Voice-AI vendors love to lead with "save 80% on customer support costs." It's not exactly wrong, but it's the wrong frame for a small business. The reason small businesses deploy voice agents isn't to lay off support staff — most of them don't have a support team to lay off. It's to capture revenue that's leaking out the back of the phone line.
Here's a working model for what a voice agent actually delivers, including the costs nobody mentions in the sales deck.
The leak nobody measures: missed and abandoned calls
The single most under-tracked number in a small business is calls that didn't get answered. Voicemail. Hold timeouts. Calls outside hours. The receptionist on lunch. These don't show up on the P&L because there's no transaction to record. The customer just hangs up and goes to a competitor.
Industry data is consistent across studies: about 30% of inbound calls to small businesses go unanswered at some point in the customer journey, and roughly 85% of those callers don't try again. They call the next result on Google.
If you're a $1M revenue business and inbound phone calls drive even 40% of that, you're losing somewhere around $100K/year to unanswered calls. That's the number that actually matters.
The honest cost stack of a voice agent
Vendors love to quote you a single number. Real cost is layered:
Build cost (one-time)
For a custom voice agent on a single workflow (order intake, lead qualification, appointment booking), expect $5K–$15K one-time depending on integration complexity. Off-the-shelf voice tools are cheaper but don't integrate with your systems — you'll either pay anyway in lost functionality or pay the integration cost in addition.
Per-minute call cost (variable)
Three components:
- Telephony (Twilio or similar): ~$0.013/min inbound
- Speech-to-text + text-to-speech: ~$0.05/min combined for a quality voice (ElevenLabs class)
- LLM inference: ~$0.02–$0.05/min depending on model and prompt size
Fully loaded: $0.08–$0.15 per minute. A 4-minute order call costs you $0.32–$0.60. A 1-minute appointment booking costs $0.10. Compare to a $20/hr employee at ~$0.33/minute — and the agent doesn't take breaks.
Retainer (monthly)
For an agent that's actively maintained — monitored, tuned, updated when your menu/inventory/process changes — expect $500–$2,500/month. Lower if it's a stable workflow, higher if it changes often or carries critical revenue.
The hidden cost: prompt drift
Voice agents need prompt updates. Your menu changes. Your hours change. A new product launches. If nobody owns this, the agent gives stale info and the trust collapses. The fix is either an internal owner who knows the prompt format or a retainer that covers it. Don't run a voice agent without one.
A worked example: the after-hours catering line
Real numbers from a deployment I shipped recently. Restaurant with $850K/yr revenue, ~$120K of which is catering. Pattern: catering inquiries come in by phone, often after hours, and 60%+ went to voicemail. Of those voicemails, the team estimated they were closing maybe 1 in 5.
Voice agent deployment:
- Build cost: $7,500 one-time
- Retainer: $800/month
- Call volume: ~120 inbound calls/month routed to the agent (mostly after-hours)
- Average call: 3.2 minutes → ~$0.35 per call → ~$42/month in usage
- Total monthly run cost: ~$842
Lead capture went from ~40% (voicemail+callback success) to ~95% (agent captures structured lead and emails it within seconds). Of those captured leads, the team's catering close rate stayed roughly the same (~25%).
Net: roughly 30 additional captured leads per month, of which ~7–8 close. Average catering ticket is around $400. That's ~$3,000/month in incremental revenue the business was previously leaking. Net of the $842 monthly cost, that's ~$2,150/month in net new gross profit, or about $26K/year. Build paid back in under 4 months.
The ROI math, generalized
If you want a quick gut check for your own business, here's the formula:
incremental annual GP = (missed calls/year) × (capture lift %) × (close rate) × (average revenue per closed call) × (gross margin)
Then subtract:
annual cost = build / 12 months amortized + retainer × 12 + per-minute cost × call volume × avg duration × 12
If the first number is at least 3× the second, the build is a clear yes. If it's between 1.5–3×, it's a yes but you should sanity-check the assumptions. If it's under 1.5×, voice agent isn't your highest-leverage spend right now.
Where voice agents don't pay back
Honest about the negative cases. Voice agents underperform when:
- Call volume is too low. Under ~40 calls/month, the build cost takes too long to amortize. You're better off with a $30/month answering service.
- The conversation is genuinely complex. Multi-step troubleshooting, regulated industries (legal, medical specifics), high-stakes negotiations. Agents help here but don't fully replace the human.
- Your callers really want a human voice. Some demographics, some industries (luxury, white-glove). Test, don't assume — but don't ignore the possibility.
- Your systems can't be integrated. If your POS or CRM has no API and you can't change vendors, the agent can talk but can't act. That's a chatbot at best.
The metric that actually matters
Most ROI conversations focus on cost savings. For small businesses, the bigger number is almost always capture rate — what percentage of inbound demand actually becomes a logged interaction your team can act on.
Pre-agent: capture rate is whatever your phone-answering coverage gets you. Often 60–70%.
Post-agent: capture rate is 95%+, with structured data, every time, with a transcript.
That delta is where the ROI lives. The cost savings on labor are a bonus.
Want to model your own?
The home page has an ROI calculator that takes a few inputs and gives you a back-of-envelope. Or book a 30-minute call and we'll do it together with your real numbers.
Either way: don't take the vendor decks at face value. The honest math is more interesting than the marketing math, and usually still works.