Switch From a Call Centre: 5 Steps, ₹2/min

Here is the whole sequence before the detail. Nothing in it requires you to cancel anything on day one.

1. Extract the script from the team, not from your memory. Sit with the people currently taking the calls and write down what they actually say, including the answers that were never in any document.
2. Decide what happens to the number. Keep the existing line and forward the overflow, or point the line at the agent and route escalations back. Both work, they fail differently, and which one is available to you depends partly on your own operator and partly on your contract's notice period.
3. Draw the escalation ladder before anything goes live. Which calls get flagged, to whose phone, and within how long. Written down, agreed by the person who will actually pick up.
4. Run both side by side for a few weeks. Start the agent on one narrow slice — after-hours only, or overflow only — while the team keeps everything it has today. Nothing is cut in this step.
5. Move only the volume you have evidence for. Read the transcripts. Move the call types the agent demonstrably handled. Leave the rest with people.

The reason step one comes first is that an AI agent is a script executor with good ears. It will say what it was told to say, consistently, at 2am on a Sunday. What it cannot do is invent the sentence your best telecaller uses to save a cancelling customer, because nobody has ever written that sentence down. Step one is the step most people skip, and skipping it is how you end up with a technically working agent that says nothing your customers recognise.

The reason step four exists is that a parallel run is the only honest test. A demo on a scripted call proves the agent can talk. A fortnight of real callers, in real accents, at real hours, with your real edge cases, proves whether it should take more. During that fortnight you are paying twice — your existing invoice plus credits — and that overlap is a genuine migration cost, not a hidden one. Budget for it rather than trying to design it away.

Steps two, three and five are the ones with operational teeth, and they get a section each below.

Step 1 — take the script off the team

What you are collecting is not a document. It is four things:

• The twenty questions. Ask the team to list the questions they answer most. In an Indian SMB the list is usually short and repetitive: are you open, what does it cost, where exactly are you, do you have it in stock, can I come Thursday, do you do EMI, is the doctor in today. This is the volume, and it is the part an agent handles well.
• The exact answers, in the exact words. Not a policy summary. The sentence the person says. If your team says "sir, the consultation is ₹500 and it adjusts against the treatment if you book the same week", that whole sentence is the asset.
• The objections and what beats them. Ask which objections they hear all day and what they say back. Record both, in their words.
• The edge cases they handle by feel. This is the residue: the caller who wants the price matched, the regular who always gets fitted in, the enquiry that is really a complaint. You are not writing these down to automate them. You are writing them down so you can recognise them and route them to a person.

If you have call recordings, use them. Twenty real recordings beat two hours of trying to remember. If your current vendor holds the recordings, ask for them in writing before you give any notice — that is a request that gets slower after a relationship changes.

Step 2 — decide what happens to the number

There are two workable patterns and one that causes trouble.

Pattern A, keep your line and forward. Your published number stays exactly where it is. Calls that go unanswered after a set number of rings, or arrive outside working hours, forward to a second number where the agent answers. Nothing about your printed material, your Google listing or your ads changes. This is the low-risk pattern and it is where most migrations should start.

Pattern B, agent first, humans behind. The published line goes to the agent, which handles the routine and transfers or flags anything else. This is the pattern that changes your cost shape most, and it is the one to move to after the parallel run, not before it.

The pattern that causes trouble is publishing a brand-new number and hoping traffic moves to it. It mostly does not: your number is printed on boards, saved in phones and sitting in old ads.

One thing neither pattern can promise on your behalf: what your line can actually do. Conditional forwarding — on busy, on no-answer, or outside set hours — is a setting on your own connection, and what is available differs by operator and by whether the line is a mobile, a landline or already on a cloud-telephony system. Porting a number, or getting a toll-free line, is a separate question again, decided by your operator rather than by us. Ask your provider what your line supports before you design the routing around it.

If we supply the number, it is ₹2,000 per 30 days, auto-debited from your credit balance. One number and one concurrent line are included in that; an extra concurrent line is ₹1,500 a month and an extra number ₹750 a month. If your existing telephony already gives you a line you can forward from, you may not need a number from us at all — check that before you buy one.

One contractual point that is yours to check, not ours to assert: outsourcing agreements commonly carry a notice period, and some carry a minimum monthly commitment that survives a reduction in volume. Read yours before you pick a date. A migration plan that assumes you can halve the seat count next Monday is a plan built on an assumption you have not verified.

Step 3 — the escalation ladder

An escalation ladder is three columns: the trigger, the destination, and the time. Write it before go-live, because writing it afterwards means the first angry caller is the test.

