The gap between what AI in operation is being sold as and what it is actually doing inside most businesses is significant. The pitch involves transformation at scale, automation of complex decisions, and competitive advantage that compounds over time.
The reality, for most operators who are being honest, involves a handful of genuinely useful tools, a longer implementation than expected, and a set of new problems that were not in the brochure.
That is not an argument against adopting AI in operations. It is an argument for being clear-eyed about what you are adopting and why, rather than moving because everyone else appears to be moving.
Where AI is genuinely adding value in operations
The clearest wins are in tasks that are high volume, relatively structured, and time-consuming for humans to do consistently. Such as:
- Summarising large amounts of information
- Drafting first versions of standard communications
- Processing and routing incoming queries
- Identifying anomalies in operational data are all areas where the return is real and relatively quick to realise.
In facilities management specifically, AI is beginning to show genuine value in predictive maintenance, where pattern recognition can flag equipment likely to fail before it does. That is a concrete operational improvement with a measurable financial return, and it is the kind of application worth exploring seriously.
Where AI in operations falls short
AI in operations performs poorly in situations that require genuine contextual judgement, relationship sensitivity, or ethical reasoning. It can produce a response that looks reasonable and be substantively wrong. It can generate a plan that appears comprehensive and misses the most important constraint. The quality of its output depends almost entirely on the quality of the input and the capability of the person reviewing what comes back.
The operational risk is not that AI will fail dramatically. It is that it will fail quietly, and that the teams using it will not have the experience or the inclination to catch the failures before they cause a problem.
McKinsey’s research on AI in service operations makes this point plainly. It says only 3% of organisations surveyed had successfully scaled a Gen AI use case in an operations-related domain, despite the widespread adoption at the surface level.
The failure of AI in operations is a management and training problem, not a technology one. If you want to understand what separates the businesses getting real results from those still running pilots, read more on McKinsey’s research on From Promising to Productive: Real Results from Gen AI in Services
How to think about it as an operator
The most useful frame is to start with a specific operational problem that is costing you time or money, and ask whether AI can help with that particular problem. Not whether AI can transform the business, or whether it can address this specific thing. That approach is less exciting as a narrative but far more likely to produce a return.
The businesses that will use AI in operations well are not those that adopt the most tools. They are those who understand their operations clearly enough to know which problems are worth solving, and are disciplined enough to test, evaluate, and build on what actually works.
That is the standard the Unfiltered with Brij podcast and blog apply to every operational question, not what sounds most impressive, but what actually holds up when you test it against a real problem in a real business.
Commonly Asked Questions
What is AI actually useful for in operations? The clearest wins are high-volume, structured, repetitive tasks like summarising information, drafting standard communications, routing queries, and flagging anomalies in data.
Where does AI in operations fall short? Anywhere that requires contextual judgement, relationship sensitivity, or ethical reasoning. AI can produce output that looks correct and is substantively wrong. The risk is not dramatic failure but a quiet failure that goes uncaught because the team reviewing it lacks the experience or inclination to question what looks reasonable. That is a training and management problem, not a technology one.
How should operators approach AI adoption? Start with a specific problem that is costing the business time or money, and ask whether AI can help with that particular thing. Not whether it can transform the business or can solve this. This is less exciting but far more likely to produce a return. Businesses that use AI well are not those with the most tools. They are those who understand their operations clearly enough to know which problems are actually worth solving.
Why is there such a gap between AI vendor pitches and operational reality? Because the incentive on the vendor side is to sell transformation, and the reality on the operator side is that transformation requires capability, time, and change management that the pitch rarely accounts for. Most operators who are being honest will tell you the implementation took longer, cost more, and produced less than the deck suggested. That does not make the technology useless. It makes the pitch a poor guide to reality.
Is AI adoption essential for operational competitiveness right now? Not universally. The businesses that will struggle are not those that move slowly on AI but those that move without clarity about what problem they are solving. A well-run operation with clear processes and strong people will outperform a poorly-run one with AI tools every time. The technology is an accelerant. It does not replace the underlying operational discipline it runs on top of.
