AI consulting spending is rising, but success rates aren’t. Satish Thiagarajan, founder and CEO of Salesforce consultancy Brysa, unpacks the structural issues which could make or break an AI transformation.
Study after study points to the same pattern: the majority of AI projects fail to deliver meaningful outcomes. The tools aren’t getting worse. The budgets aren’t shrinking. So why is the gap between investment and impact so persistent?
The answer is structural. Consulting engagements are built around the parts of AI that are easiest to scope and demonstrate: models, platforms, tech stacks. The work that actually determines whether a project succeeds, data readiness, governance frameworks, change management, gets under-scoped or treated as a follow-on activity. By the time clients realise it’s missing, they’re already deep into a pilot that isn’t going anywhere.
This isn’t about consultants cutting corners. It’s about how engagements are designed. Rapid execution is straightforward to scope and bill. A model trained and configured within weeks is easy to present as progress. The problem is that it often isn’t. What looks like delivery at the end of phase one tends to fall apart once it meets the reality of how a business actually runs.
AI doesn’t fail quickly. It stalls. Pilots stay live without scaling. Use cases get swapped out rather than built on. The organisation keeps investing but doesn’t move closer to embedding AI in how it actually operates. Up to 95% of enterprise AI pilots fail to deliver measurable business value. That’s not bad luck. It’s a pattern with a predictable cause – usually one of three things, or some combination of all of them.
Three causes of AI failure
Data infrastructure is the first. When customer, operational, and financial data sit in separate systems without a shared structure, no model can build a reliable picture of how a business operates. It fills the gaps with assumptions, and those assumptions show up in the output. Studies suggest nearly 70% of enterprise applications remain unconnected, which means the context that should inform a model’s decisions never reaches it. The result isn’t obviously wrong output. It’s output that quietly contradicts what client teams see in their day-to-day work, and once that happens, people stop trusting it. Once they stop trusting it, they stop using it.
Governance is the second. Without clear rules on how models are trained, monitored, and updated, outputs can’t be used in decisions that carry real risk. That affects accountability and compliance, but it also affects something more basic: the ability to explain an outcome when it’s questioned. In most regulated industries, that’s not optional. An AI system that produces results nobody can account for isn’t a useful system, whatever the demo looked like.
Organisational readiness is the third, and probably the most underestimated. Introducing AI changes how decisions get made and who’s responsible for them. If client teams haven’t been prepared to act on the output, and if there’s no clear ownership of where AI fits into existing processes, the output stays separate from the work it was built to support. It becomes a report nobody reads rather than a tool that changes anything.
None of this is especially complicated in theory. In practice, it requires honest scoping upfront and a willingness to do slower, less demonstrable work before the interesting parts begin. That’s exactly why it tends to get pushed to later phases. And those later phases have a habit of never quite arriving.
Structural change
Responsible AI consulting starts with these conditions, not after them. That means understanding how data is structured across a client’s systems before any model is configured. It means setting governance rules before outputs go anywhere near a real decision. It means preparing teams to act on what they’re given, rather than treating adoption as someone else’s problem once the technical work is done.
When those foundations are in place, things work differently. Outputs can be used without additional validation layers. Models can extend across the business rather than staying confined to the function they were piloted in. Client teams build confidence in what they’re seeing rather than working around it.
The engagements that get this right tend to start with harder questions: how is data structured across your systems, how do those systems interact, how are decisions currently made and by whom? They allocate time to the connective work before the model is built, not because it’s easy to sell, but because skipping it makes everything else unreliable.
As AI adoption continues to accelerate, the difference between engagements that produce outputs and those that deliver outcomes will get harder to ignore. AI doesn’t scale results. It scales whatever’s already built into the system. The organisations that understand that early, and the consultants who help them act on it, are the ones that will have something real to show for the investment.
Source link
#fix #consulting #engagement #model
