I recently spoke with the head of a contact center at a very large company that’s been around since before mobile phones.
When I asked what his key performance indicators (KPIs) for the year were, the answer was a roundabout “none.” This executive didn’t track any metric consistently on his contact center spend. A quick scan of online reviews of his company made me believe him.
Being unorganized breaks when scale and complexity rise or when the operations go through a platform shift. And we’re in the middle of just that.
Voice AI: The Platform Shift
Voice AI is triggering an enormous operating model change inside both existing and new contact centers alike:
- Established teams are redesigning workflows.
- In some organizations, new stakeholders are now accountable for customer conversations (Product, Engineering, Ops, Finance).
- Day-to-day decisions are happening faster and at higher volume.
The bar for measurement is rising because small failures can now scale instantly. In this environment, intuition isn’t enough. You need a measurable operating system.
Pick one KPI you’d bet your job on. Don't worry, you can change it as your priorities shift. But you need focus now.
There’s a saying in Silicon Valley: startups = growth. In other words, startups are containers around growth.
Similarly, I think, contact centers are containers around customer success. We should borrow operating principles from Silicon Valley and apply them to contact centers, especially during this voice AI platform shift.
Pick one metric that defines winning, add a few guardrails that keep you honest, and run weekly goals like the best AI teams to drive compounding improvement. I will now explore these in detail.
Set a Primary KPI
A. Redefine KPIs
KPIs are a small set of quantitative metrics that tell you how “healthy” your operation is. The “health” part matters because it’s easy to confuse activity with progress.
KPIs force you to confront whether performance is improving or if you’re just keeping busy.
Setting KPIs is important for two reasons:
1. Obtaining objective truth
Numbers don’t lie: assuming you’re measuring the right thing(s). They keep you realistic about where you are.
2. Creating a feedback loop and prioritization
These tell you whether changes you make (policy tweaks, routing, training, tooling, voice AI iterations) are working and what to do next week. If you set them wrong, you can steer into circles.
B. Know what makes a primary KPI “good”
A good primary KPI should do these two things:
1. Quantify value delivered (to customers and to the business).
2. Be a usable feedback mechanism (moves frequently enough that you can act on it).
If a metric is easy to move but doesn’t represent value, it’s a vanity metric.
C. Know what are bad KPIs (and why they fail)
Before you pick your KPI “North Star,” it helps to see common pitfalls, especially during voice AI rollouts, such as these.
1. “% automated” or “# of AI calls handled”
Why it’s tempting: it’s easy to measure and looks like progress.
Why it’s bad: it measures activity, not outcomes. You can increase automation while customer success declines.
How it breaks: teams over-optimize containment; escalations become more frustrated; repeat contacts rise.
2. CSAT/NPS as the steering wheel
Why it’s tempting: executives recognize it.
Why it’s bad: it’s lagging, noisy, and biased. It’s great as a check engine light, not as a steering wheel.
How it breaks: teams argue about survey methodology instead of fixing the underlying operation.
Rule of thumb: if the metric can go up while customers get worse outcomes, it cannot be your primary KPI.
D. Select better KPIs
Pick one KPI you’d bet your job on. Don’t worry, you can change it as your priorities shift. But you need focus now. Too many KPIs cause fatigue, debates, and eventually non-use of the system.
Your primary KPI should:
- Move often (weekly is ideal).
- Tie directly to your business outcome.
- Capture what success looks like to you for the near future.
If you pick right, it shouldn’t feel reductive; it should feel focused.
Install Guardrails
Once you’ve chosen a primary KPI, pick two-three secondary health metrics. These are your operational guardrails: metrics that prevent you from “gaming” the primary number. Think of them as your contact center version of product “retention” and “quality” metrics.
Here is an example of a strong default set:
1. Repeat Contact Rate (7–14 days)
If this rises, your “resolution” is probably brittle.
2. Handoff Success Rate (a.k.a. escalation friction)
Measure whether escalations land cleanly: the human receives full context, the customer doesn’t have to repeat themselves, and time-to-first-human stays within your SLA.
