no Politics = Innovation theatre. You’ve built brilliant prototypes. Nobody’s navigating the organisation to scale them. No named owner. No governance. Dead in the next budget cycle.

Proof + Politics, no Possibility = Operating the mess for less. You’ve got working tools with named owners and governance. But you’re just automating existing processes. You’ve built a faster horse. Your AI-native competitor is building a car.

Possibility + Politics, no Proof = Expensive slideware. The board has a bold AI strategy, a governance framework, and a named sponsor. None of it has been tested on real work.

Reply and tell me which P you're weakest on and I'll tell you what I see most often when that's the missing ingredient.

See you next time,

Helen