Your AI operating model should answer 35 questions
If ever there was a fast-moving, disruptive technology – this is it. Crawling caterpillar-like for decades AI was grounded in the realms of rules-based process automation, before flying from its chrysalis in 2023 to capture worldwide attention in the butterfly-like form of GenAI.
We spent 2024 recovering from the shock and reviewing the risks, but did little to assuage them. And now, in 2025, they're largely forgotten - with all attention on the AI opportunity.
Whatever ‘opportunity’ means to your organisation, you’ll face some common challenges in adopting AI sensibly – to strike the best balance of risks and rewards. Note that this balance will be specific to the organisation's mission and purpose, ownership, strategy and appetite for risk.
An exercise in operating model design is needed here - but it’s something very few people are talking about.
In case the term is new, an organisation’s ‘operating model’ is the set of decisions about how the organisation should work (AKA 'how things get done around here'). As such, it answers questions about the roles and skills that are needed, how they’re organised, what processes are followed, what oversight, measures and controls apply to those processes, the information, tools and other assets needed for all that, and the locations where it’s all done.
An operating model is about getting some control and consistency, so things can be done consistently well everywhere. But it needn't constrain creativity – indeed the best operating models I'm involved with do exactly the opposite.
It's often desirable to revisit an operating model in the light of new technology (e.g. cloud, IoT, blockchain, ...) But ubiquitous, low-cost, intelligent computing brings a few new questions too - operating model question that simply didn't arise before 2023, and that even now have garnered precious-little attention.
Organisations might like to ask themselves …
1.1 which roles are going to be critical to our AI success?
1.2 which recruitment battles do we need to win, and how are we going to do that?
1.3 given that substantial AI experience is extremely marketable, how will we retain key people who make a difference?
1.4 how do roles change more broadly around the organisation?
1.5 how are we going to train and enthuse the wider organisation quickly, so they can play their parts?
1.6 which skills will morph or decline, as AI engines pick-up tasks themselves?
2.1 what new organisational capabilities do we need (e.g. specialist market research, data preparation, procurement, design, prototyping, …)?
2.2 for now, should AI be loosely federated across business units and OpCos? or centralised? or do we run a tailored hub-and-spoke system?
2.3 who will run what parts of the resulting organisation?
2.4 what information and other resources do they need to do this well?
3.1 what tools will help us (e.g. for wireframing, LLMs, hosting, training, testing, …)?
3.2 how can we (possibly!) keep pace with AI market development?
3.3 how will we detect when there’s a better supplier solution to try?
3.4 how rapidly do we need to react to these changes?
3.5 what measures should apply in adopting supplier solutions so as to allow those supply chain swaps?
On AI as the operating model
4.1 should we consider devolving some parts of our operating model into AI itself, so processes, thresholds, controls etc. adapt intelligently?
4.2 where should this happen automatically?
4.3 where should it happen with human approval?
4.4 where should we draw the line here? and why?
5.1 which additional decisions can now become evidence-based?
5.2 which AI engines can be trusted to assist this?
5.3 how much evidence is adequate for humans to have confidence, especially given hallucination rates?
5.4 how much AI explainability is needed to enable that?
6.1 what constitutes responsibility, and to whom should we act responsibly?
6.2 how much uncertainty about negative impacts is acceptable, and can it be allowed to depend on the potential gains?
6.3 should we adopt some universal absolutes, e.g. ‘do no harm’?
6.4 how should our rules for responsibility be enshrined in policy, in employment contracts, in incentives etc.?
6.5 what committees will handle finely-balanced judgments, and with what rules and incentives?
6.6 what special treatments will apply to AI-related risks?
7.1 given the inscrutability of AI’s environmental costs, how much AI is it right to consume?
7.2 does that depend on potential gains?
8.1 what tools should be promoted, and which should be tolerated (e.g. for wireframing, data engineering, code assistance)?
8.2 what should be the basic solution delivery lifecycle?
8.3 what local adaptations are acceptable?
8.4 ... and critically, how do we develop a fully responsible, Agile culture, to get the best out of AI?
Lots of questions. But we should avoid building a new organisation here, with its own rules. The opposite should be the case - most of what’s determined in the AI operating model design will apply across the board – not just in some AI offshoot of the main organisation.
In any case, these questions should be prioritised and approached incrementally. They needn’t all be answered on day 1. To get the best out of AI, and to do so safely, we should expect to evolve our AI operating models as our understanding of AI grows. This means dedicating time and energy to nurturing it - just as we would a digital product - especially in the first 6 months, or when there are radically new market developments.
It's an active area of interest for me. So, please do get in touch if you think I can help - even if you just need a sounding board.
