ROI vs. Noise
The clarity that separates the two.
Last week I wrote about the mandate we are all living under. Use AI, move faster, cut the cost. And 95% of enterprise AI initiatives return nothing measurable. The question remains, what is the other 5% doing differently?
We all have the same models. The 5% have the clarity and specificity about the problem they are solving before spending a single token. A mandate to use AI is a direction. The clarity comes from explicitly defining the use case, the problem it solves, and how to measure success. A vague direction, without clarity of the problem and what success looks like, cannot produce a measurable ROI.
Here’s how we have been able to show return on investment:
Being clear about the use case and going deep on a problem: This pre-dates Gen AI. The use cases my team and I selected were specific problems we needed to solve. We got there by going deep on the problem before selecting the tool. That clarity applied to problems like accuracy of responses, the ability to search, translation of information, or intent classification. We worked to gain clarity from the ground up.
Making Build vs. Buy decisions: Early on, almost all vendors were solving the problem of summarizing a chat transcript. When the vendor market is flooded with solutions, it is essential for us to make build vs. buy decisions based on our unique context, compliance, and regulatory understanding. We could solve for what’s core to us, but didn’t make sense to rebuild something that we could buy off the shelf.
Aiming for boring use cases: AI output can be demo-friendly, pretty, and polished. The real returns are less glamorous in back-office automations, cutting the cost of manual efforts. The ROI comes from understanding where actual value is, without the flashy scenarios.
Simplifying the workflows: We chose to strip out the bells and whistles from the solutions, simplify the workflows, and go to the first principles of the problem we were solving. With AI, the solutions can seem limitless, and an appropriate tool choice for simple workflows can help unblock the teams.
Notice the patterns
Every one of these is really the same move. Know the problem before selecting the tool or solution. Be clear about the pain, the user, and what a good outcome looks like. Then AI can help get us there faster. If you think about it, none of this is new. It is the foundation of good product work, the part we have always known and sometimes skip when the pressure is on to just move at speed or scale.
Which brings me back to where I started. The AI mandates hand us the tool and tell us to go fast. It never says what for. The clarity is the piece we bring that separates the returns from the noise.
So the next time a mandate tells you to use AI, a more useful question might be ‘use it for what?’