CARR vs ARR: How to Read an AI Vendor's Growth Claim

What is Contracted ARR, and how is it different from ARR?

I'm Hlib Storchak. I build outbound systems for B2B founders and sales teams, and I've booked 2000+ meetings for B2B clients doing it, which means I read a lot of vendor pitch decks and case studies before I recommend a tool to a client. This one is worth walking through slowly because the two terms sound almost identical and get used almost interchangeably in the wild, which is exactly the problem.

ARR, annual recurring revenue, is meant to represent the annualized value of contracts that are signed, live, and being paid for today. CARR, contracted or committed ARR, adds in the value of contracts that are signed but not yet implemented, not yet fully onboarded, or not yet actually generating cash. A company that just closed a big enterprise deal with a four-month implementation timeline can count that deal's full annual value as CARR on day one, months before a single invoice clears.

Both numbers are legitimate things to track internally. The problem TechCrunch's reporting surfaced is companies collapsing the two into one word, ARR, in public materials, decks, and press, when the number underneath is actually CARR.

Why would a vendor report CARR as if it were ARR?

Because a bigger number raises more money at a better valuation, and multiples in AI right now are largely set against revenue growth. Scott Stevenson, the co-founder and CEO of legal AI startup Spellbook, called the pattern a "huge scam" in on-the-record comments to TechCrunch, and said "the biggest funds in the world are supporting this." Michael Marks of Celesta Capital gave the more structural version of the same point: "the valuations have gotten higher, and so the incentives are stronger." Neither of those is a fringe complaint. Both are people inside the fundraising process describing an incentive that runs in one direction, toward the bigger number, with nobody structurally required to correct it.

How big is the gap, really?

TechCrunch's reporting includes a specific range rather than a single figure, which is the more honest way to present this. One VC told the outlet they had seen a company where CARR ran 70% higher than realized ARR, meaning a headline "$100M ARR" company was collecting revenue closer to $59M. A separate example in the same reporting involved marketing materials claiming $50 million in ARR against an actual figure closer to $42 million, an $8 million gap on a number small enough that the discrepancy is easy to check once you know to look. TechCrunch also reports that sources described at least one high-profile enterprise startup that announced surpassing $100 million in ARR when only a fraction of that came from customers who were actually live and paying.

Source: "How VCs and founders use inflated 'ARR' to crown AI startups," TechCrunch, May 22, 2026, checked directly.

Is this against any rule?

No, and that is the uncomfortable part. ARR is not a GAAP metric. There is no accounting standard that defines it, no auditor who has to sign off on it before it appears in a pitch deck or a press release, and no regulator checking whether a startup's public revenue claim matches its books. TechCrunch's sources describe this as widely known inside the investing community, with Jack Newton of Clio noting that founder Scott Stevenson "did a great job of highlighting some of what you might describe as bad behavior," and Ross McNairn of Wordsmith summarizing the state of play as "there are some choppy, choppy standards out." Legal is a low bar. It just means nobody outside the company is required to catch it, which puts the burden on whoever is reading the number, an investor, a journalist, or you, evaluating whether to buy the product.

Why does this matter if I'm buying, not investing?

Because a growth number is doing the same persuasion job in a sales pitch that it does in a funding round: it tells you the product works well enough that other companies are paying for it at scale. If you're evaluating an AI SDR or a GTM AI tool and the vendor's homepage or sales deck leads with "$40M ARR" or "10,000 customers," that number is meant to lower your guard before you've asked a single question about your own use case. If a meaningful share of that number is contracted-but-not-live revenue, or logos that signed a pilot and never converted, the growth story is telling you less about product-market fit than it looks like it is. This isn't unique to AI SDR vendors, TechCrunch's reporting is about AI startups broadly, but the AI SDR and GTM AI category is exactly the kind of fast-fundraising, high-multiple market where the same incentive applies, and where a buyer has no more visibility into the real number than an investor does.

What other AI SDR vendor numbers get the same treatment?

