Content by Orvera AI – a 2026 A.I. Awards Success Suite entrant.

By Ali Merchant, CEO at Orvera AI. Orvera AI  are finalists for ‘Best Agentic AI Solution‘ at the 2026 A.I. Awards.

Every customer experience operation runs on a few agreed numbers. A handle time assumption inside the workforce plan. A resolution threshold inside the quality framework. A service level in the outsourcing agreement, with a penalty attached. A bonus gate paying an agent for hitting the target their supervisor is graded on. None of them were set carelessly. All of them were right when signed. We set thresholds like these. We staffed against them.

Then AI took the front of the queue. Every one of those numbers now describes a different population of work.

What we at Orvera call the stranded baseline is a handle time or resolution target priced against a contact mix that no longer exists, still sitting inside a contract term with years left to run. These are not metrics that drifted. They are prices. A metric that drifts earns a note in the monthly review. A price that no longer prices its object gets renegotiated.

The appetite is there. HFS Research’s Horizons: Customer Experience Services 2026, reported on August 26, 2026, has 76 percent of enterprise clients open to outcome-based or gainshare commercial models and only 5 percent who have adopted them. Fifteen times as many want it reopened as have reopened it.

Why the mix moved, and who already said so

The mix did not move because of agentic AI. It has been moving for two decades. Contact Center Pipeline’s August 14, 2025 write-up of ContactBabel’s 17th annual US Contact Center Decision-Makers’ Guide has the typical service call now running over seven minutes, 38 percent longer than when tracking began in 2012. The same page gives the reason. Organizations allowed call times to rise as customer experience mattered more, and self-service took a greater proportion of the easier short calls. That was written before an agentic system touched a queue, which is why it holds now.

We are not claiming that observation. Anyone who does is one search away from being corrected. TTEC Digital wrote up the end of the old rule that lower handle time is always better. ACXPA’s Australian benchmarking data carries the same mechanism, with handling time trending up industry-wide as automation takes the simpler contacts. Several other firms hold the same ground.

Here is the half that is genuinely ours. None of that work walks from the metric into the contract. It goes as far as the dashboard and stops. The money sits downstream of the dashboard. That gap is where the stranded baseline lives.

The arithmetic that makes it visible

This is the arithmetic we run on any operation we take over, and it takes about ten minutes.

Start with a published anchor. ACXPA puts 537 seconds of average handling time across the Australian contact center industry, on self-reported benchmarking data. Call it nine minutes and run illustrative numbers against it.

Three figures off last month’s reports drive the whole thing. Contacts that reached a person. Contacts the AI agent finished on its own. The date the handle time target was last signed. At Orvera we pull those three before anything else. Subtract the second from the first and you have the population your average now describes. Set that against the third and you have how long it has been describing something else.

Take an illustrative operation where 100,000 contacts a month reach a person. Three illustrative bands. 40,000 simple at four and a half minutes, 40,000 standard at nine and a half, 20,000 complex at seventeen. Blended, an illustrative 9.0 minutes.

In the same illustrative month, the AI agent finishes 28,000 of the simple contacts and another 8,000 standard ones that used to escalate. That is 36,000 conversations ending without a person. Take 36,000 off 100,000 and what is left is the population a human still answers. The minimum volume commitment was written against the number you started with.

What the reweighting actually does

Now hold every per-band handle time constant. Nobody works slower. Complex work goes from a fifth of the human queue to nearly a third with no extra contact arriving. The illustrative blended average moves from 9.0 minutes to 10.9.

Resolution reweights the same way. At illustrative first contact resolution rates of 92 percent simple, 76 standard and 55 complex, none of them moving, the illustrative blended figure falls from 78.2 percent to 72.4 percent.

Minutes fall and headcount follows. The illustrative plan staffs 115 agents before and 90 after. A plan built off the containment rate alone cuts deeper and comes up short.

Here is the pair worth writing down. Containment in this illustrative operation reads 36 percent. The human workload fell 22 percent. Fourteen points sit between the number reported upward and the number the operation is run on.

Where the stranded baseline costs money

We have sat on both sides of an outsourcing table, and the argument is never really about the metric. It is about which party pays for it.

Three instruments carry the damage, one threshold each.

The bonus gate fails fastest. Set the gate at an illustrative nine and a half minutes. A population averaging 10.9 clears nothing. The monthly review then reads as a productivity collapse that never happened.

The supervisor scorecard fails next, and quietly. A team lead is graded on the blended 10.9 minutes, a figure the mix produced and no coaching conversation can move. ACXPA’s 2026 Australian report puts the consequence in one sentence: “When leaders cannot distinguish between a technology failure and a performance issue, agents risk being coached for outcomes that were never within their control.”

The outsourcing service level agreement fails largest, because a penalty is attached to it. Put an illustrative first contact resolution floor of 75 percent in the paper. The reweighted figure above is 72.4 percent. Set the two side by side. Nothing in the operation got worse.

The pricing basis decides which side pays

Then there is the pricing basis, and it decides who pays. On a per-contact rate card the provider carries the whole difference between 9.0 and 10.9 minutes, because that price was set against a shorter contact. On a per-productive-hour rate card the buyer carries it, because the hours are real and the contacts behind them are fewer. Same operation, opposite party. That choice was made years before it mattered.

Orvera came out of a contact center operation rather than a software company, which is why we read these as commercial instruments first. The mix moved. The money keyed to it did not, and every threshold still sitting in that paper is the stranded baseline.

Dating a baseline instead of defending it

The fix is not complicated and it is not quick. Re-derive the thresholds against the work that reaches a human today, band by band, the way they were derived the first time. Then record the mix beside the number. A target with a stated contact mix behind it can be reopened by arithmetic. A target without one can only be reopened by argument, and argument is where the relationship gets spent.

Then write the trigger into new paper. A defined move in the containment rate reopens the affected thresholds by agreement, on a stated schedule, by a method both sides wrote down in advance. That turns a renegotiation into a calculation.

Both sides read a reopened baseline as an admission. The provider hears underperformance. The buyer hears a walk-back of a booked saving. Both readings are reasonable and both are wrong. The baseline was not missed. It was retired, by a change both sides asked for and both sides delivered.

Orvera expects the first of these to reopen in regulated inbound voice, inside the 2027 renewal cycle. Not outbound, where the mix barely moved. If that cycle closes with regulated voice baselines untouched, we were wrong and the argument goes with it.

None of that has to be settled to start. Open the workforce plan. Find the date the threshold was signed. Write the contact mix next to it. A number carrying a date and a mix is one two parties can reopen without either of them losing. That is how the stranded baseline stops being everyone’s problem and nobody’s job.