A workable starting ladder for a small business:

• Caller is upset, or uses complaint language. Agent stops trying to resolve, takes the details and the substance, and flags the call. Destination: the owner or the floor manager, by WhatsApp, immediately. Call back the same day.
• Caller asks for something outside policy — a discount, a waiver, a slot that is gone. Agent says it will have someone confirm, captures the request, flags it. Destination: whoever actually has that authority. Call back within business hours.
• Caller is a named account. Give your top accounts a number that reaches a person and keep them off the automated line entirely. This is the cheapest customer-retention decision in the whole migration.
• Agent does not understand after two attempts. Take a number and flag. Two attempts, not five. An agent that keeps asking a confused caller to repeat themselves is producing a worse outcome than a busy tone.

The thing that makes this ladder real is the destination being a specific person with a specific phone, agreed in advance. "It gets escalated" is not an escalation ladder.

Step 4 — the parallel run

Pick the narrowest slice with the clearest evidence. Two good options:

After-hours only. Everything from your closing time to your opening time goes to the agent. Your team's day is untouched. You learn what you have been losing overnight, which is a number you cannot get any other way.

Overflow only. Calls that ring out while your team is on other calls forward to the agent. Your team keeps every call it can take. You learn how much you are losing to a busy line.

Run for two to four weeks. Then read transcripts — not a dashboard summary, the actual transcripts of at least fifty calls. Sort them into three piles: handled correctly, handled but awkwardly, should never have been the agent's call. The second pile is a script fix. The third pile is your routing rule.

During the parallel run you carry both costs. If the agent handles 1,200 minutes in that fortnight, that is ₹2,400 of voice on top of whatever your existing arrangement bills, plus the number recharge and, in the first three months, the Care Plan. That is the price of finding out before you commit, and it is far cheaper than finding out afterwards.

One thing we will not do at any step, including this one: outbound calls go only to your own leads and enquiries — people who contacted you or asked to be contacted. We do not call bought or scraped lists. If your current telecalling operation runs on a purchased database, that part of the work does not move to us. On the legal framing around Indian outbound calling, see is AI calling legal in India.

The last step is a decision about proportions, and the honest version is rarely "all of it".

From your three piles, you now know which call types the agent handled correctly at your volumes, in your languages, with your callers. Move those. Typically that is first answer, timings and location, price and availability questions, booking and rescheduling, confirmation and reminder calls, and speed-to-lead callbacks on enquiries you already own — the callback that goes out within about a minute of a form or ad enquiry arriving, which is a thing no shift can do reliably at 11pm. More on that at outbound calls.

Leave with people: complaints, negotiation, exceptions, named accounts, and anything where the outcome depends on reading the person and adjusting mid-sentence.

What this usually does to your outsourcing spend is not "cancel it". It is one of three smaller things, and all three are real:

1. You stop buying the shift you were buying for coverage rather than for skill — the evening or Sunday seat that exists so the phone is answered, not because those calls are hard.
2. You stop scaling seats with volume. The next 3,000 minutes cost ₹2/min instead of another seat, because an agent is not bound to a desk or a shift.
3. You give the people you already have their harder half back. The part of the day that goes on repeating the same four sentences is the part that moves.

None of those requires anyone to lose a job, and I am not going to write a page that implies otherwise. If the outcome you want is a headcount cut, this page is not the argument for it, and an AI agent bought for that reason tends to disappoint, because the calls that are genuinely hard do not go away — they just arrive at a worse-prepared destination.

The pace that works: after-hours or overflow for a month, then one more call type a fortnight, each with the transcripts read before the next one moves. The pace that does not work: switching everything on a Monday because the contract renewal is on the Friday.

This is the section that decides whether the migration is a good idea, and it should be read before the cost tables rather than after them.

Read that honestly and the split is obvious. The AI column wins on coverage, consistency, languages and cost shape. The human column wins on judgement, emotion, and any work that has not yet been defined. Those are not the same jobs, and a migration that pretends they are is a migration that goes badly in the second month.

There is one more advantage a BPO has that deserves stating plainly: they will absorb mess. If you need bodies on undocumented work by next week and you cannot spare the hours to define what "handled" means, an outsourced seat is a genuine answer and I will say so rather than talk you out of it. Everything above assumes you can spare a few hours for step one. If you cannot, do not start.

You already know the number on the left side of this comparison — it is on your invoice. So rather than quote industry seat rates, here is our side in full, at volumes an outsourced operation would recognise. Put your own invoice beside it.

Our commercial model has three moving parts and no fourth: ₹2 per minute of talk time on simple whole-minute billing, ₹2,000 per 30 days for the phone number if we supply it, and — for founding clients — the Care Plan at ₹5,000 a month, required for the first three months and optional from month four. All three come out of one prepaid credit balance.

Read the effective-rate columns rather than the totals. At 2,000 minutes the fixed ₹7,000 dominates and you are really paying ₹5.50 a minute; at 20,000 minutes it is ₹2.35 and falling toward the base rate. An AI agent gets better the more minutes you have, which is precisely the condition a business already paying a telecalling team is in.

