Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Why teams need to stop bolting AI onto broken processes

And why re-architecting must come first.

[Photo: Getty Images]

Profile photo of Dutta Satadip

Chief Business Officer · Pebl

When an artificial intelligence initiative fails to deliver measurable returns, the typical post-mortem points to bad vendor selection, weak models, or poor data hygiene. But after decades leading business operations through major technology shifts, I’ve found the real culprit is almost always far simpler: Companies are trying to automate a process that was already fundamentally broken.

Right now, C-suites across every industry face immense board pressure to answer one question: “What is your AI strategy?” Yet too often, that pressure turns into a hasty procurement exercise. Leaders buy software, bolt it onto existing workflows, and expect magic. But if you bolt intelligence onto chaos, you don’t get transformation—you just get automated chaos.

THE VIEW FROM THE SLIDES

As executives, we naturally live in a world of slides and high-level abstraction. On a PowerPoint deck, an operational process looks remarkably clean: a neat progression from step A to step B to step C. It seems entirely reasonable to say, “Let’s build an AI agent to reconcile billing invoices across systems.”

The problem is that slides conceal operational reality. Because they flatten complex workflows, leaders often miss where the real friction lives. That’s why, before handing any process to an AI agent, I’ve learned to look beneath the deck for three simple diagnostic signals:

  • Multi-format fragmentation: The process relies on disparate data types scattered across incompatible software systems and static spreadsheets.
  • Slack thread dependency: The workflow only moves forward through constant, manual pinging, side-chats, and offline clarifications.
  • Boundary-crossing handoffs: If you ask who touches a single task and hear four different names—Sally talks to James, who talks to Lily, who talks to Joe—you aren’t looking at a process. You’re looking at an organizational game of telephone tag.

When a workflow fails these basic tests, handing it to an AI agent won’t solve the friction. It will only accelerate the confusion.

RE-ARCHITECT STRUCTURE BEFORE WORKFLOW

When executives realize operations are messy, the default instinct is to pause and launch a multi-quarter data governance project. However, in my experience, today’s models are surprisingly capable of working through loose, unstructured data. What AI cannot fix for you is a broken organizational structure.

If you change software tools without changing team structure, people instinctively slide back into doing things the exact same way under new titles. During periods of growth, companies naturally accumulate duplicate and adjacent teams with overlapping remits. Real transformation requires eliminating and compressing those redundant structures before deploying technology.

Take the traditional go-to-market engine. For years, Software-as-a-Service companies maintained separate teams for sales development representatives and business development representatives. When lead volume was overflowing, that division made sense. But maintaining two distinct roles now mostly creates artificial handoffs and extra cost.

When I evaluated this within my own team’s operations, we eliminated that artificial boundary and combined inbound and outbound responsibilities into a single role. That structural compression forced a complete re-orchestration of how work flowed. Only then did we equip that unified team with AI tools to handle research, account prep, and follow-ups.

By eliminating that structural redundancy, we reduced our cost of sales. Instead of banking those savings, we redeployed the budget directly into hiring account executives to enter new growth markets—turning operational efficiency into top-line revenue growth.

THE GOVERNANCE AND CONTEXT GAP

Fixing organizational structure clears the runway, but even when you hand a clean process to AI, new failure modes emerge that standard productivity metrics miss.

In fact, my team and I experienced this firsthand when we built an internal AI platform to help support representatives resolve tickets. On paper, the rollout was a triumph. Onboarding time for new representatives dropped from three months to three weeks, and speed metrics looked phenomenal.

Then our customer satisfaction scores started to plummet. When we investigated, we discovered a subtle quality trap. While experienced representatives knew how to feed rich context into the prompt window, newer representatives were providing thin prompts. They got fast answers, but incomplete or incorrect ones—and lacked the domain experience to spot the difference. The dashboard reported green for adoption, but customer experience was degrading underneath.

AI adoption isn’t a “set-and-forget” information technology project. It requires an active human quality assurance layer that inspects prompt context and input quality, not output speed alone. It also requires recognizing that adoption friction today is rarely a skill issue—it’s a job security issue. When employees fear that technology threatens their roles, they may engineer ways for the tool to fail. That suppressed adoption inevitably erodes the returns and outcomes of your entire AI investment. So you must address that human fear directly.

IN DEFENSE OF “GO SLOW TO GO FAST”

Re-architecting an organization before adding intelligence requires uncomfortable discipline. When I’ve led these turnarounds, clear financial returns lagged in the first six months. Improved operational evidence, however, appeared early in the form of fewer handoffs, lower rework, and faster cycle times. That’s because the re-architecting resolved the structural friction identified at the start.

To defend that initial lag to a board hungry for quick wins, you must manage expectations upfront. Show them functional prototypes and proofs-of-concept so they see what the technology can do. But hold the line on doing the hard structural work required to capture durable value.

At its core, AI shouldn’t be treated as a quick cost-cutting lever. It is an engine for changing how work gets done—freeing teams from low-leverage coordination so they can drive high-leverage judgment. When you re-architect the organization first, you don’t just save operational hours. You build a company capable of finding new life and real growth on the other side.

Dutta Satadip is chief business officer at Pebl.