3. Critical Defect Rate (compliance errors/over-refunding/etc.)
For refunds, cancellations, identity verification, regulated disclosures, or anything that can create liability.
Create an Operating Loop
A. Set weekly goals
There’s another Silicon Valley saying: do things that don’t scale.
In contact centers, that might look like:
- Listening to 20 calls yourself this week.
- Spending an hour rewriting the escalation policy yourself for the top two intents.
These aren’t long, glamorous projects with names, but they move the needle.
B. Aim for 1% - 2% weekly improvement
Set weekly goals for your primary KPI and target 1%-2% improvements. Those improvements compound (see CHART).
For example, a sustained 2% weekly improvement compounds dramatically over a year. Compare this to a large capstone project: an academic term that in business means a long, multi-team, big corporate effort that absorbs a lot of time: which doesn’t show any meaningful KPI movement until late, if at all.
That means you end the year at about 280% of baseline, without relying on a single giant capstone project. A small team (or even one operator) can often find a 2% win this week.
C. Draw the loop
1. Make the target visible
Draw a forward-looking chart for the next ~12 weeks. Put it somewhere the team sees weekly. Update it every week. Airbnb founders are famous for drawing it on the bathroom mirror of their offices, but you don’t have to go that far.
2. Set weekly prioritization
Every week, list the changes that could move the primary metric. Pick the one or two bets that are most likely to hit the weekly target.
3. Ship, measure, learn
Voice AI makes iteration fast if you have instrumentation in place to analyze calls and outcomes (intent tags, outcomes, recontacts, escalation reasons, and failure clusters).
Voice AI is bringing about not just a staffing change but a new operating model.
Early course correction and six months of focused engagement can pay dividends for years. You’ll build operational muscle for all members in your organization. You should also review your primary and secondary KPIs monthly to make sure they reflect the stage of business you’re in now.
When You Miss...
A. Constantly address the limiting factor
Missing a week (or two) is fine if you know why. Use “5 Whys” until you reach something actionable. 5 Whys is a simple root-cause method where you ask: “Why did this happen?” repeatedly (usually 5 times) until you get from the symptom to a specific, fixable cause. For example:
Problem: First Contact Success dropped from 62% to 54% this week.
1. Why did First Contact Success drop?
Because more calls were escalated to humans mid-flow.
2. Why were more calls escalated?
Because the voice AI couldn’t complete identity verification reliably.
3. Why couldn’t it complete identity verification?
Because the verification timed out so the agent started failing.
4. Why did the API time out?
Because traffic spikes caused rate-limiting and higher latency.
5. Why did rate limiting hurt us so much?
Because we had no retry mechanism and no fallback path (e.g., SMS verification).
Early course correction and six months of focused engagement can pay dividends for years.
So, the actionable fix is to ask the engineering team to add retries and a fallback verification path. Having proper instrumentation and call analytics in place helps you diagnose issues at startup speeds.
B. Understand if it is a metric problem versus an execution problem
When a KPI doesn’t move, it’s usually one of two things:
1. Metric problem: the definition is wrong or doesn’t reflect value.
2. Execution problem: you’re measuring correctly but not addressing the constraint.
It’s important to know what to fix.
C. Realize that instrumentation matters more in the voice AI era
Call analytics are no longer optional. Modern large language model (LLM)-based systems make it feasible to:
- Auto-tag intents and outcomes.
- Cluster failure patterns.
- Quantify policy violations.
- Surface the 20 calls that best explain why KPIs moved.
Finally...
Voice AI is bringing about not just a staffing change but a new operating model. The teams that win will treat their contact center like a modern growth organization:
- One metric that defines winning.
- A few guardrails that keep optimization honest.
- Weekly targets and a repeatable operating loop.
- Fast diagnosis powered by call analytics.
We can take lessons from the Silicon Valley playbook to improve our organizations and get compounding gains in customer success.