The short answer
The thirty terms below are the ones that come up in buying one: what the handover is, what the floor means, how the money is counted, and what the compliance words mean on a call.
The category
- Human + AI BPO
- An outsourced contact centre where AI voice agents make and take the calls and a trained person steps into the same call when the customer needs one, on the same queue, measured as one floor. The AI is billed per minute, the people per seat. Also called an agentic BPO.
- Agentic BPO
- A synonym for Human + AI BPO used by some buyers and some of the press. The press also says agentic process outsourcing (APO) for the wider shift, where AI runs the process and people supervise. The two words describe the same shape; we use Human + AI BPO because it says what each half does.
- Traditional BPO
- An outsourced contact centre staffed by people, billed per agent seat (Rs 28,000 to 35,000 a month in India for an English or Hindi agent). Its economics improve when more people sit longer, which is the opposite incentive to a Human + AI floor.
- Bot-only vendor
- A company that sells conversational AI software and no people. When the bot cannot finish a call, the call goes back to the client's floor. Automation rates quoted by bot-only vendors usually count the calls the bot kept and leave out the ones it dropped.
- The five tests
- Handover, Continuity, Floor, Outcome, Proof. The person sees the context as the call arrives; it is the same call with no re-verification; a human bench is on the queue from day one with the client's KPIs as the floor; the money is on one page; a live call is put on the buyer's number in the first meeting.
The handover
- Warm handover
- The AI brings a person into the live call and the person arrives with the summary, intent and last turns already on screen. The customer is told a colleague is joining and does not repeat themselves. Sometimes called a same-call transfer. The opposite is a callback.
- Callback
- The call ends and a person rings the customer later, usually with a ticket number and little context. Fine as a choice the customer makes when agents are busy; a different product when it is the only route to a person.
- Handover trigger
- The condition that brings a person in: the customer asks, disputes a fact, sounds distressed, or reaches a point a human closes better (a premium objection, a hardship, a ready buyer). Triggers are agreed before the pilot and tightened in the first weeks from the transcripts.
- Busy-floor rules
- What happens when the AI needs a person and none is free. On our platform the customer chooses: stay with the assistant, a callback at a named time, or hold with the queue position stated. The customer is never left in a silent queue.
- Context payload
- What travels with the handover: a two-line summary, the customer's intent in their own words, the last three turns, the account facts read from the CRM, and the disposition options for the queue. Identifiers arrive masked to fragments.
- Agent desk
- The screen a person works on. Either the client's existing desk, fed by webhook, or Vox Agent Desktop, Oriserve's console, at Rs 2,000 per agent per month with transcripts and wrap-up on the same screen.
- Disposition
- The recorded outcome of an attempt: promise to pay, renewed, callback booked, not interested, wrong number. Written back to the client's CRM or LMS after every attempt, by the AI or by the person, in the same fields.
- Wrap-up
- The notes and disposition an agent completes after a call. On Vox Agent Desktop the summary is drafted by the AI and the agent confirms it, which is where the wrap-up time saving comes from.
The floor and the numbers
- Floor (the people)
- The trained human bench on the same queue as the AI from day one of a programme. Drawn from one agreed source: the client's own agents, Oriserve's, or a partner's. Not hired when the bot hits a wall in month four.
- Floor (the KPI)
- The client's current numbers on the queue (CSAT, resolution, collection rate, compliance), agreed before the first call and written into the pilot as the level nothing may drop below.
- Baseline
- The client's historical performance on the chosen queue and metric, from the client's own reporting, agreed in writing before anything runs. The baseline sets the floor and, later, the reference for any outcome-linked terms.
- Ramp
- The share of calls the AI finishes, by month. On a typical programme: up to 40% by month 3, 50% by month 4, 70 to 80% by month 6. Month one is lower while the AI learns the book on live calls.
- Maturity
- The point, usually around month six, when the AI share settles in the 60 to 80% band and the people take the rest. The number we quote for a mature programme is 60 to 80% of calls finished by the AI.
- Containment
- A bot vendor's measure: the share of calls the bot kept to the end. Useful inside a programme and misleading as a headline, because it says nothing about the calls the bot dropped.
- Cost per outcome
- The total monthly bill (people, minutes, setup amortised) divided by the outcomes delivered (policies renewed, promises kept, tickets closed, leads passed live). The number to compare models on, at months 1, 3 and 6.
- Calibration weeks
- The first weeks of a programme, when the client and Oriserve read transcripts together, change scripts, grow the do-not-say list and tighten triggers. How fast the ramp climbs depends on this.
The money
- Pilot rate
- The per-minute rate for the AI during a pilot: Rs 2.50 a minute in India, with no commitment and no minimum bill. The pilot sets the baseline on live calls.
- Committed rate
- The per-minute rate once a volume is committed, stepping down in slabs to Rs 1.75 a minute at the largest. The middle slabs sit on the quote because they depend on volume.
- Seat
- The unit people are billed in: one agent, one month. On a Human + AI floor the seat count falls as the ramp climbs; on a traditional BPO it is the whole bill.
- Setup
- One-time work priced per integration: SIP connection, CRM read and write-back, agent-desk webhook. A line on the quote, not hidden in the minute rate. Smaller for a file exchange.
- Outcome-linked option
- An option, after the pilot has set a baseline, to tie part of the fee to the metric the client chooses; when the metric is not met, that part is waived. Not the default construct.
Compliance words on a call
- Contact window
- The hours a recovery call may be placed. Under the RBI directions in force from 1 January 2027, 8 am to 7 pm. A dialler setting, so a call outside it cannot be placed.
- Recording notice
- The sentence at the start of a call telling the customer it is recorded, in their language, before any account detail. Required on recovery calls from January 2027.
- Retention
- How long recordings and transcripts are kept. Set in the data-processing agreement; the RBI floor for recovery calls is six months, longer while a dispute is open.
- PII masking
- Showing identifiers as fragments on the agent screen by default, with full values behind a role and every access logged. Card numbers never reach the screen because the AI redirects them to a payment link.
- Data-processing agreement (DPA)
- The contract that sets the field scope, retention, data-principal request channel, breach clock and sub-processor list under the DPDP Act 2023. Oriserve provides it during onboarding.
- Do-not-say list
- The client's list of phrases and claims the AI may never use on a call, loaded into the script and grown during the calibration weeks. Collections programmes always have one.
Questions people ask
What is a warm handover?
The AI brings a person into the live call with the summary, intent and last turns already on screen. The customer does not repeat themselves. The opposite is a callback.
What is the difference between a Human + AI BPO and an agentic BPO?
None in shape; they are two names for the same thing. Oriserve uses Human + AI BPO because it says what each half does.
What does containment mean?
The share of calls a bot kept to the end. It says nothing about the calls the bot dropped, so cost per outcome is the better comparison.