Enterprise AI Performance: the Model vs. the Mundane

What does 'enterprise' conjure up? Imposing steel and glass skyscrapers, well-tailored executives battling in boardrooms, ambitious leaders planning go-to-market on a whiteboard? Hollywood has shaped how enterprise workplaces are viewed. Sure, we had The Office, but was Dunder Mifflin really enterprise?

So, when we talk about enterprise AI, what comes to mind: something complex, fast-paced, and mildly esoteric? The ineffable corporate flow of data and capital?

Against this backdrop, it is no wonder that the perception of successful enterprise AI deployment involves advanced models and specialist talent. We see this in the obsession with model benchmarks and endless roundtables about the agentic workforce. Consider the disappointment around GPT-5's performance and the trepidation about OpenAI's new consulting arm.

What does enterprise AI look like in practice? As someone who has spent the last decade in digital and AI leadership roles in large organisations, I can tell you – Hollywood tends to flatter us. After all, the tasks in the average enterprise context are no different in nature than those in a small business – the difference tends to be the convoluted nature of processes and systems, not to mention the overheads of aligning decisions with far larger groups of stakeholders.

Using retail as an example: restocking an item in a mom-and-pop shop might be as simple as a phone call; in a large retailer, this could entail days of effort involving approval flows and proprietary systems. The same goes for tasks like merchandising or marketing. What is interesting about these tasks is that already 12 months ago Generative AI (GenAI) models were capable of supporting them.

The bottleneck in unlocking AI's value is no longer model performance, but the ability to implement it effectively – often involving reimagining processes and integrating the AI technology with legacy systems. These two tasks are usually not shiny or exciting, and here lies the paradox. So many are waiting for the next breakthrough – the mythical AGI, the perfect reasoning model – while ignoring the transformative potential of readily available technology.

Speaking to industry peers, I often find conversation drifting to the latest innovations and model breakthroughs, instead of dwelling on the meaningful challenges regarding integration architecture, data collection, or process mapping. Cynics might see an obvious reason for this: it is far easier to talk and share 'thought leadership' than to deliver measurable impact.

This obsession with model performance is causing organisations to miss the forest for the trees. The uncomfortable truth is that for enterprise AI, often the mundane is where the money is. Those tedious tasks nobody wants to think about – invoice processing, contract review, data entry – represent billions in potential productivity gains. Current AI models are more than capable of handling them with no specialist tools required. What is needed is realism, focus, and grit.

Yet here we are, with LinkedInstagram influencers pushing low-calorie AI content to enthralled business leaders, while the average employee is stuck fighting the same broken processes day after day. We should be doing better.

Let us learn from the organisations seeing real ROI from AI – who are not chasing the bleeding edge and instead are focusing on change management over model selection; integration over innovation. The future of enterprise AI is not in the lab. It is in the back office, on the shop floor, and in the warehouse. Best of all, it is already here. We just need to stop getting distracted by all that glitters...

– Ryan

Cover image by ChatGPT.