On a typical programme the AI finishes up to 40% of calls by month 3, about 50% by month 4 and 70 to 80% by month 6, and stays in that band at maturity. Month one is lower, often well under a quarter, because the AI is learning your scripts, your objections and your edge cases on live calls while a full human bench takes the rest. A 90% figure in week one is a bot's containment number, and the floor's number is a different thing.
The curve
The curve above is the shape we put in writing. It is drawn from programmes on our platform and labelled as such; your line starts at your number. The marks are the ones we commit to in conversation: up to 40% at month 3, 50% at month 4, 70 to 80% at month 6.
Why month one is low
The AI arrives knowing the language and the mechanics of a call. It does not arrive knowing your book: which objections come up on a two-wheeler renewal in Marathi, what a borrower in Indore says when the salary is late, which three questions your customers ask about a delivery. It learns those on live calls, so the first weeks hand over more. The human bench is sized for that, and the baseline is agreed before the first call so that nothing you measure drops while the share climbs.
What the people do meanwhile
They take the handovers, with the summary on screen, which is the job they will keep doing at month six on the 20 to 30% that is theirs. And they review the first week's transcripts with us, which is how the share moves: the scripts change, the do-not-say list grows, the handover triggers tighten. A named owner on your side who can approve those changes is the single biggest factor in how fast the curve climbs.
What it means for cost
Month one costs more than month six, because the people are all there and the AI share is small. That is why the pilot runs at the pilot rate with no commitment, and why the rate steps down with committed volume once the share has climbed. Compare cost per outcome at months 1, 3 and 6, since day one tells you little; the ROI calculator lets you set the share by month and see the line.
Questions people ask
What percentage of calls can AI handle in a contact centre?
On a typical Human + AI programme, up to 40% by month 3, 50% by month 4 and 70 to 80% by month 6, with a person on the same queue for the rest.
Why is the AI share low in the first month?
The AI learns your scripts, objections and edge cases on live calls. The human bench takes the rest while the baseline protects your KPIs.
Does the share ever reach 100%?
No. The remaining 20 to 40% are the calls that need judgement: disputes, hardship, objections, customers who want to close. Those go to a person.