How Product Management Discipline Separates Lasting AI and Agent Adoption from Expensive Shelf-Ware
We’ve all read about the AI rollouts that go awry. Tools get purchased, training gets scheduled, an adoption campaign goes out, but within two months the usage curve flattens because nobody in the organization can answer a simple question:
What specific problem are we solving, and how will we know we solved it?
I’ve spent years leading teams from both an engineering and product management perspective, so I’ve seen from the trenches why this obvious question can get skipped. The urgency to "adopt AI" pushes companies straight into tool selection and training programs while the harder, slower work of defining which problems are actually worth solving never happens.
The Missing Discipline
A recent Harvard Business Review study by Amanda Pratt and Melissa Valentine examined AI adoption at a major tech company and surfaced a finding that should reframe how every operator thinks about this problem.
It was no surprise to me that the area most correlated with successful, sustained AI adoption turned out to be product management, not prompt engineering or technical fluency. The disciplines that mattered most were defining which problems are worth solving, designing structured experiments, and integrating solutions into the way work already happens.
These findings line up with what I've observed across dozens of AI and Agentic AI engagements. The companies where AI actually takes root are the ones that approach adoption with product discipline, starting with a specific workflow, identifying a measurable friction point, building a small test, and evaluating results before scaling anything.
Two Companies, Two Approaches
Consider the difference between two real patterns I see repeatedly in enterprise AI and Agentic AI work.
1) Company A purchases an AI platform, negotiates an enterprise license, builds a prompt library, and launches a change management campaign complete with lunch-and-learns, weekly tip emails, and a login dashboard to track "adoption." After three months, a handful of power users have integrated the tool into their workflows, and everyone else has moved on.
2) Company B takes a different path. Before selecting any tool, they run a structured problem-definition process across three business units. Each unit identifies its highest-friction workflow, documents the current state in detail, and defines what a measurable improvement would look like. Only then does the team evaluate which AI capabilities (if any) could address those specific problems. They run 30-day pilots with clear success criteria, and when two of the three pilots produce measurable gains, those two scale while the third gets killed early, saving months of wasted effort.
One of those pilots, for example, targeted a procurement approval workflow that averaged nine days from request to sign-off. The team mapped every handoff, identified two steps where AI-assisted document review could eliminate manual bottlenecks, and set a target of reducing cycle time to under four days. After the pilot, cycle time dropped to three and a half days. That result gave leadership concrete evidence to fund a broader rollout in procurement, and the specificity of the success made it easy to communicate across the organization.
Company B spent less money, took slightly longer to get started, and ended up with AI embedded in actual workflows producing actual results. Company A spent more, moved faster, and ended up with an expensive tool that sits mostly unused.
Why Problem-Definition Keeps Getting Skipped
The rise of AI has put immense pressure on companies to try to move fast. But the problem-definition process feels time consuming and slow. On the other hand, buying a tool and launching a training program feels like jumping quickly into action.
There's also a structural gap. Most organizations assign AI adoption to IT or to a newly created "AI team" that reports to the CTO. Those teams are good at evaluating technology. They're less practiced at the product management work of scoping problems, defining success metrics, and designing experiments within business workflows they don't own. The people closest to the workflows (operations leads, department managers, senior ICs) rarely get pulled into the problem-definition phase because the initiative is framed as a technology project, not a workflow improvement project.
Velocity without direction is just expensive motion. The organizations I work with that have the strongest AI adoption results are the ones that invested the first four to six weeks in problem definition and a Data Story / IP Moat audit before evaluating a single vendor. That initial patience created clarity that made everything downstream faster, from tool selection to pilot design to scaling decisions.
The Diagnostic Question
If you want to know whether your AI or Agentic AI adoption effort has legs, ask one question across every team that's supposed to be using AI. Can they answer, specifically, what problem they're solving and how they'll know if they've solved it?
If the answer is vague ("We're using AI to be more efficient") or circular ("We're adopting AI because we need to adopt AI"), the rollout is already in trouble. Clear problem statements are the leading indicator of whether AI adoption will stick or stall.
The companies that bring product management discipline to AI adoption, with defined problems, scoped experiments, and honest evaluation, end up with AI embedded in their actual operations. Everyone else ends up with a line item on the budget and a login dashboard nobody checks.
…
Nick Talwar is a CTO, ex-Microsoft, and a hands-on AI engineer who supports executives in navigating AI adoption. He shares insights on AI-first strategies to drive bottom-line impact.
→ Follow him on LinkedIn to catch his latest thoughts.
→ Subscribe to his free Substack for in-depth articles delivered straight to your inbox.
→ Watch the live session to see how leaders in highly regulated industries leverage AI to cut manual work and drive ROI.
Top comments (0)