AI Won't Fix a Broken Process — It Will Scale It

TL;DR

Every efficiency lever — outsourcing, shared services, offshoring, and now AI — gets deployed the same flawed way: before the underlying work is examined. The tool then embeds the problem instead of solving it. With outsourcing, there was at least a vendor to renegotiate with. With AI, the only party left to negotiate with is the organization that made the decision.

A pattern that keeps repeating

There’s a pattern that repeats with every new efficiency lever an organization adopts — outsourcing, shared services, offshoring, and now AI. The tool gets deployed before the underlying work is actually examined. And because the tool is fast and confident, it doesn’t fix the flaw in the process. It scales it.

I saw this clearly during the outsourcing of roughly 700 roles at a GCC bank. In a steering committee meeting, someone asked a simple but uncomfortable question: why hand the vendor the chance to capture efficiency gains the organization already knew existed, instead of optimizing the process first and keeping that value in-house? The honest answer was that speed mattered more than optimization. The organization chose to move fast. The vendor got an easy contract and reaped the efficiency gains the bank had already identified but left on the table.

That decision had a cost. The choice to move fast without doing the foundational work does not eliminate the cost of fixing the process. It just transfers it — in that case, to the vendor’s margin instead of the bank’s bottom line.

With AI, there is no one left to transfer the cost to

The same logic now applies to AI, except there’s an important difference: there is no vendor to transfer the cost to anymore. There is only the process — but now running faster than before, at greater scale, producing the same flawed output with considerably more confidence.

A junior analyst handed a broken process will eventually push back. They’ll notice the exception that doesn’t fit, question the approval that doesn’t make sense, or flag the report built on bad data. An AI agent will not. It executes completely, repeatedly, and at volume, with no instinct that something might be wrong.

Across the GCC, I am seeing organizations connect AI to uncleaned data, build automated reporting on flawed metrics, and deploy automation into processes that grew organically rather than by design. The AI does exactly what it’s told. The problem is what it’s told to do.

The question that should come before automation

The right question before any automation initiative is not “how do we make this faster?” It’s “if this process were executed perfectly, would it produce the right outcome?”

If the answer is no, automating it first only gets you to the wrong outcome faster.

The programs that navigate this well start by auditing the work itself — not the process diagram, which often describes how the process is supposed to work rather than how it actually runs. They look for the manual steps that exist only because an integration was never built, the approvals given without real review, and the handoffs that create delay without adding judgment. Only once that picture is clear do they ask what AI can actually do.

This sequencing is slower at the front end. It is considerably faster, cheaper, and better for customers, at every stage that follows.

What the outsourcing wave should have taught us

The outsourcing wave left plenty of evidence behind, and most organizations didn’t learn from it. The AI wave moves faster, goes deeper into the organization, and comes with higher expectations attached. With outsourcing, an organization that moved too fast could still negotiate with the vendor after the fact — adjust the contract, renegotiate scope, push back on performance. With AI, the only party left to negotiate with is the organization that made the decision to automate before it understood the work.

That’s a harder conversation to have internally, which is exactly why it’s worth having before the rollout, not after.

From the field: I led process optimization ahead of a large-scale outsourcing initiative for a Gulf retail bank — the discipline of fixing the work before handing it off is the same discipline that has to come before any AI deployment. Read more in Services: Transformation & Target Operating Model Design.