How to choose an AI calling provider: 10 questions to ask before you sign

Before you sign with any AI calling provider, get written answers to ten questions: how minutes are counted, what happens at your usage cap, what setup involves, which languages the agent speaks, what compliance tooling exists, what you can see during calls, how your data is retained and deleted, the contract terms, how scaling works, and whether you can hear a live call first. Choosing carefully matters: 80% of organisations already use some form of voice agent, yet only 21% are very satisfied with their current solution, per Deepgram and Opus Research's State of Voice AI 2025 survey of 400 business leaders.

This guide sets out what a good answer looks like for each question, so it is useful whoever you shortlist. 6 Seven Labs builds AI voice calling agents for Singapore businesses; after each benchmark we add one line on how we answer the same question, so you can hold us to our own standard.

Why does your choice of AI calling provider matter?

Because the market is growing far faster than buyer satisfaction. Conversational AI was a USD 13.2 billion market in 2024 and is projected to reach USD 49.9 billion by 2030, a 24.9% compound annual growth rate, per MarketsandMarkets (April 2024). Markets that grow this fast attract vendors of every quality, and a glossy demo can hide a very ordinary product.

Adoption has already reached small businesses: 68% of US small businesses now use AI regularly, up from 48% in July 2024, per an Intuit QuickBooks survey of more than 2,200 US small businesses (April 2025). Yet when Deepgram and Opus Research surveyed 400 business leaders for their State of Voice AI 2025 report, 80% of organisations were using some form of voice agent while only 21% were very satisfied with their current solution. Most buyers would not buy the same thing twice.

The long-term case is strong, which is exactly why selection deserves rigour now. Gartner (March 2025) predicts agentic AI will autonomously resolve 80% of common customer service issues by 2029, with about 30% lower operational costs. Today, a Gartner survey of 5,728 customers (August 2024) found only 14% of issues are fully resolved in self-service. The ten questions below tell you where a vendor sits between those two numbers.

The 10 questions at a glance

The table below is the shortlist version. The sections that follow take each question in order: what a good answer looks like from any provider, then how 6 Seven Labs answers it.

#	Question	A good answer looks like
1	How are minutes counted?	Per second, connected calls only
2	What happens at my cap?	Calling pauses; you choose a top-up or upgrade
3	What does setup involve?	Done for you, live in days, tested before launch
4	Which languages, how naturally?	Your market's languages, proven in a live demo
5	Compliance tooling, and who owns it?	DNC list, stop requests, audit trail; you own the list
6	What do I see during calls?	Live dashboard, recordings, transcripts
7	What happens to my data?	Named retention window, then permanent deletion
8	What are the contract terms?	Month to month, cancel anytime
9	How do I scale to more agents?	Per-agent plans, one subscription, prorated additions
10	Can I hear a live call first?	Yes, unconditionally, before any signature

What should you ask about pricing, setup and languages?

One benchmark first: business-grade AI voice agents typically run USD 0.10 to 1.50 per minute all-in, per a March 2026 industry comparison of published vendor pricing. Where a quote sits in that range matters less than how the meter runs, which is what the first two questions expose. For the full human-versus-AI arithmetic, see our cost guide, AI callers vs hiring a telemarketer in Singapore.

How are call minutes counted?

A good answer is precise: billing per second, and only on connected calls. The traps are rounding every call up to the next full minute and billing for unanswered dials, ring time or voicemail drops. On a list where half the dials go unanswered, those habits can double your effective rate without changing a single line of the price sheet. 6 Seven Labs bills per second on connected calls only; an unanswered dial costs nothing.

What happens when I hit my cap?

A good answer has three parts: you are warned as usage climbs, calling pauses when the included minutes are used up, and continuing is a decision you make at a published price. The bad answer is silent overage, where the dialler keeps going and the surprise arrives on the invoice. At 6 Seven Labs, calling pauses at the cap; you choose a top-up block at S$150 per 100 minutes or an immediate prorated upgrade, with current rates on our pricing page.

What does setup involve, and how long does it take?

Insist on specifics: who writes the script, who selects and tests the voice, who sets the calling rules, and how long until the first live call. The answer should be days to a few weeks, not months. Hamming's analysis of more than 4 million production voice-agent calls (January 2026) found containment typically starts at 40 to 60% in month one and matures to 75 to 85% by month six and beyond. A vendor with no tuning plan is selling you month-one performance forever. 6 Seven Labs setup is done for you in days: script, voice, calling rules and testing, all before the first live call.

Which languages, and how naturally?

