AI transformation
Operations first. AI second.
Most companies are spending millions on AI with nothing to show for it. The problem is not the technology. It is the operating model underneath it. AI amplifies whatever it lands on, so when decision rights are unclear, the cadence is weak, and the data is a mess, AI amplifies the mess. That is why so many pilots impress in a demo and then quietly die.
AI transformation is changing how a company operates with AI, not just adding tools. Done right, it starts with the operating model: where decisions are made, how work flows, and where the measurable ROI actually is. Fix that, then apply AI where the return is clearest. Sequence it the other way and you get expensive experiments instead of compounding gains.
Why most AI initiatives fail
- No clear owner, so the initiative escalates to the CEO and stalls.
- No defined outcome, so "success" means "we'll discuss next steps."
- AI bolted onto broken operations, amplifying the dysfunction.
- Pilots with no pre-commitment to scale, so they never leave the lab.
How to do it right
Everyone is pitching AI builds. Most of them are automating broken processes, and many of those processes should not be problems in the first place. So the order matters more than the tooling. Install the operating system first: clear decision rights, measurable outcomes, a cadence that forces closure, and the data discipline to trust what you are feeding the models. That is the VOOCS framework. Then build with production discipline, training and governance built in, pre-committed to scale.
And then keep it current. Foundation models turn over roughly every year, and a vendor can retire one with as little as 60 days' notice, so an AI system nobody maintains is a system already aging out. At KeyDelta this is the KeyDelta Method, one method in three acts, Advise, Build, Manage, cycling so value compounds instead of resetting: the full method is here. The execution framework underneath it: VOOCS.
Two terms I coined for this era
In 2026 I coined two terms for what is happening to software. AI-Generated Software (AIGS) is software AI builds fast: anyone can produce it now. AI-Enabled Software (AIES) is software with a standing system that keeps it secure, current, and evolving. AIGS is what gets built this week. AIES is what is still running, and still evolving, next year.
SaaS came with a vendor whose job was keeping it current. AIGS does not. Measure adoption at 90 days, not lines of code at launch. Generated is not the same as used. The full definitions live on the KeyDelta pages linked above.
The data
- MIT, State of AI in Business 2025: 95% of enterprise AI initiatives deliver no measurable value.
- SailPoint, 2025: 80% of organizations report AI agents taking unintended actions; fewer than half have governance policies for them.
- Menlo Ventures: 70%+ of enterprises now use AI in at least one business function.
- OECD: companies with proper implementation see 15 to 35% productivity improvement.
- Gartner: GenAI spending hit $37B in 2025, a 3x year-over-year increase.
- Deloitte: early adopters report average revenue increases of 15% with comparable cost savings.
The mistakes I see most
- Starting with the technology, not the business outcome. "Implement AI" is not a strategy.
- Treating AI as a side project of one team instead of a capability the whole business uses.
- Demanding perfection from AI while accepting mediocrity from manual processes that have been broken for years.
- Buying tools before fixing data definitions, so every model produces unreliable results.
- Running pilots with no pre-commitment to scale, so even the wins die in the lab.
- Shipping the build with no owner for keeping it current, so the system starts aging the day it goes live.
The keynote
I speak on this for CEO audiences, PE operating partners, and leadership offsites. The talk is Why Your AI Strategy Is Failing, and it is about getting ROI from enterprise AI by fixing operations first. See speaking.