Lean AI practices for building products people actually want

Executive overview

Most AI initiatives waste time and money because teams start with technology and work backwards to a problem. The right entry point is the human problem — then business priority, then AI.

AI adoption is a multi-year journey. Early wins matter mainly as stories that unlock funding and create momentum for larger transformation.

Asking "what can AI do?" instead of "what are we trying to achieve?" is the root cause of most failed AI investments.

What not to do

• Starting technology-first rather than identifying a concrete human or business problem
• Sprinkling AI into already-broken processes — take it as an opportunity to reimagine the process first
• Chasing incremental improvements when competitive advantage requires domain-specific, strategic application
• Expecting quarterly results from initiatives that need a multi-year horizon to materialise

Building a portfolio of AI opportunities

• Define AI ambitions first: what outcomes do you want to achieve?
• Look for areas of friction — find the pain before reaching for a solution
• Align every opportunity to a top business priority; if it's not a priority it won't get funded
• Assess whether the goal is to optimise an existing process or to completely reimagine it
• Think across departments and value chains, not just within a single function

Spending time in the problem space

• Surface the underlying problem, not just the visible symptoms
• Surveys often produce inaccurate representations — burnout manifests as overwork but the real issue may be how time is spent
• Double-click on symptoms: what is actually causing friction, not just what people report?
• Frame every opportunity in terms of the business, the user, and the broader ecosystem of stakeholders

Iterating and communicating progress

• Start with early-stage concepts; get prototypes into users' hands quickly
• Test hypotheses fast: can this process run in half the time? Can this product 10x output?
• Report improvements against a baseline, not just activity
• Frame early wins as success stories — they become the blueprint that scales AI adoption across the organisation
• Monthly progress reports should surface learnings, not just metrics; the technology is moving too fast for two-year build cycles

The human element and change management

• AI done well becomes invisible — users experience the benefit without noticing the technology
• The human side of adoption deserves as much attention as the technology, often more
• Data-driven change management — using AI to recommend the next adoption action — is itself a high-value use case
• Mindset and ways of working must change alongside the tooling

Portfolio mindset, pivoting, and parking

• You cannot identify the strongest use cases upfront — uncertainty is inherent, not a planning failure
• Treat AI initiatives as a portfolio of bets; counting cards beats gambling
• Create a culture of experimentation: prove ideas cheaply before committing resources
• Pivot when discovery reveals a more pressing priority than the original use-case list
• Park initiatives that aren't proving value — reallocating resources is a feature, not a failure
• Embrace the chaos and ambiguity; envision the future state and work backwards from it

Three priorities for any corporate AI leader

1. Human-centred: know who you are solving for and how significant their pain is
2. Brutal prioritisation: build a strong hypothesis and business case; if it is weak, return to the problem space
3. Experimentation mindset: stay open to failure, treat the work as a portfolio, and iterate toward the candidates that prove out