Why 95 percent of AI projects fail
Liam Otley at Morningside AI has spent two and a half years implementing AI for some of the world's biggest brands. His conclusion is simple and stark: 95 percent of all AI initiatives in companies deliver zero return on investment. Not because the technology is bad. Because companies do it in the wrong order.
They do it backwards
The pattern is always the same. A leadership team sees a headline, buys a tool, and tries to plug it into a chaotic system with scattered data and processes that nobody fully understands. Then they wonder why it does not work.
MIT recently published a study with the same number Otley cites: 95 percent of AI initiatives in enterprise companies deliver no measurable return. That is not a claim that the technology is overhyped. It is a claim that most companies have not done the groundwork.
The five percent who actually see a return do the opposite. They start with the process, not the technology. They map how the business actually works, including all the ugly parts, before writing a single line of code or buying a single tool. Otley describes it as becoming the world's leading expert on how a specific company operates: deep-dive interviews with everyone from department heads to frontline staff, visual process mapping, bottleneck identification, and prioritisation of where AI actually delivers value today.
That work takes time. But it is what separates the companies that succeed from those that do not.
The quick wins sit in the most boring processes
One of the strongest insights from Morningside's client work is where the quick wins actually are. Not in the impressive projects that look good in a presentation. In the boring ones.
Manual data entry. Report writing. Document lookup. Just let me check that for you. Those are the processes that save hundreds of thousands of pounds in staff time when you automate them. And they are relatively simple to build, relatively quickly, at relatively low cost.
The logic is straightforward. These processes are well defined. The input is known. The output is known. The rules are clear. That is exactly what AI agents are good at: running well-defined transactions at high volume with high precision. The more exciting projects — the predictive systems, the strategic decision tools — require infrastructure that most companies do not yet have in place. They may be right in time. But they are rarely the right first step.
Otley is direct about why most companies get stuck here: they are searching for the impressive project they read about in a newsletter, not the well-defined bottleneck that is costing them time and money every single day. That is the wrong question to ask.
What this means for your back office
Morningside works primarily with mid-market companies, fifty to five hundred employees. Exactly the group where back office functions have typically scaled linearly with volume for years, without anyone questioning why.
Every invoice reconciled by hand. Every product data update done manually. Every order status checked in one system and copied to another. These are well-defined transactions with known inputs and expected outputs. If you can specify them precisely enough, an AI agent can run them.
But you will not find those opportunities if you start with the technology. You find them when you map the processes properly, identify the bottlenecks, and then match that to what AI actually delivers in production today.
Otley open-sources his method and hands it to anyone who wants it. That is generous. But the method requires time, genuine knowledge of what AI solutions work in production, and the ability to turn process mapping into actual systems with SLA guarantees. Most companies lack that combination internally and cannot afford to wait until they have built it.
That is the gap Lights Out fills. We do the mapping. We identify the quick wins. Then we take operational responsibility with SLA. The client buys the outcome, not the journey.