Buying guide

How to choose an AI voice agent vendor: 12 questions that expose weak products.

Every vendor demo sounds great. The agent answers fast, the voice sounds human, the dashboard looks clean. Then it goes live on your phone line and meets a real caller with a real objection at a real 9 PM. This guide gives you the 12 questions that separate products built for demos from products built for your business — with the good answer and the red flag for each.

Published June 9, 2026 · 12 min read

We build custom AI voice agents for a living, so read this knowing where we sit. But most of the businesses that call us have already bought one voice agent that didn't work. The pattern is consistent: they evaluated the demo, not the product. The demo is the vendor's best 90 seconds. The product is what happens in month four when your scheduling software pushes an API update and your agent silently stops booking.

The stakes are not small. The average home-service business loses $50,000–$60,000 a year to missed calls. A voice agent that half-works doesn't recover that — it just adds a monthly fee on top of the loss. Run your own numbers in our missed-call calculator before you talk to any vendor, including us.

One more honest note before the questions. If you're under $500K in revenue, most of this guide is overkill — pick a cheap SaaS tool, accept its limits, and move on. The questions below matter most for businesses doing $1M–$20M — the HVAC and plumbing companies, dental practices, and law firms where a broken phone line costs five figures a month.

The one stat to remember

85% of callers who hit voicemail never call back. A voice agent that answers but can't book is functionally voicemail with better manners. Every question in this guide exists to find out whether the agent actually finishes the job.

Questions 1–3

Start with the demo. Most vendors fail here.

1. Can I hear it handle my price objections before I sign?

Why it matters. Generic demos use generic callers. Your callers say things like "the company down the road quoted me $89 for the same tune-up" and "my insurance should cover this." If the agent has never heard your objections, it will improvise — and improvised pricing answers cost you jobs or margin.

A good answer: "Give us your service list and your three most common objections. We'll configure a pilot this week and you can call it yourself." Then the pilot actually handles them.

A red flag: "The demo line shows you the core experience." Translation: configuration is your problem after the contract is signed.

2. Was this trained on templates, or will it be trained on my actual calls?

Why it matters. Template agents know what an average HVAC company sounds like. They don't know that you charge a trip fee north of the river, that Tuesdays are install days, or that your owner's mother-in-law calls weekly and should never be booked as a lead. The gap between average and yours is where bookings die.

A good answer: a specific training process — call recordings reviewed, service menu mapped, edge cases documented, a human signing off before launch.

A red flag: "Our AI learns on its own over time." Unsupervised learning on your live customers means your callers are the training data and your revenue is the tuition.

3. What are your containment and booking rates — and how do you measure them?

Why it matters. "Answers 100% of calls" is a vanity metric. Voicemail also answers 100% of calls. What matters is how many calls the agent resolves without human rescue (containment) and how many bookable callers actually end up on the calendar (booking rate).

A good answer: real numbers with definitions — "70–85% containment depending on industry, measured as calls resolved without transfer or callback; here's a sample report."

A red flag: no numbers, or numbers with no definition behind them. If a vendor can't define the metric, they aren't tracking it.

Questions 4–6

Then ask who does the work after launch.

4. Who fixes it when my FSM or CRM API changes?

Why it matters. This is the question that exposes the most weak products. ServiceTitan, Jobber, Housecall Pro, GoHighLevel, Dentrix, Clio — all of them push API updates. When that happens, the integration between your agent and your calendar breaks quietly. Calls still get answered. Bookings stop landing. You find out a week later when your board is empty.

A good answer: "We monitor the integration, we get alerted before you notice, and a named engineer fixes it — usually same-day." Monitoring is the key word.

A red flag: "Our integrations run through Zapier — you can update the Zap." That sentence means you just hired yourself as the integration engineer, unpaid, on call.

5. Who reviews the calls it gets wrong, and how often?

Why it matters. Every voice agent gets calls wrong. The difference between a good system and a liability is whether anyone notices. An agent that confidently quotes the wrong price 40 times before someone catches it has cost you more than it saved.

