AI Receptionist vs Call Center India: ₹2/min

Indian SMBs almost never choose between "AI and nothing". The real shortlist is: hire a telecaller, hand inbound to a small BPO, or run an AI agent. Here is how the three differ on the things that actually change the decision.

Read that table honestly and two things fall out. The AI column wins on cost shape, hours and consistency. The human columns win on judgement, emotion and everything that is not a phone call. That is not a rhetorical concession before a sales pitch — it is the actual result, and it is why the section on hybrid operating models further down matters more than the price tables.

If you want the narrower comparison against a phone menu rather than a person, that is a different question and lives at AI voice agent vs IVR.

Start with the salary, from a source that publishes its sample size and its date.

Jooble India — Telecaller salary (verified on in.jooble.org, checked 20 Jul 2026) gives an average of ₹21,007 per month, or ₹2,52,085 per year, with a range from ₹2,06,052 to ₹3,09,884 per year. It also shows ₹4,848 per week and ₹117.19 per hour. That figure is based on 4,481 reported salaries and 1,978 active job postings, and the page states its data is as of 15 Jul 2026. That combination — sample size plus a date — is why I am using it as the headline number rather than any of the higher or lower figures floating around search results.

Two corroborating sources, both with caveats I will not hide:

• Indeed India (verified on in.indeed.com, checked 20 Jul 2026) shows ₹19,958/month for "Caller", ₹16,303/month for "Telemarketer" and ₹12,433/month for "Customer Service Representative". The sample sizes are 14, 2 and 2 salaries respectively, estimated from 46 job postings, and the page shows no date. Directionally consistent with Jooble; far too thin to headline.
• Telemploy's Telecaller Salary Guide India 2026 (verified on telemploy.com, checked 20 Jul 2026) is a recruiter-platform estimate — it carries no publication date and cites only "listings on Telemploy.com and feedback from recruiters", with no methodology. Treat it as a recruiter's read of the market, not authority. On that basis it puts freshers at ₹10,000–₹18,000 in-hand, 1–2 years at ₹14,000–₹25,000, 2–4 years at ₹20,000–₹35,000 and team leads at ₹30,000–₹55,000; by city, Bangalore and Delhi NCR highest at ₹15,000–₹25,000 fixed for fresher-to-two-years and tier-3 cities at ₹8,000–₹13,000; by industry, BFSI freshers ₹16,000–₹22,000, EdTech ₹18,000–₹28,000, real estate ₹12,000–₹18,000, and international BPO ₹22,000–₹40,000 plus a ₹2,000–₹4,000 shift allowance.

The layer almost every comparison page omits

Salary is not cost-to-employ. Two statutory contributions sit on top, and neither appears on the comparison pages I read.

ESI. Employer contributes 3.25% of gross wages, employee 0.75%, total 4%. Coverage is mandatory for employees drawing monthly wages up to ₹21,000, or ₹25,000 for employees with disabilities. Rates have been effective since 1 Jul 2019 and are unchanged. Verified on cleartax.in and independently corroborated on tallysolutions.com, both checked 20 Jul 2026. Note the sharp edge here: Jooble's average of ₹21,007 sits ₹7 above the ₹21,000 coverage ceiling. A telecaller paid at exactly the national average may fall outside ESI; one paid ₹20,500 does not. Check your own payroll rather than assuming.

EPF. Employer contributes 12% of basic salary plus DA, split 8.33% to EPS and 3.67% to EPF, with the EPS portion calculated on a maximum wage of ₹15,000/month regardless of actual basic plus DA. Employee contributes 12%. There are additionally EDLI and administrative charges of 0.5% each. Verified on prashasthicorporate.com (published 17 Jun 2026), checked 20 Jul 2026. A limitation I will state rather than paper over: the EPFO's own contribution-rate PDF could not be fetched — the connection failed twice — so this rests on a secondary source that is internally inconsistent about whether EDLI and admin charges sit inside or on top of the 12%. So: take the employer contribution as 12% of basic plus DA, and treat total employer outgo as *roughly* 13% at most. Do not let anyone quote you a precise on-cost percentage sourced from a blog, including this one.

