Are Those AI Sales Numbers Real? How to Check Before You Buy

In early 2025, an AI sales startup called 11x was claiming roughly $14 million in annual recurring revenue. A former employee told TechCrunch that only about $3 million in contracts survived the three-month trial period.¹ The same reporting found the company had listed logos of businesses that were not paying customers. One of them was ZoomInfo, which said a one-month trial performed significantly worse than its human reps. It had never granted permission to use its logo.¹

That story did something lasting to how sales leaders read an AI vendor's pitch deck. The doubt used to be about the technology. Buyers wanted to know whether their leads would notice they were talking to software. That question is mostly settled now, because the conversations work. The doubt has moved to the numbers on the slide.

So the useful skill in 2026 is not spotting good AI. It is telling which numbers would survive a reference check before you sign anything.

Buyers now want a reference they can call

Walk into any evaluation this year and you will hear a version of the same request. Show me a customer who looks like me, and let me call them. That request is the whole game. A vendor who can produce three referenceable customers in your industry has told you more than any conversion-rate chart ever could.

A reference call surfaces the things a chart is built to hide:

• How long onboarding actually took, measured against the timeline on the slide
• What broke in the second month, once the novelty wore off and volume went up
• Whether anyone picked up the phone when it did

But a reference only counts if the customer is real and still active, which is precisely where the 11x story came apart.

Headline stats are cheap to print and hard to check. A headline reply rate or a fast payback figure means nothing until you know three things: the sample behind it, the time window, and whether the customer who produced it is still paying. The 11x churn figure is the real lesson there. A number pulled from a trial that ended two months later is a screenshot of a moment that has already passed.

None of this means AI sales tools underdeliver. The category is real and growing. The inbound half of it, where AI responds to leads who already raised their hand, is running live traffic at scale today. So the open question is not whether the technology works. The burden of proof simply shifted onto the vendor. The buyers who win are the ones who make the vendor carry it.

Three claims you can verify yourself

Sort every claim a vendor makes into two buckets:

• Claims that live only on the vendor's slide. Response rates, payback periods, any figure whose sample and time window you cannot see for yourself.
• Claims you can confirm independently. A registration record, a date, a customer who will pick up the phone.

Spend your evaluation time on the second bucket. Three things belong there by nature.

1. Time in production. How long has the platform actually run live traffic? A company that has been handling real sales conversations since 2017 has seen edge cases a six-month-old tool has not met yet. Aktify is one of those companies, with more than 10 million conversations behind the current models. There is no shortcut to it. A model gets good by meeting messy, unscripted conversations, and those only turn up in live traffic over years. So ask when a vendor processed its first live lead. Treat the answer as a matter of record.

2. Carrier registration. Automated texting to consumers runs through carrier compliance rules. A vendor that is properly carrier-registered can show you the registration. One that waves the question away is telling you something. This is verifiable in minutes and it protects you from a deliverability problem you would otherwise inherit.

3. Referenceable customers. These are names of people who will take a 20-minute call and tell you what actually happened, not logos sitting on a web page. Aktify's customers include Fortune 200 names, and the ones who agree to a reference call are the proof that matters.

The pattern across all three is the same. Real proof points to something outside the deck: a registry, a date, a person you can call.

Ask to see the individual leads

The strongest evidence is the kind you can inspect yourself during the pilot. An aggregate is easy to present and hard to question. A per-lead record is neither. So the question worth asking is whether you can open the reporting up and see individual leads, or only the totals a vendor chose to show you.

This is where lead-level tracking earns its keep. Aktify's Branded Link Tracking gives each lead a unique, carrier-safe short link, and every click is recorded against that specific person. You see a real lead opening a real link at a real time, inside the Aktify portal alongside the rest of the campaign. And the same signal can flow into your CRM through webhooks or a native integration when you want it there.

Set your pilot up so the definitions are settled before it starts. Pin these down with the vendor on day one, so nobody relitigates them after the fact:

• Contact rate. What counts as a contact, and what the denominator is.
• Reply rate. Whether an out-of-office or an opt-out counts as a reply.
• Booked meetings. Booked, held, or both, and who confirms which.
• Clicks. Unique per lead, or every tap including repeats.

Then check the numbers against your own calendar as the pilot runs. When the detail is inspectable and the definitions are agreed, the reference check gets a lot shorter, because you are watching the proof happen.

Send every vendor these five questions

Bring these questions to every vendor on your shortlist. Good ones answer plainly and fast, without needing a follow-up meeting to prepare the response. The answers tell you more than the polished demo ever will.

1. Time in production. When did you process your first live lead, and how many conversations have you handled since?
2. Compliance. Are you carrier-registered, and can you show the registration?
3. References. Can I speak with two or three current customers who run volume like mine?
4. Visibility. Can I see results down to the individual lead, or only as aggregate totals?
5. Retention. What is your customer retention past the first three months?

Question 5 closes the 11x loophole directly. A vendor confident in the product will tell you the retention number. One that is selling a screenshot will change the subject.

Test your shortlist before you book another demo

Take the AI sales tools already on your shortlist and run them through the five questions above before you schedule another demo. It costs an email each and it re-sorts your list fast. But send them in writing. A vendor who will put an answer in an email is a different signal from one who only says it out loud on a call. The vendor who answers the retention question with a real figure and hands you two references has already passed the test that sank 11x. The one who sends another case study PDF has answered a different question than the one you asked.

The market has plenty of AI that works. What it lacks is buyers who insist on proof they can verify. Be that buyer, and the vendor selection gets a lot simpler.

If you want proof you can watch happen, [see Aktify text a real AI conversation to your phone](/?demo=open), the same way one of your leads would experience it.

Sources

1. TechCrunch, "a16z- and Benchmark-backed 11x has been claiming customers it doesn't have," March 24, 2025. https://techcrunch.com/2025/03/24/a16z-and-benchmark-backed-11x-has-been-claiming-customers-it-doesnt-have/