A good answer: a recurring human review — "we sample calls weekly, flag failures, and push fixes." Ask to see what a tuning changelog looks like.

A red flag: "You have full access to all your recordings in the dashboard." That's not review. That's homework, assigned to you.

6. What happens when I add a service, change a price, or open a second location?

Why it matters. Your business will change during the contract. Spring promo pricing. A new technician on the dispatch list. A second office with different hours. If updating the agent requires you to log into a prompt editor and get it right yourself, updates won't happen — and the agent will drift out of sync with reality.

A good answer: "Send us the change, it's live within 24 hours, included in your fee."

A red flag: "Changes are self-serve in the portal" with a straight face, or worse — per-change professional services fees.

Want these questions answered about our own product?

Fair. We build and operate custom voice agents on a monthly retainer, and we'll answer all 12 of these on a 30-minute call — including the ones that don't flatter us.

Questions 7–9

Ask about the moments that lose customers.

7. What happens after hours — really?

Why it matters. After-hours is where the money is. The burst pipe at 11 PM, the broken crown on Saturday, the no-heat call on the coldest night of January. These are your highest-ticket calls — and the reason most owners buy a voice agent at all. Some products quietly degrade at night: no live transfer, bookings queue until morning, emergencies get a "we'll call you back."

A good answer: the same behavior at 3 AM as at 3 PM — emergency keywords trigger a hot transfer to your on-call person, routine calls book the next slot, and the caller gets an SMS confirmation.

A red flag: "After hours, the agent takes a detailed message." A detailed message is voicemail. See the 85% stat above, then read what missed calls actually cost.

8. How does it hand off to a human, and how fast?

Why it matters. No agent should handle 100% of calls. Angry customers, complex quotes, legal intake with conflict checks — these need a person. The question is whether the handoff is a warm transfer with context, or a dead end that makes the caller repeat everything.

A good answer: configurable triggers (keywords, sentiment, caller requests a human), transfer in seconds, and a call summary pushed to your CRM so your person picks up mid-context.

A red flag: handoff means "the agent texts you a transcript and the caller waits." Or the vendor talks about containment rate as if transfers are failures to be minimized at all costs. An agent that refuses to give up a call is worse than one that gives up too easily.

9. Which integrations are native, and which run through Zapier?

Why it matters. "Integrates with 6,000+ apps" almost always means Zapier or Make in the middle. Middleware adds latency, a second monthly bill, and a second point of failure that nobody owns. For booking flows — where the agent must check real availability and write a real appointment — middleware is where things break.

A good answer: a short, specific list of native integrations including yours, and honesty about what isn't native.

A red flag: the integrations page is a wall of logos. Ask specifically: "Does the agent read live availability from my ServiceTitan calendar, or does it write to a form that someone processes later?" The second one is a lead-capture widget wearing a voice agent costume.

Questions 10–12

Finally, ask about money and ownership.

10. What is the real all-in monthly cost?

Why it matters. The list price is rarely the real price. SaaS voice agents commonly bill base fee plus per-minute charges plus SMS fees plus a higher tier for the integration you actually need plus overage when call volume spikes — which is exactly when you need it most. A $299 list price routinely lands at $600–$800 in practice.

A good answer: a flat number with everything itemized, and a straight answer to "what does my bill look like in my busiest month?"

A red flag: "It depends on usage" with no worked example. We published our own math in what an AI voice agent costs — make every vendor show you theirs.

11. Who owns the prompts, the training data, and the phone number?

Why it matters. Six months in, your agent embodies real operational knowledge: your scripts, your objection handling, your call recordings, your tuned prompts. And the phone number callers dial may be the one on your trucks and your Google Business Profile. If the vendor owns all of it, leaving means starting from zero — which means you'll never leave, which is the point.