Putting it together

Assuming basic plus DA is set at 50% of gross — a common but not universal structure, so check yours:

So a telecaller near the national average costs you roughly ₹22,300–₹23,000 a month all-in, not ₹21,000 — and which end of that you land on depends on whether the wage clears the ESI ceiling. Either way it is before recruitment, before the desk, phone and connection, and before the cost of the seat being empty. The rest of this page uses ₹23,000 as the working figure, which is the conservative end for us.

Telecalling churns. That is not a slight on the people doing it; it is the structure of the job.

AVOXI (verified on avoxi.com, checked 20 Jul 2026), citing research it attributes to the Global Call Centre Research Network, reports call-centre attrition by ownership model: in-house 28%, subcontractor 38%. For comparison the same research gives USA in-house 26% and USA subcontractor 36%, small call centres 17% versus large 44%, and entry-level agents 27% versus senior agents 12%. It also states that replacing one agent costs about 16% of their gross annual earnings. The page gives no year for the underlying research, so read it as "research cited by AVOXI", not as current-year Indian data.

1840 & Company (verified on 1840andco.com, page last modified 9 Jun 2026, checked 20 Jul 2026) gives India-specific attrition of 35–45% annually nationally, with Bangalore at 40–50%, Hyderabad 35–42%, Pune 30–38% and Mumbai 38–46%.

Apply the AVOXI replacement-cost rule to the Jooble salary and you get a number worth carrying in your model:

• Gross annual earnings: ₹2,52,085
• Replacement cost at 16%: ≈ ₹40,334 per exit
• At 28% in-house attrition, expected annual churn cost: 0.28 × ₹40,334 ≈ ₹11,293/year, or roughly ₹940/month
• At 38% subcontractor attrition: ≈ ₹15,327/year, or roughly ₹1,280/month

That pushes realistic cost-to-employ for one in-house telecaller to around ₹23,000–₹24,000 a month, and it does not attempt to price the harder cost — the weeks where the seat is empty, or filled by someone still learning your business. There is no honest number for that, so I am not inventing one.

The middle path most Indian SMBs assume is cheap. The published rates say otherwise.

1840 & Company (verified on 1840andco.com, last modified 9 Jun 2026, checked 20 Jul 2026) publishes India outsourcing rates as:

I am leaving those in dollars deliberately. No exchange rate was verified on 20 Jul 2026, and quietly converting at a guessed rate would be exactly the kind of soft number this page exists to avoid. Convert at your own bank's rate and the per-seat figure will land well above a locally hired telecaller.

Three structural points about the BPO route that matter more than the rate card:

1. They sell seats, not outcomes. A dedicated seat is a person's time. If your call volume is spiky — a burst at 10am and again at 7pm, dead in between — you pay for the flat hours regardless.
2. There is usually a floor. Most outsourcers do not take on a business that needs three hours a day. The per-seat model is the floor, and it is a full seat.
3. They absorb mess, and that is a genuine advantage. This is the thing a BPO does that neither a fresh hire nor an AI does well: they will take on undocumented, judgement-heavy work starting next week without you first mapping the process. If you need bodies on unpredictable work tomorrow and cannot spare the time to define what "handled" means, a BPO seat is a real answer and I will say so plainly.

What you give up is control of quality and, per the attrition data above, subcontractor churn runs higher than in-house — 38% versus 28% in the research AVOXI cites. The person who learned your business in March may not be on your account in July.

Here is our entire commercial model. No demo required to see it, no "contact sales" gate, no tiers.

