AI Isn’t the Problem. Delivery Is.

Dave Findlay
Aug 26
3 min read

A recent MIT study found that 95% of enterprise GenAI pilots are failing to deliver measurable impact.

Two people stand puzzled by a tangled line labeled "AI Pilot" on an easel. One says, "Maybe we should define the project first."

Not because the models are broken, but because the delivery process is.

Despite the buzz, the billions in investment, and the proliferation of copilots and pilots, most AI projects are going nowhere.

The MIT study, “The GenAI Divide,” uncovered why:

“Only 5% of generative AI pilots have achieved measurable value to P&L."

The rest? Underperforming or stuck in endless experimentation.”

So what’s going wrong?

Let’s break down the key findings and show how our Define-to-Deliver model directly addresses them.

MIT Finding 1: Projects lack clear use cases and measurable business goals.

Translation: No one defined the problem.

This is where most AI projects fail before they start. A team spins up a pilot because “we should try GenAI”, but no one has articulated:

• What real-world problem it’s solving

• Why it matters to the business

• How we’ll know if it worked

✅ Define → Deliver Response:

In the Define phase, we start with:

• Business context

• User pain

• “What does success look like?”

• One-sentence problem framing everyone can agree on

If your project can’t pass that bar, it’s not ready for AI or any investment.

MIT Finding 2: Most companies plug GenAI into workflows without rethinking how the work actually happens.

Translation: They’re optimizing broken processes.

Even good AI tools won’t deliver value if you simply bolt them onto existing reporting pipelines or legacy ticketing systems.

✅ Define → Deliver Response:

The Design phase fixes this.

We don’t build on top of old process debt, we co-create new workflows that are:

• Focused on the real need (not just features)

• Designed with the business, not just for them

• Prioritized using MoSCoW to keep V1 realistic and relevant

When you design with the people who’ll use it, adoption stops being a problem.

MIT Finding 3: Executives over-index on customer-facing use cases instead of internal operations, where ROI is often higher.

Translation: Misaligned priorities.

Chasing demos and dashboards for customers might look good, but the real impact often comes from back-office improvements — process automation, internal workflows, internal decision support.

✅ Define → Deliver Response:

Our Align phase brings the right people together, not just to score ideas, but to prioritize based on:

• Business value

• Readiness

• Capacity to co-own the solution

We don’t just ask, “What’s exciting?”

We ask, “What’s executable?”

That’s what keeps initiatives moving and valuable.

MIT Finding 4: Pilots are run in isolation, without business ownership or delivery accountability.

Translation: No one’s actually responsible.

AI becomes someone’s side project. There’s no ongoing feedback loop, no path to scale, no learning built into the process.

✅ Define → Deliver Response:

In our Deliver phase, we solve for:

• Shared ownership

• Weekly feedback loops

• Embedded business champions

• Outcome measurement (not just feature delivery)

We don’t deliver “to” the business. We deliver "with" the business.

MIT Finding 5: Tools improve fast. Organizations don’t.

Translation: AI is evolving faster than the org chart.

Many teams can now generate the work (SQL, content, visuals) instantly. But it still takes weeks or months to ship anything because workflows, governance, and trust haven’t caught up.

✅ Define → Deliver Response:

Our entire methodology is built to match the speed and flexibility that modern tooling enables.

Working in 3-month delivery cycles, we define success up front, measure real usage, and adapt quickly based on feedback. This builds momentum and trust and avoids the “pilot purgatory” most GenAI efforts fall into.