Blog · AI Calling

How to set up an AI calling agent in India, step by step.

Setting up a voice agent is mostly a data and decisions job, not a telephony job. Here is the order we would do it in for an Indian sales or distribution team, and what to have ready before the first call.

The regional manager has the attendance sheet open at 9:15 am. Three reps are on leave, one beat in Nagpur has not ordered in a week, and the helpline voicemail holds eleven messages from last night. Someone in the meeting says the phrase “AI calling” and everyone nods. Then the real question lands: what do we actually have to do to switch one on?

Our answer: setting up an AI calling agent for sales in India is mostly a job of preparing your own price list, script and handoff rules, and only partly a technology job, so the work starts on your side of the table. Get those three things right and the first live call is a short step. Skip them and no voice, however natural, will save the call.

Below is the order we would follow, written as a checklist you can hand to a sales ops lead. If you first want to know what a voice agent does on a call, read what an AI voice agent for sales does and come back.

Start with one job, not a platform

The fastest setups pick one bottleneck and one agent. In our product there are two jobs on the same engine. The order agent calls outlets on beats where the rep is absent, answers the retailer helpline and sends orders to the distributor. The sales concierge answers high-value enquiries after hours and closes on a booked visit.

These are different calls with different scripts, data and success measures. Choose the one that hurts most this quarter. A distribution team with frequent rep leave usually starts with absent beats. A business with Sunday enquiries going cold starts with after-hours response, which we cover in after-hours lead response with an AI calling agent.

Write the job down in one sentence before anything else. “Call every outlet on an absent rep’s beat and book its usual order” is a job. “Use AI for sales” is not.

The price list is your first deliverable, not the script

An order call is a data lookup that happens to be spoken. The agent answers only from your knowledge base, so whatever is in that knowledge base is what the retailer hears. That makes the price list the most important file in the whole setup.

Before the first test call, have these ready in one place:

  • Every SKU with its price by zone and by packing, so a carton price is never read out as a packet price.
  • The minimum order for each SKU or outlet type.
  • Every live trade scheme, with a start and an end date.
  • One named owner who approves price changes before they reach the agent.

The product includes price list sync and a price-change review step for exactly this reason. A scheme that ended on Sunday but is still quoted on Tuesday becomes a promise someone has to honour or explain.

A voice agent says exactly what your price list says, so a messy list becomes a confident mistake at scale.

Write the script from your best salesperson’s answers

Most teams try to write a script from scratch. We think it is better to sit with your best person for an hour and record what they actually say. How do they open? What do they ask first? How do they answer “price zyada hai” or “abhi stock hai”?

For the sales concierge the shape is simple: greet, discover the need, pitch in three points, qualify, close on a visit. The budget is asked once, softly, after the value is clear. Objection handling comes from your best salesperson’s words, not from a generic template. For the order agent, the script is shorter: confirm the outlet, take the order, check it against the minimum, apply the scheme and confirm before hanging up.

Decide the handoff before the first call goes out

Every good setup has a written list of calls the agent must not finish alone. Complaints, credit questions and anything complex go to a person, with the call summary attached, so the caller does not repeat themselves. Write down who that person is during office hours and who it is after hours.

Agree the other guardrails at the same time. If a caller seriously asks whether they are talking to a human, the agent does not claim to be one; it hands over. An unhappy “don’t call me” goes on the do-not-call list before the call ends. Calling rules on consent, timing and registered preferences apply in India too, and they change. Confirm what applies to your campaigns with your legal adviser before you dial a list.

Indian callers switch languages, so test in theirs

A retailer in Pune may start in Marathi, answer a price in Hindi and say the SKU name in English. The agent supports English, Hindi, Hinglish and Marathi, with more on request, and mirrors the caller mid-call without switching just because a city name was Hindi. Our notes on AI voice agents in Hindi on sales calls go deeper.

The setup step here is practical. Pick ten real outlets or leads from different regions and run test calls in the language each one actually uses. Listen for SKU names, local brand nicknames and pack sizes. Add any word the agent struggles with to the knowledge base before the campaign, not after a complaint.

Also decide where the system runs. Options include managed cloud, your own cloud account or on-premises, with India data residency when your policy requires it. Settle this with IT before the first test call so it does not stall the launch.

Measure from the first call, or you will argue later

Every call ends with a transcript, summary, outcome and next step, and is scored on price accuracy, script, compliance and listening. A daily digest on WhatsApp and email lists orders, conversions, spend and the calls worth a listen. Agree before launch which two numbers decide whether the pilot worked.

Here is an illustrative way to size the first campaign. Say a distributor runs 6 beats with 40 outlets each, and on an average month each rep misses 2 working days. That is 6 × 2 = 12 empty beat-days a month, and 12 × 40 = 480 outlet calls that nobody makes. If half of those outlets would normally place an order, the team is missing 480 × 0.5 = 240 orders a month. These numbers are made up for the arithmetic. Replace them with your own attendance sheet and beat sizes, and you have a baseline the agent has to beat.

“Our retailers will not talk to a machine” deserves a fair test

This is the objection we hear most, and it is reasonable. Retailers value a relationship with their rep. Some will hang up on any unfamiliar caller, human or not. Others dislike calls in the middle of a rush hour.

We do not think the answer is to argue. The answer is a small, honest test: the absent-beat outlets that would otherwise get no call at all. There, the comparison is not agent versus rep. It is agent versus silence. If those outlets order, you have your evidence. If they do not, the call scores and transcripts show why, and you can fix the script or the timing. The rep still owns the relationship; the agent covers the days they cannot.

For a fuller view of where people should stay on the phone, see AI calling vs human telecalling.

What to do this week

  1. Write the one-sentence job for your first agent and choose absent beats or after-hours enquiries.
  2. Pull the current price list, minimums and live schemes into one file, and name its owner.
  3. Record an hour with your best salesperson and turn their answers into the script.
  4. Write the handoff list and the after-hours contact, and confirm calling rules with your legal adviser.
  5. Size the gap with your own attendance sheet using the arithmetic above.

None of this needs a long project. It needs one owner and an honest afternoon with your own data. When that is ready, our calling guides and playbooks cover each campaign in more detail.

ShareLinkedInXWhatsApp

Go deeper: AI Voice Agent for Sales: What It Does on a Call · AI Order-Taking Calls to Retailers: How They Work · After-Hours Lead Response With an AI Calling Agent · Covering an Absent Sales Rep’s Beat With AI Calls · all AI Calling guides

Questions

How do you set up an AI calling agent in India?

Pick one job, prepare your price list, minimums and schemes, write the script from your best salesperson's answers, and agree handoff rules before the first call. Then test in the languages your callers actually use.

What do you need before an AI calling agent goes live?

A clean price list with an owner, a short script, a list of calls that must go to a person, a do-not-call process and two numbers that decide whether the pilot worked.

Can an AI calling agent in India handle Hindi and Marathi?

AI Calling speaks English, Hindi, Hinglish and Marathi, with more on request, and mirrors the caller's language mid-call.

See it on your own data. AI Calling — Voice agents that take orders. Book a 30-minute working session with an engineer.