Autonomous Sales Agents: A Buyer's Checklist

A buyer's checklist for the questions vendor-written guides leave out — ownership, liability, disclosure, and how the contract ends.

You are about to put an autonomous sales agent in front of your pipeline, and the demo went well. It always does. The demo is a controlled environment where the agent writes three good emails about a prospect the vendor pre-enriched, and nobody asks what happens in month four when your domain reputation is in the floor and you want to leave.

Most buyer's guides for this category are written by the vendors in it. They are not useless — they converge on a sensible list of capability questions, and you should ask those. But they all answer the same question: is this agent good? The questions that decide whether the purchase hurts you are different. They are about who owns what, who is liable when something breaks, what you are legally required to say, and what you keep when you leave.

This is that checklist. It is vendor-agnostic, and we run BOSRAI through it at the end on the same terms.

What the vendor guides already cover

Read three or four of these guides and the overlap is obvious. AiSDR's buyer guide and Digital Applied's landscape piece both walk through message quality, list building, reply handling, qualification logic, intent signals, reporting and escalation rules. Ask all of it. Test it live on the demo rather than taking the recording's word for it, and bring your own prospect list so the agent is enriching someone it has not seen.

Where the guides stop is the part that costs money later. Digital Applied's own piece names the gap: no discussion of who owns enriched contact data, no indemnification or SLA detail on deliverability, no coverage of compliance beyond a nod, and break clauses mentioned as a testing signal but never as contract terms you should read.

Is it actually an autonomous sales agent?

Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. The more useful number in that release is the other one: of the thousands of vendors marketing agentic AI, Gartner estimates roughly 130 have genuine agentic capability. The rest is what they call agent washing — existing chatbots and rule engines relabelled.

A two-minute test on the demo: ask to see a run where the agent changed its plan mid-sequence based on something it observed, and the log of why. A real agent has a decision trail you can inspect. A relabelled sequencer has a branch that fired because a field matched. Neither is disqualifying — a good sequencer with strong copy may be what you need — but know which one you are paying agent prices for.

The five questions the guides leave out

1. Who owns the sending infrastructure, and who eats the reputation damage?

Many platforms provision mailboxes and secondary domains for you, which is convenient right up until those domains are burned. Then you need to know whether they were yours. Google requires bulk senders — 5,000 or more messages a day to Gmail — to authenticate with SPF, DKIM and DMARC, and to keep the spam rate reported in Postmaster Tools below 0.30%, with a recommendation to stay under 0.10% for headroom. Cross that line and it is your sending identity that suffers, not the vendor's.

Ask: are the domains registered to me? Do I have DNS control? Do I get Postmaster Tools access, or do I see deliverability only through your dashboard? Is there any SLA or remedy if your sending behaviour damages a domain I own? Most answers to the last one are no — that is fine, but it should be a known no rather than a surprise. Our piece on cold email deliverability covers the mechanics if you want the technical side.

2. What leaves with you?

Enriched contacts, reply history, meeting notes, qualification scores, the sequences you built. Some of it came from the vendor's data layer and may be licensed rather than sold — which means it may not be yours to take. Ask for the export format in writing before you sign, not after you give notice.

Two follow-ups in the same breath: is my data used to train your models, and is there a clause I can point at? And if you are acquired, what happens to my agreement?

3. What are you required to disclose?

From 2 August 2026, Article 50 of the EU AI Act requires that AI systems interacting directly with people inform them they are dealing with an AI, unless that is obvious to a reasonably well-informed person, and that the disclosure comes at the latest at the first interaction. Providers of systems generating synthetic text, audio, image or video must also mark outputs in a machine-readable format. If you sell into the EU, that is now a live obligation on outbound that an AI agent writes and sends on your behalf.

Ask the vendor how the product handles it: is there a disclosure mechanism, is it configurable by region, and who is the provider versus the deployer in your arrangement. If the answer is that nobody has looked at it, you are the one holding the obligation. We wrote up the detail in EU AI Act Article 50 and sales agents. This is informational, not legal advice — get your own counsel on your own facts.

4. What are the consent rules on the channels that are not email?

Every guide assumes email. If the agent also sends on WhatsApp, the rules are stricter and they are the platform's, not the regulator's. The WhatsApp Business Messaging Policy is explicit: you may only contact people if they have given you their phone number and you have received opt-in permission from them. Cold-scraped numbers do not qualify. Billing matters too — Meta moved the platform to per-message pricing on 1 July 2025, with marketing, utility and authentication template categories priced differently, so a channel that looks cheap per contact can behave very differently at volume.

Ask: does your product enforce opt-in, or does it let me send anyway and leave the liability with me? What happens to my WhatsApp Business account if template quality ratings drop? Our WhatsApp Business API guide has the setup and policy detail.

5. How does this end?

The question a good demo makes you forget. Minimum term. Notice period. Whether the annual discount is really a twelve-month lock. Whether onboarding fees are refundable if the agent never produced anything. Whether there is a break clause and what triggers it.

Pricing transparency is a decent proxy for how this conversation will go. Artisan's pricing page lists Team, Scale and Enterprise tiers scoped to lead volume with no figures and a "talk to sales" on each. 11x publishes no pricing at all and routes everything through a demo request. Neither is wrong — enterprise sales works that way — but if a vendor will not put a number on a page, expect the exit terms to be negotiated rather than published too. We keep a running AI SDR pricing index for what is actually listed.

The checklist

Copy this into the notes you take on the call.

If a vendor answers twelve of fifteen clearly, that is a good sign. If they are fluent on capability and vague on ownership, liability and exit, the call has already told you the most important thing.

Running BOSRAI through the same checklist

Same terms, no exceptions. BOSRAI is a human-in-the-loop, WhatsApp-native sales agent for small teams, and here is where it lands on the list above.

On agent reality: BOSRAI does not claim full autonomy. A human approves messages before they go out — the human-in-the-loop model we argue for. If you want an agent that sends unsupervised, we are not it.

On cost: pricing is published — Free at $0, Starter at $79.99, Growth at $199, Scale at $499 and Pro at $999 per month, with a discount on annual billing. You can read the numbers without a demo, which is the standard we just held other vendors to.

On compliance: WhatsApp outreach runs through the Business Platform, and its opt-in rules apply to you as they do to everyone. No vendor's product creates an exemption.

On evidence: BOSRAI has no published case studies, customer logos, benchmark numbers or G2 score. We are early, and inventing those is exactly the behaviour this checklist is designed to catch. If a number matters to your decision, ask us for it and accept "we don't have that yet" as a real answer — from us or anyone else.

The checklist is more valuable than the shortlist. Run it on three vendors and the differences that matter will separate out quickly, whether or not one of them is us.

Sources

• Gartner: Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 — cancellation prediction, stated causes, and the ~130 vendors with genuine agentic capability.
• Google: Email sender guidelines — SPF/DKIM/DMARC requirement for bulk senders and the 0.30% spam-rate threshold.
• EU AI Act, Article 50 — transparency obligations for AI systems interacting with people; applicable 2 August 2026.
• WhatsApp Business Messaging Policy — the opt-in requirement before contacting a person.
• Meta: WhatsApp Business Platform pricing — per-message billing effective 1 July 2025 and template categories.
• Artisan pricing and 11x — checked 18 September 2026; neither publishes tier prices.
• AiSDR buyer's guide and Digital Applied buyer's guide — representative of what category guides cover and omit.