Metrics Need a Partner
The Setup
I’ve been working against this ROI of AI problem one little piece at a time.
Suppliers are racing to cut costs because of the promises of AI and to be cost competitive. Buyers are squeezing harder on efficiency “because AI” going both ways, my condolences for the middle management layer faced with internal layoffs/reorg and less budgets for doing the same thing. Both sides think they’re being smart “because of AI”. But we’re racing toward the same cliff.
Suppliers lose margins. Buyers get worse outcomes dressed up as savings. The relationship between the two, the thing that used to make a supplier worth keeping, gets optimized out of the equation because nobody’s measuring for it.
I started thinking about this because I was asked to review a measurement model for a program where the default was a single-KPI scorecard. Hit the number, keep the contract. Miss the number, lose the fee. Clean. Simple. And I knew this is exactly the structure that turns you into a commodity.
My Approach + AI Role
I used AI to stress-test what happens to commercial relationships when you measure only what one side controls. I ran scenarios. What does a scorecard look like when the supplier aces every metric and the client is still unhappy? What does it look like when the output is technically flawless but strategically useless? You know… AI junk.
The answer I gravitated towards kept pointing me to the same direction. I’ve seen programs hit 98% on every operational target and still lose the renewal. Because 98% response time compliance doesn’t tell you whether the responses were worth reading. A single-party KPI measures throughput. Throughput is the thing AI replaces first. If your entire commercial relationship is built on proving you can hit a throughput target, you’ve priced yourself into the race to the bottom.
What Actually Happened
So I designed something different. Two metrics, intentionally set up to pull against each other.
One measures quality from the supplier’s side. Are the things you’re producing actually useful? Not volume. Not speed. Useful. The kind of useful where someone on the buyer’s side picks it up and does something with it.
The other measures responsiveness from the buyer’s side. When something useful lands, how fast does the organization act on it? Because the supplier can deliver in 24 hours, but if it sits in an inbox for two weeks, the velocity stays flat. That number only moves when both sides are working.
Neither metric alone tells you anything reliable.
Say your team delivers 20 recommendations a month. Your quality score shows 85% acceptance. Hooray! The client picks up and uses most of what you produce. Looks great on paper. But the average time from delivered to acted on is six weeks. The quality is there. And we’ve missed several important moments the recommendation would have impacted. The buyer’s organization can’t act fast enough to capture the value. Quarterly business KPIs were flat. That’s not a supplier failure. It’s a system problem. A single-KPI scorecard would never surface it because the supplier’s number looks pristine.
Now flip it. Acceptance rate drops to 30%, but when something does land, the client moves on it within 48 hours. Moved their first quarterly business KPIs by 2 points. The organization is responsive and fast, but two-thirds of what you’re producing isn’t useful enough to warrant that speed. Different problem. Same scorecard would miss it entirely.
The only way to read either scenario honestly is to hold both numbers next to each other. High quality with slow velocity means the work is good but the organization can’t absorb it. Fast velocity with low quality means the organization is eager but the inputs aren’t worth acting on.
You need both numbers to read the relationship. And reading the relationship, that’s the whole point.
The Real Insight
A single KPI makes you a vendor. Two metrics that can only be interpreted together make you a partner.
I keep thinking about client satisfaction scores, because they’re supposed to capture the relationship. But satisfaction surveys fail the same way. You can shower a client with freebies, extra stuff, and discounts. You say yes to everything. The score still tanks because the underlying work isn’t oriented toward a shared outcome. The client knows something isn’t working even when they can’t articulate what. You can’t generosity your way to trust. You can’t volume your way to true value.
The two-lever design forces a different conversation. Not “did you hit your number” but “are we both contributing to something that’s working?” The quarterly review stops being a performative ritual and starts being a signal-and-meaning-making moment. That conversation, the one where both sides are accountable to what the program actually produces, is the thing AI can’t prompt its way into.
Try This If…
Look at whatever scorecard or measurement framework you’re working with right now. Ask one question: does this only measure what I control?
If yes, you’re measuring throughput. Throughput is commoditizable.
Try adding one metric that requires both sides to perform. One number that stays flat unless the relationship is functioning. If you’re an agency, maybe it’s how often your strategic recommendations get picked up and acted on within a defined window. If you’re in-house, maybe it’s how quickly a vendor’s output translates into a decision your team actually makes. The specifics depend on your context. The principle is the same: build one measure that neither side can move alone.
It doesn’t need to be perfect in Year 1. Build it as a discovery step first. Let both sides learn what “good” looks like together. The scorecard matures as the trust matures because we’re pointed towards a common goal.
The measurement should follow the value. Not the other way around.
Systems Lens
This connects to something I’ve been writing about in my pricing series: we keep measuring what’s easy and calling it what’s valuable. Goodhart’s Law applied to commercial relationships. The moment you optimize for the metric instead of the thing the metric indicates, you’re in trouble.
AI is already widening this problem because of our frantic pace to “get ahead”. The transactional layer of every supplier relationship is getting cheaper and faster. If your measurement model only captures that layer, you’ve handed the buyer a reason to replace you. And the buyer has handed themselves a reason to get worse outcomes, because they’ve told the market (and their middle management) that cost and speed are the only things they value.
Both sides can do better at the same time. Efficiency and shared ownership of outcomes aren’t competing priorities. They’re two layers of the same relationship. Measure both, and the conversation changes. Measure only one, and the race continues.
I think we all sense it, but I don’t think either side sees the collision coming.