The moment we’re in with AI is one we’ve been in before. Everyone is scrambling to adopt and few key players are ready to deliberately and intentionally move the right pieces for sustainable growth. I watched a clip of Jeff Bezos doing an interview with Jay Leno, who asked why the company kept losing money. Long-term vision sometimes means being misunderstood.
Most AI initiatives fail because they solve the wrong problems. The problems that new technologies and accelerations create won’t be solved by speed alone. Leaders need to drive vision and alignment. Create clarity and drive towards the underlying problems that companies had before LLM technology became widely available.
Teams are cramming AI into existing products and high friction processes instead of asking whether the processes are even necessary, let alone AI ready. Recent studies show only 26% of companies move beyond proofs of concept to generate tangible value, with failure rates consistently hitting 70-85% reportedly. But these statistics are only painting part of the picture. Most teams are solving for technology adoption rather than business impact.
When AI falls into the Low-Impact Death Spiral
Matt LeMay calls this the Low-Impact Death Spiral, shipping outputs without clear goals until complexity makes value driven work harder. Most AI initiatives are following the same path.
We’ve seen this before with digital transformation. Organisations want to keep things small and easy for teams adopting AI, so they avoid the harder changes necessary to change data, relationships and infrastructure. This leads to more clutter, fewer goals hit, and more barriers to change.
Companies that adopted tools without redesigning their ways of working and cultures will fall short. The harder work, changing how the organisation runs, keeps getting deferred. It doesn't fail. It just never starts.
It becomes fairly easy to see how this happens. Organisations invest in keeping up without modernising old processes and ways of working. They track AI feature usage instead of business capability enhancement. They measure task completion times while missing decision quality improvements. Organisations will fail to create value if they look at productivity on an individual basis.
Software is a team sport. Product and engineering leaders need to guide the conversation and make sure the unit of measure is a team when talking about productivity. Balance the leading indicators of time saved or increased productivity with the outcome and team level metrics we’ve built the discipline around in cloud and software development for the last 20 years.
The ones that thrived didn't just adopt new technologies. Often they redesigned their models around digital-native principles. We’re at a turning point where it’s important to do the same thing again. We need new ways to think about long-term vision, value and the problems we actually need to solve as leaders.
Problems that actually matter
Here's what separates successful AI implementations from the failed experiments: they don't start with technology. They start with human capability gaps that constrain business growth.
I wrote about how good execution can be the operating system for AI Adoption. Organisations with clear processes, structure and process flow will have the competitive advantages of tomorrow because they’ll ship and adopt more consistently, more frequently, with clearer impact and at a lower cost basis.
Applied to AI, this means:
Start with Capability Gaps, Not Technology Opportunities Instead of "Where can we implement AI?" ask "What work consumes the most strategic time that could be better spent elsewhere?" Asking this question shifts the focus from technology deployment to human capability.
Measure outcomes, not outputs.
The measurement problem with AI mirrors early internet adoption: companies tracked website visits while missing the fundamental shift in customer interaction preferences. They measured digital activity, not business transformation. Instead of tracking AI feature usage, measure decision quality improvements.
Decision quality is going to look different in every organisation. It’ll be composed of the inputs, the outputs and how decisions flow in your organisation. Both human and machine decisions. One of the benefits of the age of LLM technology is it forces focus on the work itself instead of the work around the work. Context, decisions, clarity and measurement are even more important now than they were three years ago.
Decision quality hinges on a few key questions:
What is your context?
How do you and others know what’s important?
How do you and your teams act on it?
How do you measure success?
How do you know the right decision was made even if the results weren’t great to start?
Think in bigger horizons. Early wins might look modest, but capability building accelerates everything else. For tomorrow’s leading organisations, AI value comes through incremental gains at scale, rather than one-off fixes. Those are the questions that leadership teams need to focus on to keep humans and machines moving in the right direction.
Product leaders should anchor team AI impact conversations on fundamentals. The fundamentals we are (or should be) measuring now, but with an AI lens. Efficiency metrics like throughput and cycle time are evergreen. I’ll do a whole series on software team measurement ideas in the near future, because it’s an area I’ve been thinking about often the last few months.
Financial impact that matters:
Revenue Attribution: Customer actions directly tied to AI-enhanced experiences, not AI feature usage. Revenue attribution puts the emphasis on margin tied to AI, not on AI feature usage.
Strategic Cost Avoidance: Eliminated dependencies and prevented escalations, not just reduced labour hours. IKEA is a brilliant example of turning costs into a new business model through LLMs.
Capability Leverage: Teams handling increased complexity without proportional resource growth. Capability leverage shows whether a team is covering more ground with the same headcount.
Employee Experience as Leading Indicator: When internal AI makes employees more capable at strategic work, it creates sustainable advantage. When it just makes them faster at routine work, it creates temporary efficiency.
The goal is creating hard-to-copy value, and the more effective and strategically capable your average software team is, the harder it is to copy. Leaders in tech should be showing up to create the operating processes and systems that make these augmentation practices core infrastructure, as that’s where I think the biggest value in software IP is headed, not the code base, but the execution processes.
Building sustainable AI value
When it comes to optimising for AI, don’t optimise for the next quarter, optimise for the next decade. That mindset separates companies building lasting value from those chasing short-term metrics.
Teams that adopt this lens create long term sustainable outcomes, not just automations. They make better decisions faster, share knowledge more effectively, and increase decision velocity across the organisation. Companies making AI intrinsic to operations see 20% to 30% gains in productivity, speed to market and revenue through incremental value at scale, according to PwC research. These aren't one-time efficiency gains. The benefits compound as teams get better at strategic work.
Early wins might look modest, but when they enhance decision-making and human capability, they multiply over time. The teams and companies that get this right accept they'll look slow, or wrong, for a while. That's what long-term vision costs.
That’s what this newsletter is all about, and i’m grateful you’re with me.
What's one thing your team measures today that you'd measure differently starting from scratch? Reply and tell me. I read every one.
Thank you for reading.
Intentionaut is a fortnightly letter from Saielle DaSilva on design, product, data and AI. What's working, what isn't, and how to keep your bearings.
If this was useful, forward it to one person who'd get something from it. That's how it grows.
Saielle DaSilva
Putting the soft back in software