AI adoption inside large organisations still feels early, even though the hype is loud.
Most teams have only had broad access to modern generative AI for a couple of years, so maturity shows up as “pockets of excellence” rather than a consistent enterprise capability. In regulated industries like banking, that gap is more visible because security, compliance, and governance requirements are non-negotiable.
As BBD’s Matthew Barnard notes on the latest episode of Tech Unboxed with BBD, enterprise adoption remains in its infancy: “I see a bit of pockets of excellence, but if you take the average across the whole organisation, I think we’re still very early in the curve. And with things changing so rapidly in AI, there’s not time for maturity to set in.”
Leaders also keep circling the same hard question: ROI. Everyone can see experiments, pilots, and internal rollouts like Copilot, but many struggle to measure business value in a way that survives scrutiny from risk teams, finance, and the board. Barnard highlights this industry-wide hurdle: “I see across most of our clients, everybody’s still asking the ROI question: ‘We’re doing these things, we’re learning, but are we actually seeing any ROI at the moment? How do we measure that?’ Those are big unanswered questions.”
Navigating the buy, build, or boost decision
A practical way to frame AI strategy is the buy, build, or boost decision:
- Buy means off-the-shelf software or SaaS that you can adopt quickly
- Build means custom engineering and integration when you need something unique
- Boost is the hybrid that is showing up most often: using existing platforms, packages, and data, then adding integrations or an abstraction layer to extend what you already own
For big enterprises, boosting often beats launching a huge new program because there is already a large application estate with underused licenses and partially adopted tools. Boosting focuses on sweating existing assets, improving efficiency, and meeting new requirements without reinventing everything. “The boost part does actually have AI as part of it… It allows us to interrogate systems, interrogate requirements, and to build new systems on top of existing systems in a well-engineered way,” Barnard explains during the discussion.
Overcoming the “two-day demo, three-month production” gap
A common misconception is that AI automatically makes software creation trivial, so buying products becomes unnecessary. AI can accelerate parts of development, but it does not eliminate the end-to-end work: requirements, architecture, security reviews, deployment pipelines, production support, and operating costs.
The biggest friction appears when teams move from a great demo to a production-grade system. Governance and security controls were designed for safety, but many policies have not yet adapted to the pace of AI change, creating a frustrating gap between expectation and delivery. Barnard captures this common enterprise tension in the episode: “That’s where the frustration currently in organisations is building up: they see the demo—two days later, bang, got a really nice demo. Okay, cool, now let’s take it to production. Three months later, they’re going, ‘Why is it not in production yet?’ Well, the security questions, architecture questions, and costs of running various models haven’t really been answered for production-like environments.”
Protecting your core differentiation
Differentiation also does not come from the model alone. Even with converging model capabilities, outcomes vary because organisations provide different data, context, prompts, constraints, and operational workflows.
The real “crown jewels” are not a single piece of code, but the core value proposition: the unique way you serve customers, and how that is translated into business processes and systems. You can let AI help execute, but you should not outsource ownership of that value. As Barnard emphasizes: “You as an organisation need to own that core value proposition you have, and how your actual end clients are expecting to see that. That to me is more like IP and understanding processes rather than necessarily a piece of code.”
A shift toward incremental, outcomes-based procurement
For procurement and leadership, this shifts evaluation away from simplistic licensing comparisons toward outcomes-based thinking: define the change you want, deliver it in small increments, measure impact, and avoid long commitments in a market that can be out of date next month.
When asked what single piece of advice he would give an enterprise about to sign a major agreement, Barnard advises on the podcast: “Don’t sign for a long time. Things are changing so rapidly. Try to commit to whatever you do in really small chunks about what you’re achieving, because whatever you sign up for today may be out of date tomorrow or out of date next month.”
Want to dive deeper into enterprise AI implementation strategies and hear the full discussion on buy, build, or boost? Listen to the full episode of Tech Unboxed with BBD featuring Matthew Barnard below.