You want the languages your market actually speaks, proven live rather than with a studio sample. Ask which languages are production-ready today versus merely supported, and listen for natural intonation and turn-taking in each. In a multilingual market like Singapore, an agent that only handles English only covers part of your list. Our own agents speak English, Mandarin and Malay naturally today; the enterprise-grade voice platform underneath supports 20+ languages.

What should you ask about compliance, data and proof?

The remaining six questions cover what goes wrong after the honeymoon: compliance, visibility, data, lock-in, scale and proof.

What compliance tooling exists, and who owns list compliance?

A good answer names four tools: an automatic do-not-call list, stop requests that are captured and honoured, control over calling hours, and a full audit trail of recordings and transcripts. Just as important, an honest vendor tells you plainly that legal responsibility for a clean list stays with you, the list owner. Our guide to PDPA and DNC compliance for outbound calling in Singapore covers the rules in depth. 6 Seven Labs provides all four tools, and our FAQ states plainly that the client owns list compliance.

What do I see while calls are happening?

Listen for a live dashboard, not a weekly report: every call with its outcome, recording and transcript, plus usage you can check at any moment. Visibility turns a campaign from a black box into something you can steer mid-week; our guide to outbound calling metrics that matter covers which numbers to watch. 6 Seven Labs clients get a live portal with calls, recordings, transcripts, a leads board, usage bars and self-serve billing.

How is my data retained and deleted after cancellation?

The right answer names a retention window, commits to permanent deletion after it, and leaves you time to export first. Customer concern here is not a niche worry: 71% of US adults think increased AI use will make personal information less secure, per Pew Research Center (February 2026), and Singapore's PDPA imposes its own retention and protection obligations. Call recordings are personal data, and a vendor with no deletion story is a risk you would be choosing on purpose. At 6 Seven Labs, operational data is retained for 30 days after cancellation and then permanently deleted, with invoices kept as legally required.

What are the contract terms?

A good answer, in a market this young, is month-to-month with cancellation from your own account and access kept to the end of the paid period. A 12-month lock-in shifts the risk of a disappointing product onto you; the satisfaction figures above are the argument for keeping the exit open. 6 Seven Labs is month-to-month with no lock-in: cancel anytime from the portal and keep full access to the period end.

What does scaling to more agents look like?

A good answer prices additional agents clearly, keeps everything on one subscription with one invoice, and prorates mid-cycle additions. Ask before you need it; renegotiating while a campaign waits is how bad terms get signed. At 6 Seven Labs, plans are per agent on one subscription, and an agent added mid-cycle is prorated for the remainder of the period.

Can I hear a live call before signing?

A good answer is yes, unconditionally. A live call exposes what no deck can: the pause before the first word, how the agent handles an interruption, what happens when someone objects or asks whether they are speaking to an AI. If a vendor will only play polished recordings, ask why. For a walkthrough of a strong call end to end, read how AI appointment setting actually works. 6 Seven Labs demos are live: book a 20-minute demo and hear an agent take a real call.

Frequently asked questions

How much does an AI calling agent cost per minute?

Business-grade AI voice agents typically run USD 0.10 to 1.50 per minute all-in, per a GetVoIP comparison from March 2026. Plans usually bundle minutes into a monthly fee, so divide the plan price by included minutes to compare vendors, and confirm whether unanswered dials are billed.

What is a good containment rate for a new voice agent?

Hamming's analysis of more than 4 million production voice-agent calls, published January 2026, found containment typically runs 40 to 60 percent in month one and matures to 75 to 85 percent by month six. Treat month-one numbers as a starting point and ask who does the ongoing tuning.

Should I sign a long contract with an AI calling provider?

Avoid long lock-ins while the market matures: only 21 percent of organisations are very satisfied with their current voice agents, per Deepgram and Opus Research in 2025. Month-to-month terms let you leave if quality disappoints. 6 Seven Labs is month-to-month, cancel anytime, with access kept to the period end.

Who is responsible for DNC compliance when using an AI calling provider?

The list owner, meaning your business, keeps legal responsibility for calling lists in Singapore and most other jurisdictions. A good provider supplies tooling that makes the duty easier: an automatic do-not-call list, honoured stop requests, calling-hours control and a full audit trail of recordings and transcripts.

How long should AI calling agent setup take?

Days to a few weeks for a done-for-you build, covering the script, voice, calling rules and testing. Be wary of quotes measured in months for a single use case, and of instant self-serve launches with no testing, because untested agents are where poor first impressions come from.