AI Strategy
Lots of AI activity. Not enough business impact.
The useful question is not where you can use AI. It is where the business is constrained, and how human plus artificial intelligence can improve flow through that constraint.
Why do AI pilots stall?
AI pilots stall when they prove a tool can produce something, but never prove that a workflow, decision, customer experience, or business result changed. The gap is rarely lack of AI activity. It is weak value realization.Move beyond deliverables
AI work has too much uncertainty for annual project logic. Treat early AI effort as buying information: what workflow will change, who will use it, what evidence would earn more funding, and what should stop.
Context makes AI useful
Useful AI depends on boundaries, examples, data, decision rules, and workflow knowledge. Prompt tricks help less than a clear picture of the work the AI is supposed to change.
Confidence by phase
You cannot guess your way to AI value. Move from upfront commitments to staged funding where confidence is earned phase-by-phase through concrete evidence.
What should your AI strategy answer?
These are the questions that separate AI theater from useful AI transformation.
Where is the constraint?
Ask 'Where would we hire AI to move the needle?' rather than 'What tool should we buy?' Start with your business constraints.
What outcome should move?
Ground AI initiatives in OKRs that measure real business impact, not just pilot completions or feature launches.
Who owns adoption?
Tool ownership may sit in IT, but value usually shows up in Sales, Legal, Marketing, Finance, or Operations. Adoption needs a business owner, not just a rollout plan.
What should we learn first?
Run short, evidence-driven experiments with clear hypotheses to minimize the cost of learning.
What evidence changes funding?
Shift from schedule policing to evidence-based governance that manages risk through fast feedback.
How will the workflow change?
If AI saves five minutes in one step but adds a day of review downstream, nothing improved. Name the workflow change before you scale.
Patterns worth keeping in view
Useful if you are trying to move beyond pilots and tool rollout
What do the better AI organizations do differently?
The organizations winning with AI are not just using better models. They are redesigning workflows, clarifying ownership, limiting active bets, and reviewing evidence before they scale."Context is the real training data."
The more clarity you give teams about users and success criteria, the smarter their AI experimentation becomes.
"Don’t automate a broken process."
Applying AI to a legacy waterfall system just makes you fail faster. Fix the flow before you accelerate the work.
"Deliver outcomes, not scope."
AI value is found in solving problems (churn, speed, quality), not in meeting a project plan.
Common Questions
AI strategy questions leaders are already asking
Short answers, with the deeper articles linked where they help.
Why are our AI pilots not producing ROI?
Most AI pilots stall because they prove that a tool can produce something, not that the business workflow changed in a measurable way. Start by asking where the business is constrained, who owns adoption, what evidence would justify more investment, and what should stop. For a deeper dive, read Why Your AI Effort Has Activity But Not Impact and If AI Coding Made Engineering Faster, Why Isn't The Business Faster?.
How do we move from AI adoption to business impact?
Do not treat adoption as the finish line. Tool usage, training completion, and pilot counts are useful signals, but they are not value. The useful move is to connect AI work to a specific business constraint, redesign the workflow around that constraint, and review evidence of behavior change or business improvement. Start with Why Your AI Effort Has Activity But Not Impact.
Where should we start with AI transformation?
Start where traction is missing, not where the tool is easiest to apply. Look for work waiting on decisions, adoption, product judgment, review, data access, or value evidence. Then aim human and artificial intelligence at that constraint. If AI coding is already changing engineering, read If AI Coding Made Engineering Faster, Why Isn't The Business Faster?.
What operating model is needed for AI value realization?
You need a way to see AI bets as a portfolio, limit work in progress, fund learning in stages, clarify outcome ownership, review confidence instead of status, and treat adoption as product adoption. The operating model matters because AI creates speed, not automatic value. For a concrete executive example, read From Personal Productivity To AI Operating-Model Change.
What changes when AI coding makes engineering faster?
The bottleneck often moves away from coding and toward product judgment, review, launch, adoption, and value evidence. That means the leadership question shifts from "how much faster are developers?" to "where does this new speed improve flow to value?" Read If AI Coding Made Engineering Faster, Why Isn't The Business Faster? and Is Spec-Driven Development a Step Forward or Back for Product Development?.
Did AI make Agile obsolete?
No. AI makes some agile mechanics look optional because production is getting cheaper, but it makes the original purpose of agility more important: learning, steering, adapting, and turning capability into value. The bottleneck moved from building software to helping organizations absorb what can now be built. Read AI Didn't Kill Agile. It Moved the Bottleneck.
How should we set goals for AI agents?
Be careful not to define every agent goal as an output condition. "Tests pass" and "backlog empty" are useful, but the bigger question is whether the work changed user behavior, workflow performance, or business results. That requires better observability. Read AI Agents Can Now Run Toward Goals - Are Yours Worth Running Toward?.
What does Yuval help with?
I help leaders sort out where AI can actually move business results. Sometimes that means portfolio work: which AI bets deserve funding, what evidence changes confidence, and how adoption will happen. Sometimes it means product and engineering flow: why coding got faster but specs, review, release, or adoption did not. The two service pages show what those conversations look like: AI Transformation Advisory and AI Product Development Lifecycle Advisory.
Trying to make sense of your AI next move?
The pages below go deeper into the two patterns I see most often: AI portfolios that need better value evidence, and product or engineering systems stretched by faster code generation.