Part 1 of Tenet’s Enterprise Operating Models with AI series 

Most organisations investing in AI today have a use case register, a phased deployment plan and a prioritised list of processes planned for automation. 

The assumption running underneath is that the current operating model is still fit for purpose and the tools simply change. The expectation is that injecting AI into enough processes will eventually transform the organisation. 

This assumption is worth examining more carefully. 

Where is the AI productivity going? 

In practice, teams adopt an AI tool, improve individual outputs, compress routine tasks and the pilot is declared a success.  

However, when you look for where those gains are showing up at the enterprise level, it becomes much harder to track. Time saved at the individual level is real but often does not compound into improved commercial outcomes or reduced operating costs at the enterprise level.  

This is a common conundrum. The AI tools are working. The pilots are delivering. But the enterprise is not much better off, because the structure around those tools has not changed. Nobody has redesigned the work itself, the roles, or the governance. Bolting AI onto an operating model that was designed for a different era is not the same as designing an operating model ready for AI. 

AI amplifies whatever is in your operating model today 

Operating models encode choices about how resources are organised, where decisions sit, who holds accountability, and how performance gets measured. Most of those choices were made when human decisioning and execution were the primary constraints. AI shares neither. 

When you apply AI to an existing operating model, it scales both the good and the bad. Well-designed processes get faster. But so do the structural problems the model was already carrying: unclear accountabilities, process workarounds, data inconsistencies, and governance gaps that people were quietly managing around. 

This shows up most visibly in complex legacy environments. Multi-country operations where the same process runs differently across regions. Businesses that have grown through acquisition where system and data integration was deferred. In these environments, AI does not smooth over the inconsistencies. It surfaces them at a speed the organisation is not prepared to deal with. AI running on inconsistent data produces confident errors at scale. Data foundations are not a parallel workstream to address eventually. They are a prerequisite to get AI to work properly. 

Redesigning enterprise operating models from first principles 

The organisations getting real traction are asking a different question before they deploy anything: if we were designing this operation from scratch, with AI as a native capability, what would it actually look like? 

Take a finance function. A conventional team spends most of its capacity on transaction processing, reconciliations, variance reporting, and period-end close. Analysts process, seniors review, managers approve. The headcount and hierarchy exist because of the volume and rhythm of that work. 

Designed from first principles, finance exists to allocate the organisation’s capital towards where it creates most value, manage risk before it materialises, and build the financial capacity to act on opportunities. Today, most finance functions still spend the majority of their time recording and explaining the past. 

With AI handling the execution layer, reconciliations run continuously, period close compresses from days to hours, and exceptions surface as they occur rather than during a manual review cycle. The transactional burden that consumed most of the function's capacity is largely removed. 

What replaces it is the work finance was always meant to do. Real-time commercial intelligence. Scenario analysis at the point decisions are being made. Working capital managed actively rather than periodically. Risks identified and mitigated before it compounds. The analysis lag between data and decision narrows, and better informed decisions produce better financial outcomes over time. 

The human work heavily shifts towards judgment, interpretation, and commercial decision support. The skills required are different and old reporting lines make less sense. The governance model also needs to account for agents executing tasks, with explicit decisions required on who owns accountability for setting and monitoring the parameters those agents operate within. 

Enterprise value from AI does not come from automating current roles one task at a time. It comes from going back to the purpose of the function, understanding what the processes need to produce, being clear about what requires a human, and building the model around that answer. 

Your starting point determines what investments to prioritise  

Organisations sit at different levels of AI adoption readiness, and those positions require genuinely different approaches. Being unclear about where you are starting from usually means investing in the wrong things for your actual context. 

Some organisations are building AI capability on modern foundations, with clean data architecture, integrated systems, and flexible infrastructure. These organisations can make AI-enabled operating model design choices faster, with fewer dependencies and shorter time to value. 

Most established enterprises are working from complex legacy environments where systems often predate current AI capability, data definitions are inconsistent across the business, and governance frameworks have not been revisited in years. For these organisations, an AI operating model redesign is still the right thing to do. The sequencing just needs to be honest about what has to be in place first. 

If your AI pilots are consistently performing but failing to scale, that is usually a signal that the foundations the pilot assumed do not exist across the broader environment. Clean data, integrated systems, and consistent process definitions are the conditions most pilots are built on. They are rarely the conditions that exist across the entire business. 

Being realistic about your starting position also means you stop benchmarking against organisations whose context is fundamentally different from yours and start investing in initiatives that directly address your organisation’s unique constraints.  

What has to be true to realise enterprise value from AI