The founding-client trade, stated plainly

The ₹40,000 build is free for ten founding clients in return for two things: the Care Plan for the first three months at the founding rate of ₹5,000/month, and an honest testimonial plus a short case study after go-live. Honest means honest — if the agent underperforms for your business, that is what the case study will say. Ten accurate write-ups are worth more than fifty flattering ones.

For a detailed rate menu, where one is warranted

Most businesses never need this. Where a breakdown is genuinely warranted: Bharat Standard ₹2/min, Studio HD ₹3/min, Global covering 99 languages ₹4–5/min, Ultra realtime ₹10/min, call recording adds ₹0.10/min. These are voice quality and language-reach options, not subscription tiers — the advertised base you will actually run on is ₹2/min. Full detail lives on pricing.

Who builds it

I do. I am Rohith Sriramula, a laid-off engineer working from home in Paloncha, Bhadradri Kothagudem, Telangana. There is no office, no support floor and no headcount, and I am not going to pretend otherwise on a page about honest costs. What that buys you is that the person who wrote your call flow is the person who answers when it breaks. If you need enterprise onboarding, a named SLA or compliance paperwork, read enterprise first — and the "when you should not buy from us" section below.

The most useful thing I read while researching this page was written by a competitor, and it argues against the category's own advertising.

Caller Digital's voice AI pricing page for India (verified on caller.digital, no visible publish or update date, checked 20 Jul 2026) states that headline advertised rates in India run ₹2–₹12/min, with ₹3–₹6/min the most common mid-market band — but that the effective rate in production is ₹6–₹25/min. It says explicitly that a ₹3/min quote typically costs ₹6–₹9/min once everything lands. Their component breakdown:

Two caveats. First, that telephony base figure is described second-hand as a regulated base rate; no regulator's page was fetched, so do not attribute it to TRAI directly. Second — and this is the part worth sitting with — that same pricing page does not publish caller.digital's own rate. It lists "Tier 1: transparent per-minute pricing" and "Tier 2 (above 100,000 minutes/month): outcome-based pricing" with no numbers, and directs readers to book a demo. A page that documents the industry's hidden fees, and then hides its own, is the clearest possible argument for reading the invoice not the headline.

The one vendor I found publishing a complete public rate card is Aixclerate (verified on aixclerate.com, checked 20 Jul 2026):

Add-ons: extra business number ₹699, call recording ₹0.20/min, additional concurrency ₹499. Two months free on annual billing. Doing the arithmetic on their own published figures: ₹9,999 ÷ 500 minutes = about ₹20/min if you use exactly the allowance, and ₹24,999 ÷ 2,000 = about ₹12.50/min. Use less than the allowance and the effective rate rises; use more and you pay ₹7/min on top.

And HuskyVoice's AI-versus-human receptionist comparison page publishes no cost figures at all on the comparison itself, routing readers to a separate pricing page. We keep a like-for-like breakdown at Dvaarik vs HuskyVoice, Dvaarik vs Bolna AI and Dvaarik vs Vapi.

Where this leaves us honestly: publishing a price is not a moat. Aixclerate does it too. The difference worth arguing is the *shape* — no subscription, no minimum monthly usage commitment, credits that never expire, and a ₹10,000 first load against the ₹50,000–₹5,00,000 minimum commitments caller.digital documents as normal in this market.

Assumptions, stated so you can argue with them: an average handled conversation of 3 minutes, one in-house telecaller at ₹23,000 all-in per the maths above, and Dvaarik at ₹2/min plus the ₹5,000 Care Plan and ₹2,000 number recharge during months one to three.

The AI column is lower in absolute monthly spend at every row. That holds until roughly 2,600 calls a month, where ₹7,000 + ₹6 per call crosses ₹23,000 — working: (₹23,000 − ₹7,000) ÷ ₹6 ≈ 2,667 calls. Past that, one salary is cheaper than our usage bill, if one person can absorb the volume.

