How to choose an AI outreach system: the seven questions that actually matter

Every AI sales tool demo looks impressive. Prospects get found, messages get written, replies get handled, meetings appear in the calendar. Thirty minutes later you have seen a system that appears to do the entire job of a sales development rep, and the pricing page suddenly looks reasonable.

Then a meaningful number of those programmes quietly stop within a year.

The tools are rarely the reason. The reason is that the demo answers questions about capability, and the things that actually determine whether outbound works are questions about ownership, judgement and accountability. Those never come up in a demo unless you raise them.

So here are the seven we would ask any vendor, including us. Take them into every conversation you have with anyone in this category.

1. Who decides who gets contacted, and are they a person?

Almost every failed outbound programme we have looked at failed at the list rather than the message.

Targeting is a judgement problem. Somebody has to know that a company is too small to buy, that a job title means something different in this industry than the next one, that this account is actually a competitor, that this person started three months ago and has no budget yet. Software does not have that instinct, and it does not develop it by processing more contacts.

This matters more with automation than without it, not less. If your target definition is loose, a system will contact the wrong people faster and more consistently than any human could manage. Bad targeting does not get diluted at scale. It gets multiplied.

Ask directly: who builds the list, what criteria do they use, and can I see it before anything is sent?

2. Do I approve the messages before they go out?

There is a version of this technology where copy is generated and sent without anyone reading it. Avoid it.

Roughly half of B2B buyers stop engaging the moment a message feels automated, and by 2026 most decision makers have received thousands of them. They recognise the shape before they read the content: the flattering opener, the pivot to a problem they supposedly have, the tidy value proposition, the low-friction ask.

The fix is not clever prompting. It is that a person who understands your market reads every sequence before it is live, and keeps editing it as replies come in. That is a small amount of work which removes almost all of the reputational risk, and any serious vendor should build it into the process rather than treat it as an optional extra.

Ask: can I see and change every message before it sends, and how easily can I change it later?

3. What happens the moment somebody replies with a real question?

This is the question that separates the systems that work from the ones that look good in a demo.

The work before a conversation is genuinely repetitive. Building the list, sending the first message, remembering to follow up on day nine. Removing that from a salesperson's week is the entire value of automation, and it is real value.

The moment someone replies with a real question, the situation reverses. Now it depends on knowing what your product does not do, where the pricing has flexibility, whether this prospect resembles a customer you have already served well, and whether the deal is even worth pursuing. That is not a processing task.

A good system handles the neutral and administrative replies, the polite questions, the scheduling, the "circle back in September" messages, and gets a human involved when there is real buying intent. A weak system treats every reply the same way and burns your best opportunities on generic answers.

Ask: what does the system reply to on its own, what does it escalate, and can I control where that line sits?

4. Who is accountable for meetings held?

When outbound is run by a team, someone is answerable for the number. When it is run by a platform, accountability tends to disappear into the dashboard.

Six months in, the reporting shows thousands of contacts reached, hundreds of replies, and a pipeline that has not moved. Nobody adjusted the targeting, because nobody's job was to notice. Nobody rewrote the sequence, because technically it was running.

The metric to hold everyone to is qualified meetings held. Not messages sent, not connection acceptance rate, not reply rate. Those are inputs, and they are easy to make look healthy while producing nothing.

Ask: which single number do we review together, how often, and what happens when it is bad?

5. What is the commitment, and is it the right length?

Outbound needs roughly three months to be judged fairly. A few weeks to configure and get the targeting right, a few more to reach full sending volume, and a few beyond that before replies turn into meetings in any consistent way. Judging a programme at week four tells you almost nothing.

That cuts both ways. Three months is the minimum honest evaluation period, and it is also close to the maximum you should have to commit before you can walk away.

Entry pricing for autonomous AI sales agents commonly runs between $500 and $1,200 a month, with enterprise agreements reaching five figures annually and twelve month commitments attached. That is a significant bet placed before there is any evidence, with no exit at the point you would actually know something.

Ask: what is the minimum term, what does the exit look like, and why is it that length?

6. What does this cost per qualified meeting?

Monthly fee is the wrong comparison. A cheaper programme that produces fewer meetings is not cheaper.

Take the total monthly cost, including tools, data, seats, internal time and setup amortised across the term. Divide it by qualified meetings held per month, using a conservative figure rather than the vendor's best case. Most B2B outbound programmes convert somewhere between 4 and 10 percent of contacted decision makers into meetings.

Then check it against your own economics. If your average deal is worth €20,000 and you close one in five qualified meetings, each meeting carries €4,000 of expected value and most options on the market pay for themselves comfortably. If your average deal is €2,000 and you close one in ten, a large retainer will never make sense no matter how good the vendor is.

For reference, a fully loaded in-house SDR in the US runs roughly $9,000 to $14,000 per month once you include salary, tooling, management and ramp. Agency programmes run $2,500 to $15,000 per month, or $150 to $600 per mainstream B2B meeting. We wrote up the full breakdown separately if you want the numbers in one place.

Ask: what does a realistic month look like, and what is the resulting cost per meeting held?

7. What does it do when nothing is happening?

The most expensive outbound programme is the one that stops.

Most in-house efforts do not fail because the strategy was wrong. They fail because prospecting is the first thing dropped when the quarter gets busy, and the pipeline gap appears two quarters later when it is far more expensive to fix.

The strongest argument for a system rather than a person is not that it is cheaper. It is that it keeps running in the weeks when everyone is heads down on delivery, and that most B2B deals require five or more follow-ups, which is exactly the point where human effort tends to give out.

Ask: what keeps running when my team goes quiet, and what happens to a conversation that goes cold in month two?

The honest summary

Software is very good at reach, consistency and memory. It is not good at judgement, and it is not good at the conversation once a real opportunity opens. Any vendor who tells you otherwise is describing a demo rather than a quarter.

The systems that work in 2026 split the job along that line. People decide who to target and what to say. The system executes at a scale and consistency no individual can maintain. Every sequence is approved before it sends. When somebody is genuinely interested, a person takes it from there. Somebody is answerable for meetings held. The commitment is long enough to judge honestly and short enough to leave.

That is the model we built SalesBot on, and those seven questions are ones we are happy to answer directly rather than after a signature. If you want to see what the answers look like for your specific market, including how many relevant decision makers are actually reachable in your segment, book a demo here.