Stop Looking for AI Use Cases. Start Looking at AI Outcomes.
I think we’ve got enterprise AI adoption fundamentally backwards.
Walk into almost any large organisation today and ask about its AI strategy, and you’ll probably be shown a long list of use cases. Each carefully assessed for feasibility, business value and implementation complexity, with the inevitable low-hanging fruit prioritised first.
It all sounds perfectly sensible.
Except there’s a fundamental problem.
We don’t run businesses as collections of isolated use cases. We run a collection of outcomes that cumulatively are the business.
And I increasingly believe our obsession with identifying individual AI use cases is causing us to miss a much bigger opportunity.
The illusion of low-hanging fruit
Imagine a bank wants to improve its mortgage application process.
A typical AI discovery exercise might identify opportunities to summarise documents, draft customer correspondence or help underwriters interpret policies.
Three perfectly reasonable use cases. Three potential successes.
But the mortgage application still travels through the same departments, handoffs, approval stages and queues.
We’ve made individual activities faster without necessarily improving the overall business outcome.
There’s a principle in computing known as Tesler’s Law, or the Law of Conservation of Complexity. It suggests that systems contain inherent complexity that can’t be eliminated but rather moved around.
I think this is particularly relevant to enterprise AI.
Automating an individual task doesn’t necessarily remove its surrounding complexity. It might simply move that complexity elsewhere, introducing new validation requirements, integrations or dependencies.
The easiest activity to automate isn’t necessarily the one preventing the business from achieving its outcome.
Yet our fixation on low-hanging fruit means we may be systematically avoiding the complex relationships and dependencies where the greatest value exists.
We’ve mistaken the use case for the architecture
I think there’s a deeper issue.
We’ve started treating the AI use case as the fundamental building block of our operating model.
Find a task. Build an AI solution. Connect it to existing systems. Deploy it. Repeat.
Each implementation becomes another isolated capability with predefined inputs, outputs and responsibilities.
We’re taking one of the most flexible technologies we’ve ever developed and implementing it through some of our most rigid design patterns.
AI becomes another technology inserted into an existing process rather than an opportunity to rethink how the operation works.
The result is what I’d call use-case paralysis.
Organisations become increasingly capable of deploying individual AI solutions but struggle to translate those successes into meaningful business transformation.
Operational continuity is what we’re missing
What if we started with the outcome instead?
Rather than asking which activities in a mortgage application could benefit from AI, ask what the best possible mortgage application experience would look like if we redesigned how it was delivered.
Which decisions are necessary? Which controls must remain? Where is human judgement important? Which systems need to remain authoritative? And which activities only exist because of historical limitations in how we’ve organised work?
Some parts of the operation will be flexible. Others won’t.
Get David Fearne’s stories in your inbox
Join Medium for free to get updates from this writer.
And that’s absolutely fine.
Operational continuity doesn’t mean preserving today’s processes. It means preserving the ability to deliver the business outcome while allowing how it’s delivered to evolve.
The objective isn’t to insert AI into every available activity. It’s to create an operating model where humans, AI and traditional systems can work together, with responsibilities changing as capabilities improve.
The business continues delivering its outcomes, even as the way it operates changes dramatically.
From isolated automations to AI coworkers
This connects directly with something I’ve written about previously: rethinking the client-server paradigm for enterprise AI.
I don’t believe the future is an ever-growing collection of AI applications, each designed to perform a narrowly defined task.
Instead, I see AI increasingly operating as a coworker alongside people and traditional enterprise systems.
Systems of record remain authoritative. Systems of execution continue performing their essential functions. Humans contribute judgement, expertise and accountability.
But AI introduces a flexible layer of intelligence that can coordinate work, interact with systems, engage people and adapt how outcomes are achieved.
Rather than repeatedly engineering isolated automations, we design the working relationships between these participants, supported by appropriate governance, permissions and controls.
As AI capabilities improve, responsibilities can shift without constantly redesigning the entire operation.
The flexibility of AI becomes a characteristic of the operating model itself, rather than something we try to contain within individual use cases.
Start small, but don’t think small
The obvious objection is that this sounds like an enormous transformation programme.
But I think we’re confusing starting small with thinking small.
There’s nothing wrong with incremental delivery. The mistake is selecting those increments before understanding the outcome we’re trying to achieve.
Starting with an end-to-end business outcome doesn’t mean transforming everything at once.
It means that every improvement, however small, contributes towards something meaningful.
And importantly, we measure success against what the business actually cares about.
Not how many documents AI summarised or how many individual tasks were automated, but whether customers are served faster, costs are reduced, quality improves and the business becomes more adaptable.
The redesign we actually need
I suspect one of the biggest barriers to enterprise AI transformation isn’t failed AI use cases.
It’s successful ones.
Successful implementations that demonstrate local efficiency while reinforcing an operating model that remains fundamentally unchanged.
We celebrate individual successes and wonder why the overall business isn’t transforming.
I think it’s time to change the starting point.
Start with the business outcome. Understand the complete operation. Identify what must remain stable and what can evolve. Then design how human colleagues, AI coworkers and traditional systems can collectively deliver that outcome.
Use cases will still have a place, but they should emerge from this design rather than dictate it.
The use cases should emerge from the business outcome. The business outcome shouldn’t have to emerge from a collection of use cases.
The real opportunity isn’t to build a business with hundreds of successful AI use cases.
It’s to build a business that can continuously evolve how it works without losing its ability to deliver value.
Stop designing use cases for AI to execute. Start designing businesses in which humans, AI and traditional systems work together to deliver outcomes.
