Every contact center leader we talk to comes in with the same priorities: reduce handle time, deflect more calls, keep customers away from a human.

Those are the metrics the business judges them on. A function under constant cost scrutiny optimizes for the numbers that keep it funded.

That framing keeps most AI voice agents anchored to cost avoidance and out of the revenue conversation entirely.

It's the wrong scorecard. An inbound contact center is one of the few places a customer chooses to talk to you at the moment they're ready to buy, cancel, or walk. That makes it a revenue engine sitting inside a cost center. The shift that matters is treating it as one: measuring it on the dollars it moves, not the calls it deflects. Get the scorecard right and what you expect from your AI vendor changes with it.

Contact Centers Are Expensive. So Everyone Optimizes to Spend Less.

The average inbound call costs $7.16 to handle, per ContactBabel's 2025 US Contact Center Decision-Makers' Guide, a five-year high. Labor accounts for roughly 70% of that; the rest goes to infrastructure, QA, routing technology, and management overhead.

At significant volume, the contact center is a seven- or eight-figure line item for a mid-sized company and a nine-figure one at enterprise scale. The business treats it as fixed overhead. Finance knows the number. Operations manages around it.

How AI Voice Agents Got Trapped in the Deflection Frame

AI got sold as cost avoidance because it's the easiest ROI to model. Automate 30% of call volume, multiply by $7 per call, and you have a projected savings figure before the meeting ends. The math works without any attribution model or coordination with the CRO.

Vendors built their pitches around it. Contact center leaders bought on that basis. The KPIs in most deployments were designed around it.

Vendors set the bar low, and buyers have had no better alternative. Across the market, the same figures get quoted as good: early deployments run 20-40% containment, reaching 40-70% only once they mature. At the low end, 80% of calls still route to a human. With cost avoidance as the only target, there's no pressure to do better, so the CFO saw savings, the project got renewed, and the AI agent stayed mediocre.

(By the way, our deployments run well above that range.)

What Changes When AI Voice Agent Containment Gets Good Enough to Trust

Going from 20% to 65-70% containment is a capability change, not just a cost improvement. At 20%, the AI handles routine queries and routes everything else. At 65-70% on a high-value call type, it handles complex, conversational interactions well enough that customers complete their task without asking for a human. Those two agents have different business cases.

Plenty of inbound calls aren't sales. The ones that are (new orders, renewals, cancellations, upsells) carry the revenue, and a deflection-first deployment routes exactly those to a human or loses them to abandonment.

A major national consumer brand is what that looks like in practice. They ran a proof-of-concept with Simple AI and went from signed agreement to live in production in 10 days. Containment hit 50% in week one, 61% by the end of a three-week POC, and 70% on inbound sales, their highest-value call type. Their own in-house build had stalled at 20% after six months. Less than a year in, the program has saved $2.6M and cut queue abandonment from 9% to 2%, and at their December peak each point of abandonment is worth roughly $1M in recovered sales.

Human agents on that same sales line run a 13:1 revenue-to-cost ratio. The AI runs 19 to 22 to 1. That gap comes from two structural advantages no rep can match, however good they are.

Some of that is availability: a call that abandons in queue is a lost sale, and cutting abandonment from 9% to 2% recovers revenue you can put a number on. The rest is consistency: the AI runs the same play on call 50,000 as on call one, including during the holiday surge, when seasonal and temp staff handle most of the volume and upsell performance usually drops off.

A national home-services brand deployed Simple AI for outbound and generated six figures in two days of limited testing. Their operator focused less on the volume than on the range: the AI handled inbound leads, service calls, cancellations, and rescheduling without rigid scripting.

"Unlike some bots that follow a rigid, straight-line script, this bot can pivot based on the customer's responses. One bot can handle multiple call types, including interested inbound leads, service calls, cancellations, and rescheduling."

Agents built to hit 20% containment don't behave this way. The target shapes the design.

The Organizational Block Keeping This Math Off the Table

In most organizations, contact centers report into Operations or Customer Service, where the mandate is to control cost. Sales and the CRO own the revenue number. When an AI voice agent lifts revenue on inbound calls, that lift lands on someone else's scorecard, not the contact center's.

So the team running the AI optimizes for the one thing they're measured on, which is cost. Anything that carries revenue risk with no revenue upside gets cut from scope. The AI handles low-stakes calls; everything complex or high-value routes to a human.

Marketing makes it harder. One customer told us their marketing team was among the biggest internal adversaries of the rollout, and the reasoning held up: marketing is measured on CAC, they paid to generate those inbound leads, and they didn't want hard-won demand routed to an AI agent that might lose the sale or fail the upsell. A cost-first deployment proves that fear justified; an agent built and measured for revenue is what removes it.

The fix is governance. Operations, Sales, and Marketing need shared accountability for what happens on AI-handled calls. That conversation is harder than buying a platform, and most companies haven't had it. The ones that do end up with an advantage that compounds.

The KPIs Worth Adding to Your Contact Center AI Dashboard

If you're running or evaluating AI voice agents and only tracking containment rate and average handle time, you're measuring inputs. Revenue metrics to add:

• Revenue per AI-handled call vs. revenue per human-handled call

• Upsell rate across AI agents, trained human agents, and seasonal workers

• Conversion rate on AI-handled sales calls

• Revenue recovered from calls that previously abandoned in queue

Cost metrics that mean more alongside revenue:

• Cost-to-revenue ratio by call path

• Queue abandonment rate before and after AI deployment

• Seasonal hiring reduction and its effect on annual labor spend

Cost metrics matter, but running them alongside revenue metrics makes the full picture visible to the executives who need to act on it.

What Happens to the People Working the Phones

The deflection framing skips what happens to human agents when the deployment is genuinely effective.

When AI voice agents handle high-volume, lower-complexity calls, human agents get more complex work: calls that require judgment, situations where experience and empathy matter.

"The average live call now requires a higher skill level."

That's from the customer referenced above. First-year attrition at contact centers runs as high as 69-73%, and most churn happens before an agent is fully productive. Give agents work that requires judgment instead of script-reading, and tenure tends to improve.

Change the Scorecard, Change the Mindset

The deflection scorecard does more than undersell the contact center. It lets vendors win by clearing a low bar. When 20% containment counts as success, nobody has to build anything better, and plenty of vendors are happy to wrap a mediocre model in a dashboard and call it AI.

That changes the moment a contact center stops measuring itself as a cost to be minimized and starts tracking revenue per call, recovered abandons, and upsell rate. Goals that reflect the revenue moving through the phone lines raise the bar for the vendor too. An agent good enough to trust with a high-value sales call is a different product than one built to deflect, and you get it by demanding it.

The mindset shift comes first: treat the contact center as a revenue engine, hold the AI to that standard, and expect your vendor to clear it. The operators already working this way are pulling ahead, and the gap widens every quarter.

Cat Li is co-founder and CEO of Simple AI, which builds AI voice agents for enterprise contact centers. To see how the revenue math applies to your operation, book a demo at usesimple.ai.