AI Implementation for Enterprise: Why Generic Consulting Fails and What to Do Instead

Generic AI consulting has a comfortable business model. It sells a strategy, presents a deck, and leaves before the hard part begins.

That hard part is where enterprise AI actually lives. The deck is not the product. The working system is, and most engagements never ship one.

This piece makes an argument, not a survey. Enterprise AI is an integration and governance problem, and generic consulting is built to solve neither. Here is the case, and the model that works instead.

The deck is the product, and that is the problem

Generic consulting optimises for the deliverable that is easiest to sell. A strategy document is low-risk, high-margin, and finished on a fixed date. A working system is none of those things. It is messy, it touches systems nobody fully understands, and it is never quite done. So the incentive quietly bends toward the artifact that closes the engagement cleanly, rather than the one that changes how the business runs.

The pattern is familiar to anyone who has sat through it. A large organisation commissions an AI strategy. Twelve weeks later it has a polished assessment, a prioritised use-case matrix, and a maturity model with the company plotted neatly on it. The slides are excellent. Nobody disputes a word. And six months on, nothing is in production, because the deck was the end of the engagement rather than the start of a build.

It shows in the outcomes. MIT found that ninety-five percent of generative AI pilots deliver no measurable impact on the bottom line, and the cause is rarely the model. It is what happens, or fails to happen, after the strategy changes hands.

What you get versus what enterprise AI needs

Set the two columns side by side and the mismatch is hard to miss.

The rows are not equivalent in effort, and that is the whole point. A model recommendation is an afternoon of judgement. A workflow redesigned around that model is months of work alongside the people who do the job today. Generic consulting prices and delivers the left column, then leaves the right column as your problem. Enterprise value lives almost entirely on the right.

Why the failures are structural, not accidental

Three patterns recur across stalled enterprise programs. They are baked into the model, not the team that runs it.

• The engagement ends at the roadmap, so you are left owning a plan while someone else owns the delivery.
• It automates the process you already have, when the value was in redesigning that process in the first place.
• Governance arrives only after an incident forces the conversation, which is the most expensive moment to start it.

Notice what these have in common. Each one defers the expensive, uncertain work and front-loads the work that is easy to invoice. That is not incompetence. It is the model behaving exactly as designed. The trouble is that the deferred work is the work that produces value, so a program built this way can spend its entire budget and still have nothing running.

What redesigning a workflow actually means

Workflow redesign is the phrase that gets nodded at and then ignored, so it is worth making concrete. Take an insurer’s claims process. The generic move is to add a model that reads claim documents faster and leave every downstream step exactly as it was. The output is a slightly quicker version of the old process, and the gains are marginal.

The redesign move is different. It asks what the process would look like if the model had existed from the start. Routine claims clear automatically above a confidence threshold. Adjusters stop reviewing everything and start reviewing only the exceptions the model flags. The handoffs, the queues, and the roles all change shape. That is where the step-change in cost and speed comes from, and it is precisely the part a strategy deck cannot do for you.

What to do instead

The counter-model starts where generic consulting stops. McKinsey’s high performers share one habit above all others. They redesign workflows rather than bolt models onto the ones they already run. That single move correlates more strongly with bottom-line impact than anything else tested.

In practice it comes down to four commitments, and each one is harder to sell than a slide.

• Buy systems, not slideware, and insist on something that actually runs in production.
• Redesign one high-value workflow end to end before scaling across functions.
• Build governance into the build itself, with audit trails and human-in-the-loop rules from day one.
• Tie the engagement to a measurable business outcome, not a count of deliverables or billed hours.

None of these are exotic. They are simply harder to sell than a deck, because each one commits the partner to an outcome they cannot guarantee on a slide. That is the trade you are making. You give up the comfort of a fixed-scope document for the value of a system that actually runs.

Where autonomous components are involved, the integration and the governance are the build, not the trim. That is the whole premise behind our agentic AI integration work, where how the system behaves under failure matters more than how it performs in a demo.

The enterprise difference

Scale does not change the fundamentals. It changes the cost of getting them wrong.

The wider picture of scope and sequencing sits in our overview of AI implementation services. What sets the enterprise apart is consequence. A stalled integration or an ungoverned model is not a contained nuisance at this scale. It is a board-level problem with a board-level price.

Generic consulting is comfortable because it asks for little and risks less. Enterprise AI rewards the opposite posture entirely: scoped delivery, redesigned work, and a partner who stays accountable for systems that run.