A voice bot is measured on the calls it kept away from a person. We are measured on the ones that ended in money.
That single difference explains everything else on this page. If your target is containment, a good voice bot is a reasonable buy. If your target is recovery, renewal or retention, containment is the wrong number to chase, because the calls that decide money are exactly the ones a bot hands back.
Deflect the predictable query, keep it off the floor, and report the containment percentage upward. Done well, that is real value on simple traffic.
Resolve what can be resolved, then put the hard one in front of a person with everything they need, and take responsibility for the outcome either way.
Seven questions worth asking either of us.
| What you are comparing | A voice bot | Oriserve |
|---|---|---|
| The number it optimises | Containment. How many calls never reached a human. | Outcome. How many calls ended in money recovered, a policy renewed, a customer kept. |
| What happens at the limit | The caller is told the query cannot be handled, and either drops or restarts in a queue. | The conversation transfers to one of your agents with the full context, mid-call, without the customer repeating themselves. |
| Conversation shape | Menu logic wearing a voice. It follows the tree it was given. | It listens, reasons against your system of record, answers, and takes an action on the account. |
| When the caller switches language | Usually breaks, or forces a language choice at the start and holds you to it. | Handles mid-sentence switching, which is how people actually speak on an Indian line. |
| What your QA team sees | A sample, scored by hand, weeks later. | Every call recorded, transcribed and scored, so the audit answer is the whole book. |
| How you are billed | Per minute or per session, whatever the call achieved. | Attached to the outcome, with a baseline agreed before we start. |
| What breaks it | Anything the tree did not anticipate. | The same things that break a human agent: bad data, an unreachable number, a customer who needs a decision only your team can authorise. |
Containment and outcome are not the same number.
The bot answered, the customer did not get what they needed, nobody was called back, and the payment did not happen. In the containment column that is a success. In the recovery column it is an account that has aged another week. This is the single most expensive measurement error in the category, and it survives because containment is easy to report and outcome is not.
Push automation past the point where the conversation needs judgement and quality falls off a cliff, because the remaining calls are the exceptions, the disputes and the negotiations. We stopped trying to beat that plateau and built for it instead: AI runs the volume, a person closes what needs closing, and the handoff between them is treated as a product rather than a fallback.
We would not sell against these.
- Single-intent, high-volume, low-value traffic. Balance enquiries, one-time passwords, delivery status. If the answer is one field from one system, you do not need reasoning.
- Pure cost-per-contact targets. If your mandate is to reduce cost per contact and nobody is measuring what the contact achieved, a cheaper minute wins the argument.
- A bot that already works. Replacing something functional to consolidate vendors is a procurement preference, not an outcome. Put us on a process the bot cannot finish instead.
- No human floor at all. Our model assumes you have agents to hand the hard calls to. If you have none and want none, we are the wrong shape.
Nobody has to rip anything out to find out.
Pick the process the bot gives up on
Usually collections, renewals or a save desk. The test is only meaningful where money is decided.
Agree the baseline before anything runs
What the current outcome rate is, on what volume, at what cost. Written down, agreed by both sides, so nobody argues about the denominator later.
Run in parallel, not instead
Your bot keeps its traffic. We take a defined slice of the same book. Same period, same list quality, same rules.
Compare outcome rate and cost per outcome
Not containment. Not minutes. If we do not move the outcome, there is nothing to discuss and you have lost a few weeks, not a platform.
We already have a voice bot. Do we rip it out?
No, and we would tell you if you should. If your bot handles balance enquiries and one-time passwords at low cost and your team is happy, leave it. The case for changing is not the simple calls; it is the calls where money is decided and your bot currently gives up. Run us on one of those and compare the outcome rate, not the containment rate.
Is this just a better bot?
The difference that matters is what happens when the AI cannot finish. A bot ends the call. We hand it to a person with the context, and the outcome still counts. That is why we report an outcome rate rather than an automation rate.
Our bot vendor also says it uses large language models now.
Most of them do. The question worth asking in that meeting is what happens to the call the model cannot resolve, and who is accountable for the number at the end of the month. Ask for the outcome rate on a book like yours and see what comes back.
What about the cost per call?
Ours may be higher per minute than a menu-driven bot, and lower per outcome. Those are different questions, and only one of them appears in your P&L. The calculator lets you put your own numbers against both.
How long before we know?
A pilot on one process against an agreed baseline. You will see the outcome rate inside the first few weeks, because the calls are real and the book is yours.
Put your number in and listen.
Reading about the difference between a bot and a conversation is a poor substitute for hearing one. Our AI will call you, in the accent and the language mix your customers use, and you can try to break it.