Whether one person can is not something I can assert — there is no verified capacity data here. The only public number I found is a claim by caller.digital on its India answering-service page (published 14 Jul 2026, checked 20 Jul 2026) that one receptionist handles 1,100–1,400 calls a month. That page cites no source for it, so treat it as their assertion, not an established figure. If your own logs show you need two people at 2,000 calls, the human line doubles to ₹46,000 — but that is your measurement to make, not my assumption to bake into a table.

The same caller.digital page also asserts a receptionist salary of ₹18,000–₹30,000/month, a four-receptionist desk at ₹92,000/month, and per-handled-call costs of ₹9–₹22 for AI versus ₹17–₹22 for a human. None of those carry a cited source on the page either. I am reporting them as claims because they are widely repeated, not because they are verified.

From month four, the floor drops

The Care Plan is required for the first three months and optional from month four. Drop it and your fixed floor falls from ₹7,000 to ₹2,000 — just the number recharge. At 100 calls a month that changes cost per call from ₹76 to ₹26: (₹600 + ₹2,000) ÷ 100. At 300 calls it becomes ₹12.67: (₹1,800 + ₹2,000) ÷ 300.

The asymmetry the table cannot show

One telecaller covers roughly one shift, six days. The agent is not bound to a shift or a desk — it answers at 2am on a Sunday at the same ₹2/min. So the per-call comparison above is not like-for-like in the human's favour: to match the coverage you would be pricing two or three shifts plus the ₹2,000–₹4,000 shift allowance Telemploy describes for night work. If a meaningful share of your enquiries arrive after 8pm or on Sunday, the gap widens further in our direction.

Not a token paragraph. These are the situations where I would tell you to keep or hire the person.

The caller is already angry

Someone who has been overcharged, kept waiting, or let down does not want efficient. They want to be heard by a person who can say "that should not have happened" and mean it. An agent that answers in one ring and processes the complaint correctly still fails that call. Our agents do not try to argue an upset caller down — they take the details and flag the call so you can ring back yourself.

Judgement calls and exceptions to policy

"Can you waive the cancellation fee this once?" "My mother is unwell, can you fit me in tomorrow?" Every business runs on exceptions its own owner makes by feel. An AI agent applies your rules. It cannot decide when to break one, and I would not build it to.

Negotiation, discounting and complaint resolution

Anything where the outcome depends on reading the other person and adjusting the offer mid-sentence. A person does this. A scripted agent, however good the voice, does not.

Relationship accounts

Your twelve biggest customers, the ones who ring and expect to be recognised. Those calls should reach you or your senior person directly — give them a number that does, and keep them off the automated line entirely.

Everything that is not a phone call

This is the one comparison pages forget entirely. A receptionist greets walk-ins, manages the waiting room, handles the courier, chases the pending file, and reads the mood of the space. If the person you are considering replacing does any of that, you are not comparing like with like — you are comparing a phone line to a job.

Written straight, because a page arguing for honest costs cannot end with a pitch that pretends we win everywhere.

Your call volume is low enough that someone absorbs it inside their existing job. If you take 60 calls a month and your front desk answers them between other tasks, the AI is a solution to a problem you do not have. We would still cost less than a salary in absolute terms — that is not the point. At 100 calls a month during the required Care Plan period we cost about ₹76 per handled call, which is real money for a problem measured in minor annoyance. Come back when you are missing calls you can name.

You need social proof or an enterprise vendor. We are pre-revenue with founding client slots open. We have no case studies, no third-party review profile, no compliance certifications and no named support SLA. Established Indian platforms have years of accumulated authority and enterprise logos. If your buying process requires any of that, you should buy from them, and I would rather say so than waste your evaluation cycle. Our rules also forbid manufacturing testimonials or statistics to fill the gap, which is why you will not find any on this site.

You need bodies on messy work tomorrow. A BPO will take on undocumented, judgement-heavy work with no build phase and absorb complexity we would need mapped up front. At $1,200–$2,400 per dedicated seat per month (1840 & Company, last modified 9 Jun 2026, checked 20 Jul 2026), that is a real answer to a real need.

