Most agencies selling voice automation treat NPS as an afterthought. The platform answers calls, books the appointment, job done. But your SMB clients care about one question above everything else: did that AI make my customer happier or angrier? If you can't measure satisfaction on AI-handled calls, you're flying blind when retention renewals come around.

We built post-call NPS into VoxReach six months ago. Since then we've collected just over 1,000 completed surveys across medical practices, trade businesses, and service providers. The patterns surprised us. Here's what we learned and how agencies can use NPS data to hold onto clients longer.

How we collect NPS without annoying callers

The survey fires 90 seconds after the AI agent ends the call. One SMS: "Quick question - how likely are you to recommend [business name] after that call? Reply 0-10." No app download, no link, no login. Plain text reply.

Response rate averages 11.4 per cent across all industries. That's low compared to email NPS campaigns, but these are inbound callers who just spoke to a robot. The fact that one in nine bothers to reply tells you something: people have opinions about AI calls.

We ask one follow-up if the score is six or below: "What went wrong?" Free text, optional. About half the detractors answer. Promoters (nine or ten) get a thank-you message and nothing else.

Response rates vary wildly by vertical

Medical and allied health lead at 16.2 per cent response rate. Patients expect follow-up texts from clinics, so one more message doesn't feel intrusive. Trades and home services sit at 9.1 per cent. Retail and hospitality bottom out at 6.8 per cent.

The lesson for agencies: if you're pitching a cafe or a boutique, don't promise robust NPS data. You'll get some signal, but not enough for monthly reporting. Medical practices, physio clinics, dental offices - those verticals give you clean feedback loops your client can actually act on.

One podiatry practice in Newcastle hits 22 per cent response rate because the principal mentions the survey during the AI call: "You'll get a quick text after this - let us know how the booking went." Priming works.

What scores actually look like

Median NPS across the full dataset is +38. That's the net score: percentage of promoters minus percentage of detractors. For context, live reception teams in similar industries typically score +45 to +55 based on widely-reported customer service benchmarks, so AI is close but not quite at human parity yet.

The distribution breaks down like this:

• Promoters (9-10): 52 per cent
• Passives (7-8): 28 per cent
• Detractors (0-6): 20 per cent

Twenty per cent detractors sounds bad until you read the comments. Half the low scores aren't about the AI - they're complaints about pricing, wait times, or past service issues. The call we listened to last Tuesday scored a two, and the caller's comment was "Too expensive and I had to wait three weeks." The AI agent booked the appointment perfectly. The customer just hated the business model.

Agencies need to set this expectation up front. NPS measures overall sentiment, not AI performance. A detractor score might mean your client needs to fix their pricing or hire more staff, not that the voice agent failed.

High NPS correlates with twelve-month retention

This is the number that matters for your agency's trail commissions. We tracked 87 VoxReach accounts that have been live for at least a year. Businesses with a rolling 90-day NPS above +50 have a 91 per cent retention rate at month twelve. Accounts below +20 churn at 41 per cent.

The mechanism is obvious: happy callers become repeat customers, repeat customers keep the business growing, growing businesses renew automation contracts. Low NPS means the AI is either poorly configured or the business itself has problems. Either way, that account is at risk.

White-label agencies should build quarterly NPS reviews into every retainer. Pull the data, flag detractor comments, propose script tweaks or escalation rule changes. That fifteen-minute call every quarter is the difference between passive income and a cancelled contract.

What to do with this

If you're selling VoxReach to clients, turn on post-call NPS from day one. Set the survey delay to 90 seconds and keep the message under 160 characters so it fits in one SMS.

Pull the report monthly. If NPS drops below +30 two months in a row, book a script review. Low scores usually trace back to one of three things: the AI can't handle a common edge case, hold music is too long, or the business changed their booking policy and forgot to update the agent.

Use promoter scores in your own marketing. A +55 NPS after six months is a case study. A testimonial from a business owner saying "our AI receptionist scores higher than our old front desk" is worth more than any feature list.

NPS won't tell you if the AI can parse complex questions or transfer calls cleanly. But it will tell you whether your client's customers are happy. That's the metric that determines whether you still have a client in twelve months.

Sign up at app.voxreach.com.au/signup and see the NPS dashboard yourself. calls on your own number, surveys included.