Call centre ROI calculator

How this is calculated

A fully-loaded agent costs a fixed amount per day whatever happens. Outbound divides that day by connected conversations, so every unanswered dial loads into the ones that connect. Inbound divides it by calls handled, and every inbound call connects by definition. The AI side pays a per-minute rate on connected minutes plus the full human cost of every conversation it escalates.

Two modes, because outbound and inbound unit costs are not the same quantity and comparing them directly is how ROI numbers end up meaningless:

OUTBOUND — denominator is connected conversations
costPerDial     = repDayCost / dialsPerRepDay
costPerConnect  = costPerDial / connectRate
humanMonthly    = dials × costPerDial
occupancy       = (connects × handleMin) / (repDays × 480)

INBOUND — denominator is calls handled
callsPerRepDay  = (480 × occupancy) / handleMinutes
costPerCall     = repDayCost / callsPerRepDay
humanMonthly    = calls × costPerCall

BOTH — the AI side pays for minutes AND for what it hands back
aiMonthly       = (units × handleMin × aiRate) + (units × escalationRate × humanCostPerUnit)

Worked example

20,000 outbound dials a month at a 12% connect rate is 2,400 conversations. A fully-loaded $300 agent-day across 45 dials is $6.67 per dial, and because roughly eight dials are needed per connect, $55.56 per connected conversation — about $133,000 a month across 20 full-time people. With AI at $0.06 a connected minute and 15% of conversations escalated to a human, the same volume costs about $20,600: an annual difference near $1.35M, or 84% of the human cost.

Run the same 20,000 as inbound calls at a 6-minute handle time and 83% occupancy and each agent handles about 66 calls a shift, so a call costs $4.56 — roughly a twelfth of the outbound figure. Same agent, same day rate, different denominator.

Why escalation is the input that matters most

Set escalation to zero and this model reports a saving multiple above 200×. That is arithmetically correct — an outbound rep spends under a tenth of the paid day talking, and AI has no idle to pay for — and nobody should believe it. Real deployments hand some share of conversations to a person, and an escalated conversation costs a full human conversation on top of the AI minutes. At 15% the multiple lands near 6.5×; at 30%, near 3.3×. Those are numbers you can take to a finance review. The tool warns you when the multiple stops being plausible rather than quietly printing it.

What this excludes

Integration work, prompt and workflow design, QA, and the ramp before an agent performs well — all real, all excluded. This is a steady-state running-cost comparison, not a first-year business case. It also starts from the agent-day figure you enter, so it will read lower than published cost-per-call benchmarks that bundle technology, supervision and facilities. Load those into your agent-day cost if you want the comparison to be like for like.

The currency toggle picks a separate figure for each currency rather than converting one into the other, so the rupee model reflects Indian agent economics rather than a dollar figure at today's exchange rate.

Where the defaults come from

Every input above is editable — these are starting points, not claims about your operation. Benchmarks current as of July 2026.

Frequently asked questions

How do you calculate call centre ROI?

Take the fully-loaded cost of an agent-day, divide by the units that agent actually produces in a day, and compare against what the same volume costs with AI. The subtlety is the denominator: for outbound it is connected conversations, which is far fewer than dials; for inbound it is calls handled, which depends on handle time and occupancy. Getting the denominator wrong is what makes most ROI numbers meaningless.

Why is outbound cost per conversation so much higher than inbound cost per call?

Because of the denominator, not the rate. At a 12% connect rate roughly eight dials are paid for to produce one conversation, and a short outbound call cannot fill a shift the way an inbound queue does — outbound reps often spend under 10% of paid time actually talking. The same agent at the same day rate produces a figure around ten times higher on outbound. Both numbers are correct; they measure different things.

Why does this show a lower cost per call than the $6–12 figures I've seen?

Published inbound cost-per-call benchmarks include technology, supervision, facilities and overhead that a bare agent-day rate does not. This model starts from the agent-day you enter, so it lands lower unless you load those costs into the figure yourself. It also means that once the two bases are aligned, the outbound number in this tool is the conservative one.

What escalation rate should I use?

Whatever your own data says. The defaults are 15% for outbound and 25% for inbound, and the number matters enormously: at 0% escalation the model produces a saving multiple above 200×, which is arithmetically true and completely unbelievable. Modelling escalation honestly is the difference between a calculator and a brochure — every escalated conversation costs a full human conversation on top of the AI minutes.

Does this include implementation cost?

No. It compares running costs — agent-days against per-minute AI plus escalations. Integration work, prompt and workflow design, QA, and the ramp period before an agent performs well are all real and all excluded. Treat the output as the steady-state difference, not a first-year business case.

Where do the default numbers come from?

The agent-day cost, dials per rep-day and connect rate come from the same sourced model that powers our own cost pages, with each value's basis and vintage recorded. Handle time defaults to published contact-centre figures. Every one is editable, and the honest use of this tool is to replace them with your own payroll and CRM data.

Can I switch to rupees?

Yes — the currency toggle switches between USD and INR. It picks a separate figure for each currency rather than applying an exchange rate, so the rupee model reflects Indian agent costs rather than a converted American one.

Related reading