Entrepreneur
Companies are treating artificial intelligence like Victorian-era physicians treated leeches: as a universal remedy to be liberally applied regardless of the actual problem. Board meetings across the country feature some variation of “We need an AI strategy” without first asking “What specific problem are we trying to solve?” The results are predictably underwhelming.
Anyway, here we are with executives demanding AI solutions for problems that don’t exist while ignoring problems that AI might actually solve.
This is expensive in ways that rarely show up on quarterly reports. Companies pour millions into AI initiatives that generate impressive demos and dismal results. They’re writing checks that their data infrastructure can’t cash. And nobody seems to notice the pattern.
Related: How to Avoid Wasting Millions on AI
The technology-first trap
The typical corporate AI journey follows a depressingly predictable path. First, an executive attends a conference where competitors boast about their AI initiatives. Panic ensues. A mandate comes down: “Implement AI across all departments.” Teams scramble to find use cases to justify the technology that’s already been selected. Consultants arrive with slide decks. Pilots are launched. Demos are built. Press releases are drafted. And a year later, when someone asks about ROI, everyone stares intently at their shoes.
This backward approach of starting with the solution instead of the problem explains why so many AI projects fail. It’s like buying an expensive hammer and then wandering around looking for nails. Sometimes you find them! More often, you discover your actual problems require screwdrivers.
The thing is, technology-first strategies make for great headlines but terrible business outcomes. They mistake motion for progress. They value novelty over utility. And often, solutions are harder to build and use than they look.
The data delusion
There’s a curious cognitive dissonance in how organizations think about their data. Ask any technical leader about the quality of their company’s data, and they’ll grimace knowingly. Yet, companies approve AI projects that assume pristine, comprehensive datasets magically exist somewhere in their systems.
Machine learning doesn’t just need data. It needs meaningful patterns in good data. A learning algorithm trained on garbage doesn’t become intelligent; it becomes extraordinarily efficient at producing highly confident garbage.
This disconnect between data reality and AI ambitions leads to an endless cycle of disappointment. Projects begin with enthusiastic predictions about what AI could accomplish with theoretical data. They end with engineers explaining why the actual data couldn’t support those predictions. Next time will be different, everyone assures themselves. It never is.
Related: Nobody Wants Another Useless AI Tool — Here’s What to Build Instead
The implementation gap
The most sophisticated AI solution in the world is worthless if it isn’t integrated into actual workflows. Yet, companies routinely invest millions in algorithms while allocating roughly seventeen dollars and thirty cents to ensuring people actually use them.
They build AI solutions that require perfect participation from employees who weren’t consulted during development, don’t understand the models and haven’t been trained to use the tools. This is roughly equivalent to installing a Formula 1 engine in a car without modifying the transmission, then wondering why the vehicle keeps breaking down.
Look, technology adoption isn’t a technical problem. It’s a human one. Humans are notoriously resistant to changing established behaviors, especially when the benefits aren’t immediately obvious to them. An AI solution that requires significant workflow changes without delivering obvious, immediate benefits is dead on arrival. Nobody wants to admit this, but it’s true.
Reversing the strategy
What would a reverse-engineered AI strategy look like? Start with identifying specific, measurable business problems where current approaches are falling short. Validate these problems through rigorous analysis, not executive intuition. Evaluate whether these problems actually require AI or might be better solved through simpler solutions. Consider the organizational changes needed to implement any solution. Then, and only then, evaluate what data and technology might address the validated problems.
A better implementation framework
Effective AI implementation requires inverting the typical approach:
-
Problems before solutions: Identify and validate specific business challenges with measurable impact
-
Data reality check: Audit existing data quality and collection processes before assuming AI feasibility
-
Simplicity test: Determine whether simpler, non-AI approaches might solve the problem more effectively
-
Organizational readiness: Assess whether workflows and teams are prepared to integrate AI solutions
-
Incremental implementation: Start with small-scale pilots focused on narrow, well-defined problems
Related: When Should You Not Invest in AI?
Training algorithms on flawed data is like building a house on quicksand. The architecture might be impeccable, but that won’t matter much when everything sinks. Companies proudly announce their AI initiatives with roughly the same level of strategic clarity as medieval alchemists had about turning lead into gold. The main difference is that alchemists spent less money.
Perhaps the most valuable AI implementation strategy is simply reversing the question. Instead of asking “How can we use AI?” try asking “What specific problems are worth solving, and might AI be the right approach for some of them?” This reframing doesn’t make for impressive conference keynotes. It doesn’t generate the same press coverage or conference speaking slots. But it does tend to produce solutions that actually work, which seems like a reasonable goal for multi-million-dollar technology investments.
Read the full article here