A good answer: "The number is yours or portable to you. Your recordings and data export on request. Prompt logic is documented and you get a copy."

A red flag: the contract is silent on all three. Silence in a contract always resolves in the vendor's favor.

12. What happens if we part ways?

Why it matters. Every vendor relationship ends eventually. The exit terms tell you how confident the vendor is in earning your renewal versus trapping it.

A good answer: a defined offboarding process — number ported, data exported, a transition window where the line keeps working. Vendors confident in their product make leaving easy.

A red flag: auto-renewing annual contracts with 90-day cancellation windows, "data export available for a fee," or — the worst — the phone number is non-portable. That last one isn't a product decision. It's a hostage strategy.

The vendor landscape

Who you'll actually be evaluating.

Most SMB owners end up comparing four categories. Home-services specialists like Avoca and Sameday integrate deeply with FSM platforms and know trade-business call flows well; their limits show up when your business doesn't match their template. We've broken down both — see our Avoca alternatives and Sameday alternatives pages for the specifics.

Budget-tier tools like ServiceAgent target smaller trade businesses at lower price points — a reasonable entry if you're below the revenue where edge cases get expensive (our ServiceAgent alternatives breakdown covers the trade-offs). DIY platforms like Synthflow give you the building blocks to assemble your own agent; powerful if you have technical staff, a part-time job if you don't (details in our Synthflow alternatives guide).

The fourth category is custom build-and-operate — what we do. Higher monthly cost, but the 12 questions above are answered by design: a named team trains the agent on your calls, monitors your integrations, reviews failures weekly, and writes the answers to questions 10–12 into the contract. If you're also weighing a human answering service against all of this, read AI receptionist vs answering service first.

Common questions

Questions buyers ask us about vendor selection.

What should I ask an AI voice agent vendor before signing?

If you only have time for four: ask to hear the agent handle your own price objections live, ask who fixes integrations when your CRM or FSM API changes, ask for the real all-in monthly cost including minutes and integration fees, and ask who owns the prompts, training data, and phone number if you leave. Weak products fail at least two of those four.

How much does an AI voice agent cost per month?

SaaS tools list at $99–$500/mo, but all-in cost with minutes, SMS, and integration tiers typically lands at $300–$800/mo. Custom build-and-operate agents run $800–$2,500/mo and include integration work, weekly tuning, and a human team. Full breakdown in our pricing guide.

Are Avoca, Sameday, ServiceAgent, and Synthflow good products?

They're legitimate products serving different segments. Avoca and Sameday specialize in home services with FSM integrations. ServiceAgent targets smaller trade businesses at a lower price. Synthflow is a DIY platform for people who want to build their own. None are wrong — each fits a different business. The 12 questions in this guide will tell you which fits yours.

What's the biggest red flag when evaluating vendors?

A vendor who will only demo a generic agent. If they can't or won't configure a pilot on your actual services before you sign, you're buying the demo, not the product. Close behind: ticket-only support, Zapier-only integrations, and contracts silent on data and number ownership.

Should I pick SaaS or custom?

Standard call flows, one location, and a willingness to be your own admin — SaaS is the cheaper path. Non-standard pricing, emergency routing, multiple integrations, or no time to troubleshoot — custom usually pays the difference back in recovered bookings. We wrote the long honest version in custom AI vs SaaS voice bots.

How do I test an agent before buying?

Call it and act like your hardest customer. Push back on price. Ask for an appointment at 9 PM Sunday. Interrupt it. Ask for a human. Then have the vendor configure a pilot on your actual services and run the same gauntlet. The gap between the demo and the pilot is the product.

Bring these 12 questions to every vendor. Including us.

Book a 30-minute call and we'll walk through them one by one for your business — and tell you straight if a $99 SaaS tool is actually the right answer for where you are. If a voice agent is only one piece of what you need, we'll show you how it fits alongside workflow automations and a custom dashboard in one monthly partnership.

Book a free pipeline audit See the system