Most voice-agent ROI calculators are theatre. They ask three vague questions, multiply some optimistic numbers, and produce a "five-figure annual savings" pitch designed to get you to a sales call. Here are the six inputs that actually determine whether your voice agent pays back in 90 days or never — with the maths anyone can do on the back of an envelope.
The honest model: six inputs, one equation
Strip away the marketing and a voice agent's ROI comes down to a single equation:
(Calls handled per month × value per call) − ongoing run-cost = monthly contribution
Monthly contribution × 12 ÷ build cost = first-year ROI
"Value per call" is where most calculators cheat. They quote the full revenue of a converted customer as if every call converts. Reality is messier. Here are the six numbers you actually need:
1. Inbound call volume (calls/month)
How many calls hit your business in a typical month? Count both answered and missed. Your phone system or CRM should give you the number — and if it doesn't, that's reason enough to fix the visibility before you fix the volume.
2. Missed-call rate (%)
Of those calls, what percentage go to voicemail or get dropped? Most AU SMBs we work with underestimate this by 30–60%. After-hours, weekend, and "everyone's on another call" volume adds up fast. If you don't know your real number, assume 25% as a starting point — it's typically conservative.
3. Conversion rate on a connected call (%)
Of the calls a human (or AI) actually answers, what percentage become a customer, booking, or qualified lead? For inbound enquiries with intent (someone calling about a service they actively want), 25–50% is normal. For cold inbound (general enquiries), 10–20%.
4. Customer lifetime value (CLV)
The total value of a customer over their relationship with you. For a one-off purchase business, this is average order value. For a services business with repeat work or retainers, multiply average annual value by typical relationship years. Don't use revenue here — use margin, after the cost of delivering the service.
5. The AI agent's effective conversion rate (%)
This is where calculators lie most. AI agents don't convert at human-level on every call — they convert at 50–80% of human level on calls they're well-designed for, and worse on edge cases. Start with 65% of your human-conversion baseline as a realistic planning number.
6. Build cost and ongoing run-cost
Build cost is one-off ($7.5K–$25K for typical AU voice agent builds). Ongoing cost is the LLM, telephony, and any platform fees — usually $150–$900/month at SMB volumes.
Worked example
Take a 25-person Australian professional services firm. Real numbers (anonymised) from one of our recent builds:
- Inbound calls/month: 320
- Missed-call rate: 28% (90 missed calls/month — mostly after-hours and weekends)
- Human conversion rate (when they connect): 38%
- CLV (margin): $4,200 per new client
- AI agent effective conversion (65% of human): 24.7%
- Build cost: $18,500 (voice + chat, CRM hand-off, AU-English voice)
- Run cost: $480/month
The maths
Calls recovered by the AI per month: 90 × (assume it handles 80% of those that would otherwise miss) = 72 connected calls.
Calls converted: 72 × 24.7% = 17.8 new clients/month.
Margin recovered: 17.8 × $4,200 = $74,760/month.
Net of run cost: $74,760 − $480 = $74,280/month.
Payback period: $18,500 ÷ $74,280 = 0.25 months — about 8 days.
First-year ROI: ($74,280 × 12) ÷ $18,500 = 48× first-year return on the build.
Those numbers look insane. They are also broadly accurate for this client. The reason they're not the norm is most businesses don't have this much after-hours volume going to voicemail. Adjust the volume down, and the maths still works — at a third of the volume, the ROI is still 16×.
Where this maths breaks
If your customer lifetime value is genuinely low (sub-$200 per customer), or your conversion is in single digits even for warm inbound, or your call volume is under 50/month, a voice agent probably doesn't pay back. We'll tell you so on a discovery call — and recommend a workflow automation or chat-only build instead.
Sensitivity analysis — what changes the answer
| If you change… | By | Effect on payback |
|---|---|---|
| Call volume | +25% | Payback shortens proportionally |
| Missed-call rate | 30% → 50% | Recovered revenue jumps 67% |
| AI conversion vs human | 65% → 80% | Recovered revenue jumps 23% |
| CLV | 2× higher | Payback halves |
| Build cost | $18K → $30K | Payback extends by ~$12K worth of conversion |
The two inputs that most affect the answer are missed-call rate and CLV. If you only have time to measure two things accurately before a discovery call, measure those.
The build/buy/integrate options
Off-the-shelf (cheapest, fastest, weakest fit)
A no-code voice agent platform like Synthflow or Bland can be running in a day for $50–$300/month. They work for simple use cases. They fail when you need real CRM hand-off, custom routing logic, or anything beyond their default flows.
Pick this when: you want to test the idea before committing to a custom build, OR your use case is genuinely simple (FAQ answering, after-hours message-taking only).
Custom build (recommended for most SMBs)
$7.5K–$25K one-off, $150–$900/month ongoing. Tuned to your business, your tone of voice, your CRM, your routing rules. Owned by you — code and prompts handed over at the end.
Pick this when: the maths above gives you a payback under 6 months and you want a solution that compounds (every conversation you capture is data that improves the next one).
Enterprise platform (overkill for most SMBs)
Genesys, NICE CXone, Five9 with AI add-ons. $50K+ implementations, $5K+/month run rate. Built for contact centres, not 25-person firms.
Pick this when: you're a contact centre — not before.
When voice agents don't make sense
Honesty matters. Voice agents are wrong for:
- Complex sales requiring relationship-building. Enterprise sales, financial advice, anything where the call IS the relationship.
- Sensitive intake (mental health crisis lines, abuse reporting). Use AI to triage written enquiries, not to talk to people in crisis.
- Industries where AI disclosure damages trust. Some sectors — bespoke luxury, exclusive professional services — still benefit from "you only ever speak to a partner". Run a survey before assuming.
- Below ~50 calls/month. The fixed cost of building is too high relative to the volume of conversations to optimise.
A free spreadsheet (if you want to model your own)
The model above runs in Excel or Sheets in about 15 minutes. Six input cells, three calculation cells, one output. We'll share a free template on request — or run the numbers for you for free during a discovery call.
Key takeaways
- Six inputs determine voice-agent ROI: call volume, missed-call rate, human conversion rate, CLV, AI's effective conversion rate, and build/run cost.
- Most calculators inflate ROI by assuming AI converts as well as humans (it doesn't — assume 60–80% of human baseline).
- The two highest-leverage inputs are missed-call rate and CLV. Measure them before you spec the build.
- For most AU SMBs with 200+ calls/month and CLV over $1,500, voice agents pay back in under 90 days.
- Off-the-shelf tools are fine for testing the idea; custom builds compound value over time.
- Below 50 calls/month, build something else first.
Want help putting any of this into practice?
30-minute discovery call. We'll map your highest-leverage AI opportunities to revenue or time saved — no obligation.