AI Doesn’t Have an Intelligence Problem. Enterprises Have a Decision Friction Problem

Introduction

For the last two years, most enterprise AI conversations have been about intelligence. Which model performs better. Which one has the longest context window. Which coding assistant writes cleaner code. These questions dominate boardroom discussions and technology roadmaps. 

They are also, mostly, the wrong questions. Most enterprises are not short on intelligence. They are short on decisions moving fast enough to use them. Every purchase request, product change, customer escalation, compliance approval, and engineering review must pass through several layers of people and process before anything happens. Each layer is individually reasonable. Together they create friction that slows an organisation down more than any technology gap ever could. 

That is the real reason so many AI initiatives fail to produce the returns leadership expects. McKinsey’s State of AI research found that 88% of organisations now use AI regularly in at least one business function, yet only around a third have scaled it beyond isolated pilots, and just 6% report significant enterprise-wide financial impact. Technology is not a constraint. The organisation is. 

What Is Decision Friction?

Decision friction is the accumulated delay created by approvals, handoffs, and coordination steps that individually seem reasonable but collectively slow an organisation down. It is not caused by any single bad process. It builds up from many small, defensible steps, each one adding a bit of wait time before a decision can actually be made. 

In practice, decision friction shows up as: 

• Waiting for approvals that could have been resolved faster with the right context available upfront 

• Searching for information that already exists somewhere in the organisation, just not where the person looking for it can find it 

• Following up on requests that stalled without anyone actively blocking them 

• Escalating routine issues that should not need escalation in the first place 

• Sitting in meetings whose only real purpose is handing work from one team to another 

Where the time actually goes

Ask people in most companies where their time goes, and the pattern above is what comes back, consistently, regardless of department or seniority. 

None of that creates value. It persists because organisations have grown more complex over time, and the processes built to manage that complexity have not kept pace. Herbert Simon, who won the Nobel Prize in economics for this exact idea, called it bounded rationality: people cannot process every option before deciding, so they lean on approvals and hierarchy to manage uncertainty. Those structures worked well for decades. Today they often delay themselves. 

The gap between AI adoption and AI value is not a model problem. It is decision friction that AI happens to make visible rather than create. 

AI is better at speeding up decisions than replacing people

Early enthusiasm about AI centred on replacing repetitive work. The more durable value turns out to be somewhere else: reducing decision friction, the accumulated drag of approvals and handoffs, so routine decisions happen faster without removing the person making the call. 

Take a customer support escalation. Without AI, it might pass through three or four teams before landing with the right person. An AI-assisted workflow can classify the request, pull the relevant documentation, surface similar past cases, and suggest next steps in seconds. The engineer still makes the call. The organisation just removes the unnecessary delay getting there. The same logic applies in procurement, finance, HR, manufacturing, and software delivery. AI does not replace expertise. It shortens the distance between a problem and the person equipped to solve it. 

What this looks like when it actually works

Picture a mid-market GCC handling change requests on a product platform. Under the old model, a change request lands in a queue, waits for triage, gets assigned to an engineer who has to read through prior documentation to understand context, gets flagged for a design review, and eventually reaches implementation days or weeks after it was first raised. None of the individual steps are unreasonable. The sum of them is the problem. 

An AI-assisted version of the same workflow looks different in a specific way: the triage, context retrieval, and initial impact assessment happen automatically at the moment the request is logged. The engineer who picks it up starts with a summary of relevant prior decisions and affected components rather than a blank page. The design review still happens, because it should, but it happens with the groundwork already done. The total decision friction in the process drops even though no single human decision was removed from the loop. 

That distinction matters because it is easy to build AI tooling that looks impressive in a demo and changes nothing about how long a request takes to resolve. The measure that matters is not whether AI touched the workflow. It is whether the workflow got faster without anyone cutting a corner they should not have cut. 

Why more AI agents creates a new coordination problem

A different kind of friction shows once companies deploy AI agents across functions rather than a single assistant. One agent handles customer service, another supports developers, a third reviews contracts, a fourth generates reports. Each performs well on its own. 

The challenge is getting them to work together. Individually capable systems that do not coordinate simply recreate the same silos AI was supposed to remove, just with software instead of people. Getting multi-agent systems to actually coordinate, rather than just coexist, is a genuinely hard architectural problem, and one that most organisations underestimate until they are already deep into an implementation. 

What changes first

None of this requires a company-wide AI strategy document before anything can start. It usually starts smaller: pick the one workflow where delay is most visible and most expensive, map where the actual friction sits, and design the narrowest AI-assisted intervention that removes it. Organisations that try to fix decision friction everywhere at once tend to end up with the same fragmented, uncoordinated mess described above, just with AI attached to each piece of it. 

What Pratiti does about this

Pratiti’s AI-assisted software engineering work sits closer to innovation consulting services than to a standard AI tooling engagement. It is built around reducing decision friction inside a real workflow, not adding another AI tool on top of an unchanged process. We work through where decisions are actually getting stuck in an engineering or GCC delivery workflow, then design the AI-assisted process around that specific source of friction. 

That distinction matters in practice. An AI code review tool that nobody has integrated into the review workflow does not speed anything up. It just becomes another tab. Our AI-assisted SDLC approach is built at the workflow level for exactly this reason, integrating AI assistance into planning, review, and testing rather than leaving it as a tool individual engineers may or may not use. 

For enterprises evaluating where AI can help, the useful question is rarely “where could we add AI.” It is “where is decision friction actually slowing work down.” That second question consistently surfaces the opportunities that generic automation efforts miss. 

Struggling to turn AI adoption into measurable delivery gains? Pratiti works with enterprises and GCCs in India to find out where decisions are getting stuck and build AI-assisted workflows around that specific bottleneck. Explore our AI-assisted engineering services →  or  talk to our team →

Frequently Asked Questions

Why do most enterprise AI initiatives fail to deliver expected value?

Most AI initiatives stall because the organisation around them has not changed, not because the underlying model is weak. McKinsey’s 2025-2026 State of AI research found 88% of organisations use AI regularly but only about a third have scaled it past isolated pilots, with just 6% reporting significant enterprise-wide value. The gap is workflow redesign, not model capability. 

What is decision friction in an enterprise context?

Decision friction is the accumulated delay created by approvals, handoffs, and coordination steps that individually seem reasonable but collectively slow an organisation down. It shows up waiting for sign-off, searching for information, and meetings whose only function is transferring work between teams. 

How does AI reduce decision friction without replacing people?

AI-assisted workflows classify requests, surface relevant context, and suggest next steps automatically, removing the manual search and handoff time that normally precedes a decision. The person still makes the final call. What changes is how quickly they have what they need to make it well.