Every support call your team handles carries a cost that most operations heads underestimate. The salary on the payroll is visible. The infrastructure, supervision, training, repeat calls, and idle time that sit beneath it are less visible. Together, they produce a cost-per-call figure that is significantly higher than most finance teams realize and almost entirely reducible for the call types that make up the majority of e-commerce support volume.
Let’s breaks down the real components of cost-per-call, shows where AI voice bots change the model, and walks through a sample calculation that illustrates what the reduction looks like in practice.
What Goes Into Cost-Per-Call
Cost-per-call is not just agent salary divided by calls handled. It is a blended figure that includes every input required to produce a handled call.
For a human agent model, the components are:
- Agent compensation: Salary, PF, incentives, and any variable pay. For a mid-level support agent in an Indian e-commerce operation, this typically runs ₹20,000 to ₹35,000 per month.
- Training and onboarding: A new agent takes 2 to 4 weeks to reach full productivity. During that period, they handle fewer calls, make more errors, and require supervisor time. The cost of this period is real and recurring, every time attrition forces a replacement.
- Supervision and quality: For every 10 to 15 agents, you carry a team lead or supervisor. Their time is a cost that spreads across every call the team handles.
- Infrastructure: Telephony, CRM access, workstation, internet, and seat costs. These are fixed regardless of call volume.
- Idle time: Agents are paid for shifts, not calls. During low-volume windows, lunch breaks, and between-call wrap-up time, the cost continues. Idle time typically accounts for 20 to 35% of total agent hours in e-commerce support operations.
- Repeat contacts: According to a 2025 cross-industry analysis, the real cost per issue is often 2.3 times the cost-per-contact benchmark, because many issues require multiple contacts to resolve. In e-commerce, repeat WISMO calls and unresolved delivery queries are common cost multipliers.
When all of these are factored in, a cost-per-call figure of ₹120 to ₹180 for an in-house Indian e-commerce support operation is a realistic working range for manually handled calls. This is the number a voice bot model is being compared against.
How AI Voice Bots Change the Cost Structure
An AI voice bot does not eliminate cost. It replaces a high fixed-cost model with a low variable-cost model. The components look like this.
- Platform and integration cost: There is a fixed cost for the voice bot platform, OMS integration, and CRM connectivity. This spreads across call volume. At 10,000 calls per month, the platform cost per call is significantly lower than at 1,000.
- Escalation cost: Not every call stays with the bot. Calls that escalate to a human agent carry the full agent cost for that interaction. The escalation rate determines how much of the human cost model remains in the blended figure.
- No idle time, no training cost, no attrition cost: The bot does not have shift patterns or lunch breaks. It does not require onboarding time. It does not resign. These cost lines disappear from the model entirely for bot-handled calls.
The resulting cost-per-bot-call, at typical e-commerce automation rates, falls in the ₹40 to ₹60 range depending on call duration, platform costs, and escalation rate.
The Automation Rate Is the Key Variable
The single biggest lever in the cost reduction model is automation rate, the percentage of calls the bot handles without human escalation.
For e-commerce operations, the call types that drive the highest automation rates are:
- WISMO and order status calls: These are highly scriptable, require no judgment, and follow predictable flows. You can easily achieve automation rates of 80 to 90% for WISMO calls type with a well-configured bot.
- COD confirmation calls: Outbound COD confirmation is almost entirely automatable. The bot calls, the customer confirms or cancels, and the interaction ends. Escalation rates are low.
- Delivery update and availability confirmation calls: Standard delivery notifications and slot confirmations follow a fixed flow. Automation rates above 85% are typical.
- Return initiation and refund status calls: Standard return flows are scriptable. Complex return disputes require human handling, but they represent a minority of return-related contacts.
Call types that lower the overall automation rate include complex complaints, payment disputes, and anything requiring empathy or resolution authority. These are precisely the calls that should reach human agents. Freeing agents from the high-volume repetitive calls means they are available for these conversations without a queue.
Conclusion
Cost-per-call reduction through AI voice bots is not a theoretical benefit. It is a numbers exercise. The inputs are your call volume, your current fully loaded cost per call, your target automation rate, and your escalation handling cost. The output is a monthly saving figure that finance heads can evaluate against platform investment.
The ₹150 to ₹50 model is illustrative. Your number may be higher or lower. But for any e-commerce operation handling thousands of repetitive voice interactions every month, the direction of the reduction is consistent, and the levers are well understood.
FAQs
Cost-per-call is the fully loaded cost of handling one inbound or outbound support call. It includes agent compensation, supervision, training, infrastructure, idle time, and repeat contacts.
By replacing the high fixed-cost human agent model with a low variable-cost automated model for repetitive call types. The bot handles calls at a lower price than human agents. Platform costs spread across volume. Idle time, training, and attrition costs are eliminated. Escalations still carry agent cost, but at a much lower frequency.
Not necessarily or immediately. The more accurate outcome is redeployment. Agents move from handling repetitive queries to handling escalations, complex complaints, and high-value interactions. Over time, headcount planning becomes more precise because you are staffing for escalation volume rather than total call volume.
The economics become strong at around 5,000 or more calls per month, where platform costs spread sufficiently across interactions. At 10,000 to 20,000 calls per month, the monthly savings in a fully loaded cost model typically exceed platform investment by a significant margin.
