The number I can defend

People keep asking me for one number. They want me to say Maya AI resolves 80 percent of support at first touch. It would look good on a slide. It would round nicely.

I report it as a range. 70 to 80 percent. I do that on purpose, and I want to explain why, because the reasoning is the whole point.

What “first touch” actually means

Maya is the customer support platform I built at ESC Partners. It is two-sided. There is a customer-facing assistant that handles inbound across voice, chat, SMS, email, and mobile, in 26 languages. And there is an internal console for human agents, plugged into Oracle CCS, that loads the full context before a person picks up. Right now dozens of agents are running it in pilot at one US utility.

First touch means the very first time a customer reaches out. They send a message or make a call, and Maya is what answers. No human has looked at it yet. So “first-touch resolution” is asking a narrow question: of all the inbound that arrives cold, how much does Maya close on its own, without a human ever stepping in?

That word “resolution” is where it gets slippery. A lot of vendors count a conversation as resolved the moment the bot replies. I do not. A reply is not a resolution. If a customer asks the same thing again an hour later, or calls in angry the next day, that first touch did not resolve anything. It just delayed the work.

What counts as handled, and what doesn’t

So here is what I count as handled. The customer asked something. Maya answered it. The customer got what they needed and did not come back about the same issue. Bill questions. Outage status. Account changes. The repetitive volume that makes up most support. When Maya answers those and the thread closes for good, that counts.

Here is what does not count, even when it is tempting.

A handoff to a human is not a resolution for Maya. When Maya cannot do something, it routes the customer to an agent and loads the context into the console so the human starts informed instead of cold. That is good product. It saves the agent real time. But Maya did not resolve it. The human did. I do not get to claim that one.

An escalation is not a resolution. If a customer is upset, or the request is sensitive, or there is anything that should have a person on it, Maya hands off on purpose. That is the system working correctly. It is still not a number I get to put in the resolved column.

A reply with no real answer does not count. If Maya responds but the customer has to ask again, that first touch failed. I count the follow-up against it.

And then there are the edge cases. The weird ones. The phrasing nobody anticipated, the customer who asks three unrelated things in one message, the question that needs data Maya does not have access to. Those live in the gap between handled and not handled, and how you score them moves the number. That gap is real, and pretending it is not real is how you end up with a number you cannot defend.

Why a range is more honest than a point

When I add all of that up, honestly, the answer is not a single number. It is a band. On a clean week with familiar volume, it sits near the top. On a week with an outage event or a billing cycle that drives unusual questions, it sits lower. The truth lives in a range, so I report a range.

I could collapse it to 80 percent. I would just have to pick the friendly definition. Count replies as resolutions. Count handoffs as wins because Maya “touched” it. Drop the messy edge cases out of the denominator. Every one of those choices is defensible on its own, and stacked together they turn a real 70 to 80 into a clean 80 that does not survive a hard question.

I would rather be able to survive the hard question. If you ask me how I measure it, what counts as handled, and where the edge cases live, I have an answer for every part. That is what I mean when I say it is the number I can defend. The range is not me hedging. The range is the honest shape of the data.

Why honesty in metrics is the differentiator now

Here is the thing about AI products in 2026. The demo is no longer the hard part. Anyone can stand up something that looks like it works for five minutes. The hard part is the gap between the demo and the deployment, and that gap is exactly where inflated numbers hide.

Utilities are not buying a demo. They are putting Maya in front of real customers who have real billing problems and real outages. When the number on the slide does not match what the agents see on the floor, trust is gone, and you do not get it back. I have watched dozens of agents use this console with live customers. They know immediately whether a number is real.

So honesty is not a virtue I am performing. It is the most durable thing I can offer. A vendor who tells you 80 and delivers 72 has a churn problem. A vendor who tells you 70 to 80 and lands at 78 has a reference customer. Over a year, the honest number compounds and the inflated one collapses.

That is why I will not give you a bigger number than the one I can defend. Not because I am modest about what Maya does. Because the defensible number is worth more.