The credit balance, stated properly

All of the above is exclusive of GST. 18% GST applies when you load credits, so ₹17,000 of usable balance is billed ₹20,060, and ₹27,000 is billed ₹31,860. The minimum first load is ₹10,000, billed ₹11,800 — but if your first month's run rate is ₹17,000, load ₹17,000, not the minimum. Later top-ups start at ₹500. Credits never expire, and at a zero balance everything pauses gracefully rather than producing a surprise bill.

The month-one debit order is worth seeing once: on a ₹10,000 load, ₹2,000 goes to the number recharge and ₹5,000 to the Care Plan, leaving roughly ₹3,000 for actual talk time — about 1,500 minutes. Whether 1,500 minutes is a fortnight or a quarter for you is a question only your own invoice answers, so read your current minutes off it and load to match your first month rather than to the minimum.

The build

The custom build is ₹40,000 one-time, and it is free for our ten founding clients. The trade is stated plainly and never as fine print: the free build is in return for (a) the Care Plan for the first three months at the founding rate of ₹5,000 a month, and (b) an honest testimonial and a short case study after go-live. Honest means honest — if it underperforms for your business, that is what the write-up will say.

During the parallel run in step four you carry this and your existing invoice. That overlap is a real cost and it is the correct cost. Full detail at pricing; if you want to size the loss you are currently carrying before spending anything, the missed-call calculator will structure it.

If you are switching, you should shortlist more than one vendor. Here is what the Indian market published on its own pages when we checked it on 21 Jul 2026, with each vendor's own qualifiers attached — because in this category the qualifier is usually the story.

Exactly one vendor on that board — SquadStack — publishes no rate at all. Most of the others do publish something; the work is reading what the number is attached to. Three patterns to watch for when you shortlist: a low headline rate that lives on a quote-only enterprise tier, a per-minute rate that excludes telephony so a carrier bill lands on top, and a per-minute rate that sits above a mandatory monthly plan.

One caution about research, learned the hard way. A rival's comparison page is not a source for a third party's price. One such comparison page lists Bolna at ₹8–11/min, while Bolna's own pricing page publishes 6.00¢/min displayed as ₹5.52/min. We do not republish rivals' figures for other vendors, and neither should you — take every number off the vendor's own page, with a date. Our dated, primary-sourced version is the India AI voice agent pricing index.

Our claim here is about terms, not about being the lowest number on the board: ₹2/min from the very first minute, no subscription, no minimum volume, no lock-in. Competitor pricing moves — several rates on this board changed within weeks of being recorded — so re-verify before you sign anything, including with us.

Four things reliably need handling, and none of them are technical.

The undocumented knowledge. Covered in step one, repeated here because it is the failure mode. When a team leaves an account, the reasons behind the answers leave with them. Do step one while the relationship is good.

The handover of assets. Call recordings, the lead sheet, the dispositions, the contact list, the DLT registrations if bulk outbound is involved, and any integrations pointed at your CRM. Ask for all of it in writing, early, and confirm the format. "We will export it" and "here is a CSV with the fields you need" are different commitments.

The escalation gap in week one. The agent will flag calls, and someone has to actually pick them up. In the first fortnight that person should be you or your manager, not a rota. If flagged calls sit unanswered for a day, you have replaced a slow answer with a silent one, which is worse.

The customers who notice. Some callers will realise it is an agent, and a few will dislike it. The mitigations that work are being straightforward about it, keeping the transfer path short, and giving your named accounts a human number. The mitigation that does not work is trying to make the agent pass for a person.

When you should not switch

Most of your calls are the hard kind. If the majority of your inbound is complaints, disputes and negotiation rather than "are you open, what does it cost, can I book Thursday", the agent handles the minority and flags the rest, and the economics stop working.

You cannot spare the hours for step one. A vendor team will take on undefined work with no mapping phase. We cannot, and pretending otherwise would just move the failure to month two.

Your outbound runs on bought lists. We call only your own leads and enquiries. That part of your operation does not migrate to us.

You need an enterprise vendor. We are pre-revenue with founding slots open — no case studies yet, no third-party review profile, no compliance certifications, no named support SLA. If your procurement requires those, buy from an established platform. I would rather say that than spend your evaluation cycle. We do not invent testimonials, ratings or statistics to fill that gap either, which is why you will not find any here.

Your contract makes it expensive right now. If your notice period or minimum commitment means you pay for seats you are not using for another quarter, start the parallel run anyway and time the reduction to the contract, not to your enthusiasm.

What we commit to

The number: ₹2 per minute, whole-minute billing, permanently — not a launch discount that expires. Chat at ₹2 per conversation. WhatsApp message fees billed by Meta directly to your own account with zero markup from us. No subscription, no minimum monthly usage, credits that never expire. I build each agent personally, which is why it is days rather than months, and also why a business needing a support floor should read enterprise first. The process is at how it works, the founding programme at pilot.

If you want to start, send your rough monthly minutes and what your team currently handles to WhatsApp on +91 93923 98750 — the message to send is your problem, not a request for a demo. You will get back the arithmetic for your own volumes, including the case for staying where you are if that is what the numbers say.