ARR is the clearest example because it now has a named investigation behind it, but the same instinct, report the number that includes the most activity rather than the number that reflects the most outcome, shows up elsewhere in this category too. A vendor's own internal case study describing leads contacted and opportunities created, for example, tells you about volume the vendor's system generated, not necessarily revenue a customer closed. Self-reported "X% lift" claims with no disclosed sample size or methodology are the same pattern one level down: a real number, generalized past what it can support, because nobody outside the company has to check the math before it goes on a landing page. I've written before about how to pressure-test a vendor's self-reported lift claim for the same reason: the honest fix isn't assuming every number is fake, it's asking what, specifically, was measured, and by whom.

A worked example: reading a vendor's pitch deck stat

Say an AI SDR vendor's sales deck states "$30M ARR, 400 customers, 3x growth year over year" as the opening slide. Here's how I'd actually read that in a first call, and it takes about three questions, not a forensic audit:

1. "Is that ARR or CARR, and can you show the split?" A vendor with nothing to hide answers this in one sentence. A vendor that gets vague or reframes the question is telling you something.
2. "Of the 400 customers, how many have been live and paying for more than 90 days?" Logos signed last month with a long implementation runway inflate the count without telling you anything about renewal behavior yet.
3. "What's your logo churn and net revenue retention over the last four quarters?" Growth without a churn number attached is half a story. A vendor confident in its retention will usually offer this before you ask twice.

None of these questions accuse the vendor of anything. They're the same questions a competent investor would ask before wiring a check, and a legitimate vendor answers them without friction. The vendors worth worrying about are the ones where the growth number gets repeated three times in the deck and none of these three questions get a straight answer.

The mistake I see most when clients evaluate a vendor's traction

The mistake isn't naivety, most founders and sales leaders I work with are plenty skeptical of marketing copy in the abstract. It's that the skepticism doesn't survive contact with a big enough number. A vendor claiming "50,000 companies use us" or "$100M ARR" gets treated as a solved question, the product must work at that scale, rather than as a claim that deserves the same three questions as a smaller vendor's more modest pitch. The size of the number is doing the persuading, not the substance behind it, and that's exactly backwards. The bigger the number, the more it's worth five minutes of asking what's actually inside it, because a bigger number is also a bigger incentive to have rounded up.

A pressure-test checklist before you trust a growth number

1. Ask whether the number is ARR or CARR, by name. If the vendor hasn't heard the distinction, that's informative on its own.
2. Ask what share of customers are live and paying versus signed and onboarding. A healthy split should be easy for the vendor to state.
3. Ask for churn and retention alongside growth. Growth with no retention context is an incomplete number by design.
4. Check whether the claim appears consistently across the vendor's own materials. A number that shifts between the homepage, the deck, and the last press release is worth asking about directly.
5. Treat a round, headline-friendly number with extra scrutiny. $50M reads better than $42M, and TechCrunch's own example shows that gap is exactly the kind that gets rounded away.
6. Ask for a reference customer who has been live for a full year, not a recent logo. A year of real usage tells you more than a signature does.
7. Remember none of this is audited. ARR and CARR are not GAAP metrics, so the number in the deck is only as reliable as the person who built the slide.

What this changes about how you negotiate a pilot

Once you accept that a vendor's headline growth number might be softer than it looks, the practical response isn't to distrust every vendor equally, it's to shift what you're buying on. Ask for a structured pilot with a defined success metric tied to your own outcome, meetings booked, positive reply rate, pipeline sourced, rather than signing a full annual contract on the strength of the vendor's growth story. A vendor confident in its real numbers will usually accept a shorter initial term or a usage-based ramp, because their own retention data supports it. A vendor that pushes hard for a 12-month commitment upfront and resists a smaller pilot is asking you to underwrite their growth story with your budget before you've seen the product work on your own list. This is the same logic I use when I take over a client's stack: prove it on a small slice of real volume before committing the whole budget, regardless of how the vendor's deck reads.

When I'd take a growth number closer to face value

Not every big number is soft. A vendor that publishes ARR and CARR as two separate, labeled figures, names reference customers who'll confirm live usage on a call, and shares retention data without being pushed, has already cleared the bar most of this article is about. The scrutiny here isn't a reason to assume every AI SDR vendor is lying about growth, it's a reason to ask the same handful of questions regardless of how impressive the headline sounds, and to notice when a vendor answers them easily versus when the conversation gets vague. If you'd rather have someone run that scrutiny as a standing part of how your team evaluates every tool in the stack, not just remember to ask once, that's closer to a fractional GTM function than a one-time vendor call.

FAQ