Most Organisations Are Measuring the Wrong Things in AI
- aiomniconversation
- May 15
- 4 min read
The number of AI pilots an organisation has run tells you almost nothing about whether it is transforming. Neither does the number of use cases deployed, the sophistication of the interface, or the size of the AI team. These are activity metrics. They feel like progress because they are visible, countable, and easy to report upward. The organisations that are genuinely ahead are not the ones with the most activity. They are the ones that have asked harder questions: does this create measurable value for the customer?, does it create equivalent value for the business?, and can it be repeated reliably at scale? Until those three questions have clear answers, what you have is experimentation. Experimentation is not transformation.
That argument runs through my recent conversation with Moush Verma, Senior AI Lead at Santander UK, on The AI Adoption Podcast. Moush leads customer AI at one of the UK’s
largest retail banks, which means she is not theorising. She is addressing the structural, organisational, and philosophical challenges of embedding AI at scale inside a complex financial institution. The conversation covered a lot of ground, but the thread that connected everything was the same: genuine transformation requires foundations that most organisations have not yet built.
The Interface Is Not the Product
There is a temptation, entirely understandable, to measure AI progress by how impressive it looks. The fluency of a conversational interface. The coherence of a co-pilot experience. The polish of a customer-facing assistant. These things matter. They matter because customer experience matters. But they are not the transformation. They are the surface.
Moush put it plainly: a polished interface on weak foundations is never going to produce genuine organisational transformation. The market, she argued, has become so captivated by what AI sounds like that we have stopped asking whether the underlying guidance or outcome is actually better, rather than just more polished and pretty.
This matters enormously for leaders who are under pressure to demonstrate AI progress quickly. The path of least resistance is to invest in the interface layer: deploy a capable-looking assistant, collect positive feedback, and report success. The problem is that none of that tells you whether the underlying process is better, whether the data foundation can sustain scale, or whether the value created is repeatable. A compelling interface can be impressive without being transformative. Recognising that distinction is one of the more important judgements a leadership team can make right now.
Kate Jones episode link https://open.spotify.com/episode/1LqAcRFxrA1uScZWOeNRdh?si=CfS_db5QQ6GWrTFVkYlbAQ
The Scaling Trap
The most common mistake Moush sees organisations make is scaling use cases before they have scaled the capability to sustain them. She used a direct analogy. She loves to cook, and she experiments freely at home. She would not, however, claim that her ability to produce one impressive meal proves she could run a professional kitchen at consistent quality and volume. The question that matters is not can you do it once. It is can you do it repeatedly, responsibly, and at scale, which are the hallmarks of true maturity.
This is where the proof of concept trap closes in. Organisations accumulate pilots. Each one looks promising in isolation. The volume grows. Somewhere along the way, volume gets mistaken for maturity. But volume of proof of concepts is not maturity. The organisations that have moved beyond this are the ones that have built a repeatable, scalable foundation, and then measured whether each initiative delivers the outcomes it was supposed to deliver, with the measurement criteria set at the beginning of the programme, not retrofitted at the end.
The structural barrier, as Moush described it, is the fragmentation of data ownership. In most large organisations, and certainly in most large banks, data sits in silos owned by different functions with different priorities, different timelines, and different skills. There is no intelligent data fabric connecting them. Customers experience this directly whenever they are transferred between departments and have to explain their situation from scratch. The data was there. The organisation simply never built the architecture to use it coherently.
The Philosophical Question Organisations Keep Skipping
Technical capability is necessary but not sufficient. Moush made a case that I found genuinely refreshing in its directness: the organisations that will lead in two to three years will be the ones that asked philosophical questions, not just technical ones.
The question she posed is one that deserves more board-level attention than it currently gets: is AI in this organisation simply there to industrialise existing practices more aggressively, or is it there to create clearer guidance, stronger trust, and more meaningful value for customers and society? The two paths produce fundamentally different organisations. One compounds existing habits at greater speed. The other uses the removal of low-value friction to do things that were previously impossible: reaching a customer at a life-changing moment before they have to ask for help, offering proactive reassurance when a payment fails rather than waiting for a complaint, designing around the weight of the moment rather than the speed of the transaction.
Moush was clear that the winners will use AI in a way that is not just technically scalable but ethically grounded and socially sustainable. They will know where automation adds value and where human explanation and accountability must remain explicit. They will not simply have more AI. They will have better intelligence, applied more coherently, contextually, and responsibly at scale. For any leader currently reviewing their AI strategy, that sentence is worth sitting with.
The episode with Moush Verma is a reminder that the conversation about AI maturity in business needs to shift. Shifting it starts with a simple but uncomfortable question: are we measuring the right things? If the answer is not yet, then the most valuable thing a leadership team can do is agree on the customer outcomes that matter, set the measurement criteria now, involve the people who will be affected by the change from the beginning, and build the foundations before the interface.
That sequence is not glamorous. It does not produce the kind of announcements that generate press coverage. But it is the sequence that produces transformation that lasts.
Listen to the full conversation with Moush Verma on
Spotify: open.spotify.com/show/296zibtjU3w4ANuUsPSu2D and
Comments