Businesses are under pressure to show what they’re doing with AI – and to demonstrate results. Yet promising pilots often falter when the technology has to work inside a complex enterprise.
Enterprise systems come with legacy dependencies, sensitive data, established controls and people who need dependable outcomes. An impressive model is only a starting point.
In a conversation about AI in the enterprise, BBD’s Dieter Rosch and Russell Davidson explore where AI is genuinely creating value, why some projects never make it beyond experimentation and what organisations should prioritise instead.
The organisations seeing the greatest return from AI aren’t necessarily the ones adopting the most technology. They’re the ones solving the right business problems.
Why enterprise AI conversations often start in the wrong place
Which model? Which platform? Should we build agents? These questions often come before anyone has defined the problem.
When an initiative begins with a tool, teams can end up searching for a problem that justifies it. A technically impressive pilot may generate excitement without changing a business outcome. Meanwhile, the underlying bottleneck remains untouched.
Start instead with what costs the business time, money or opportunity, then decide whether AI – or another change – can improve it.
That distinction is central to AI business value. Before investing, organisations need to understand their existing systems, processes and constraints. A structured technology assessment can help expose those dependencies and give the initiative a realistic starting point.
Key takeaway: Enterprise AI succeeds when organisations focus on business outcomes rather than technology adoption.
The most valuable AI use cases are usually the least glamorous
The strongest opportunities don’t always involve reinventing a product or launching a customer-facing AI assistant. Often, they’re buried in work the business already does every day.
Think of fraud detection that helps investigators prioritise suspicious activity, workflow support that puts relevant information in front of employees, or process automation that reduces repetitive administration. Small improvements across thousands of transactions can add up.
The point isn’t to automate everything. It’s to find the steps where better predictions, faster access to information or fewer manual hand-offs genuinely improve performance. That often means enhancing existing software and operational workflows rather than replacing them wholesale.
“The best AI opportunities are often hiding inside existing operational processes.”
Why AI projects are really data and process projects
A capable model cannot compensate for information that’s incomplete, inconsistent or inaccessible. Nor can it repair a process nobody has properly mapped.
AI adoption depends on the less visible foundations: usable data, clear process ownership, integration with existing systems and a plan for how people will work differently. If an AI tool produces a recommendation but employees must still re-enter information into three other systems, the supposed efficiency gain quickly disappears.
Before deployment, teams must establish which data is trustworthy, where decisions happen, when humans intervene and how results enter everyday operations.
Successful AI implementation needs engineering discipline as well as modelling expertise. Treating AI as a standalone technology purchase overlooks the very conditions that determine whether it will be useful.
Governance has become a business requirement
As AI moves closer to operational and customer decisions, governance becomes inseparable from value. A result that arrives faster isn’t necessarily better if nobody can explain it, challenge it or take responsibility for it.
Can the organisation trace how an outcome was reached? Who is accountable when a recommendation is wrong? What information was used, and what happens when the system encounters something unexpected?
Leaders must define risk, accountability and oversight, especially in regulated environments. Explainable AI can support that accountability by making model outputs easier to interpret, but explainability is only one part of the picture. Human review, monitoring, security controls and clear escalation paths matter too.
Governance should make useful AI safer to deploy and easier to trust – not become an afterthought once a pilot is ready to scale.
What causes AI initiatives to stall?
Pilots benefit from narrow datasets and controlled conditions. Production is less forgiving.
Poor data quality can undermine results. Unclear ownership leaves nobody accountable for the outcome. Weak adoption means an apparently successful tool sits unused. And integration, security and operational costs can make scaling a pilot far harder than building it.
The gap between promising AI value and sustained AI value realization is therefore rarely explained by model performance alone. It’s an execution challenge – and one worth confronting early rather than after the investment has been made.
A simpler way to think about AI adoption
Begin with one problem, give it an accountable owner and define success before testing a solution.
That could mean reducing the time needed to process a case, lowering false-positive rates, improving response quality or cutting manual rework. Measure the baseline, run a contained experiment and compare outcomes. Strengthen and scale what works; learn quickly from what doesn’t.
For organisations adopting AI in business, this creates a practical path to enterprise AI adoption without making technology deployment an end in itself. It also forces a useful distinction: AI value comes from sustained improvements to how the business operates, not from the number of tools employees can access.
That’s the essence of AI for business value: making the return visible through better decisions, less friction, lower risk or greater capacity.
Key takeaway: Successful AI adoption starts with a measurable business challenge, not a company-wide AI mandate.
Watch the full discussion
The full conversation goes deeper into where AI delivers measurable returns, common implementation mistakes, governance considerations and practical starting points. Dieter and Russell also explore real-world enterprise examples, including the complexities behind seemingly straightforward use cases.
Which problems deserve investment? How should automation and human judgement work together? And what does it really take to move from pilot to production? These questions rarely have answers that fit neatly into a product demo. Hearing the reasoning behind the trade-offs is what makes the discussion valuable.
Watch the full conversation with Dieter Rosch and Russell Davidson to learn where AI is creating tangible business value and how organisations can move beyond experimentation towards measurable outcomes.