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

We are more transparent than most, not uniquely so. Aixclerate publishes a full public rate card with add-ons. Our argument is the structure — no subscription, no minimum usage commitment, credits that never expire — not a claim to be the only honest vendor in India.

One method limitation, restated. The EPFO's own rate table could not be fetched for this page. The statutory PF maths above rests on a secondary source. Verify against your payroll provider before you build a business case on it.

For most businesses reading this, the answer is not "replace the person". It is a split, and the split is not complicated.

Give the AI:

1. First answer on every inbound call, so nothing rings out
2. Overflow when calls land while your person is already busy
3. After hours, Sundays and holidays
4. The repeat questions — timings, price, location, availability, directions
5. Booking, confirmation and the reminder (booking)
6. Speed-to-lead callbacks on your own ad enquiries, typically inside about 60 seconds (speed-to-lead)
7. WhatsApp and Instagram replies that currently sit unanswered overnight (WhatsApp, Instagram)

Keep for the human:

1. Anything flagged, and every genuinely upset caller
2. Exceptions, discounts and negotiation
3. Named relationship accounts, on a number that reaches a person
4. Walk-ins and everything happening in the room

What this does to the salary line is usually not "remove it". It is "stop hiring the second one", or "stop paying a night shift", or "give the person you have back the two hours a day they currently spend saying the same four sentences". That is a smaller claim than most vendors make and a more defensible one.

One constraint that is ours, not a limitation of the category: outbound calls go only to your own leads and enquiries. We do not call purchased lists. Calling hours are respected and the agent identifies your business at the start of the call. If your plan depends on cold-dialling a bought database, we are not the vendor. More on the split at inbound vs outbound AI calling.

A worked example (invented, and labelled as such)

Sundara Skin & Hair, Warangal — an invented business used only to show the arithmetic. No such client exists; the numbers below are illustrative inputs, not observed results. Assume 420 enquiry calls a month, average 3 minutes, and assume a third arrive after 8pm or on Sunday — an assumption, not a measurement, so check your own logs. Currently one telecaller on a day shift, and the owner's phone after hours.

The existing telecaller stays under Option C. What changes is that the evening hire does not happen and the after-hours calls stop reaching the owner's personal phone. The honest caveat: the saving here comes from not adding a second salary, not from the AI being free. If your volume is low enough that the current person already handles the evenings, Option C is buying coverage you may not need yet.

Five steps, none of which require talking to a vendor.

1. Count. Two weeks of call logs. Total calls, missed calls, calls outside working hours, and average duration. Everything above is arithmetic on these four numbers, and guessing them is why most of these decisions go wrong. The missed-call calculator will structure it if you want.
2. Classify. Tag a sample of 50 calls as routine (timings, price, availability, booking) or hard (complaint, negotiation, exception). If hard is more than about a third, weight toward people.
3. Price all three. Telecaller at your city's band plus employer PF and, where the wage is under the ceiling, ESI. BPO at published per-seat rates. AI at ₹2/min against your actual minutes plus the fixed floor. Use your numbers, not the illustrative ones here.
4. Ask every AI vendor for the effective rate, in writing. Platform fee, telephony markup, language surcharge, storage, minimum monthly commitment, setup, and what happens to unused credits. Caller.digital's own pricing page documents that a ₹3/min headline commonly becomes ₹6–₹9/min in production — that is vendor-side confirmation that the headline is not the invoice.
5. Test it on your worst call. Not the easy one. Take the enquiry your team handles badly at 9pm on a Sunday and see whether the agent handles it or takes the details and flags it for you. Either outcome is useful; only one of them is a reason not to buy.

If you want the sector-specific version of this maths, we have written it up for clinics, salons, real estate and restaurants. Details on the build itself are at how it works and the pilot.