This article is based on an interview with Atera Chief Marketing Officer Nir Elharar for The Agile Brand Guide’s Expert Mode series.
Marketing leaders have spent roughly two years buying AI on the strength of demos. A vendor shows the generation, the clean chat window, the workflow that runs beautifully on stage, and the room nods. The purchase gets justified on what the thing can do. What it actually did gets deferred to a quarterly review nobody ends up scheduling.
Nir Elharar, Chief Marketing Officer at Atera, argues that era is over — and not because vendors got more honest. Buyers got burned. Atera sells an autonomous IT agent called Robin and prices it against incidents resolved rather than seats, so Elharar has an obvious commercial interest in an outcome standard. The more useful part of his argument is where he points it next. The same standard is coming for marketing, and most marketing measurement isn’t built to survive it.
The Demo Stopped Being the Product
For a stretch, “AI-powered” was the claim, not a description of one. Elharar’s account of why that stopped working is refreshingly unsentimental. It isn’t that buyers got smarter. It’s that they spent money and had nothing to show anyone.
For a long time, an AI vendor got measured on the demo: a slick generation, a smooth chat interface, and buyers nodded along because there wasn’t yet a good way to hold anyone accountable for what happened after the demo ended. That’s changed, mostly because buyers have been burned. They’ve bought “AI-powered” a few times now without much to show for it afterward, and they’re done paying for potential.
The mechanism he describes is a ratchet, and it’s worth sitting with. Once an organization has one agentic AI deployment it can actually point at, that deployment sets the floor for every AI purchase that follows. Capability stops being interesting on its own. For a CMO this runs in both directions. On the buying side, your first real agent deployment becomes the internal benchmark whether you intended it to or not, so it’s worth picking a use case you’d be willing to be measured against later. On the selling side, the same ratchet applies to your own claims — including the ones your product marketing team is drafting right now.
A Belief Is Harder to Grade Than a Tagline
Atera launched “Let People Work” in March 2026 as a full brand platform: TV, out-of-home, an experiential takeover at the HIMSS healthcare tech conference, timed alongside the next generation of Robin. The claim underneath it is structural rather than promotional — that IT friction is a design failure and not a fact of working life. Claims like that are easy to launch and genuinely hard to grade. Elharar says he’s watching for unprompted “language echo,” prospects using the phrase before anyone feeds it to them. Then he described something more rigorous.
I’m also disaggregating brand lift from message lift: standard brand tracking tells you if people remember Atera, but what I actually need is a belief-attribution measure, whether people who’ve seen the campaign agree with the underlying claim that most IT friction is avoidable at a higher rate than people who haven’t, independent of whether they can name us as the source. If awareness moves but belief-agreement doesn’t, we’ve bought attention, not conviction.
That’s a design a marketing team can actually build. Brand awareness tracking asks whether people remember you. Belief-attribution asks whether your argument moved, with or without your logo attached to it, which means adding one agreement statement to a tracker you’re probably already fielding and cutting the results by exposure. The uncomfortable implication is that a category-level belief helps your competitors too. Elharar seems fine with that trade, and the failure mode he names is the one worth pinning above a CMO’s desk: a campaign that wins creative awards while pipeline stays flat. Attention and conviction come apart early, and it’s easy to mistake the first for the second for a couple of quarters.
Write the Test That Could Fail
Ask most marketing organizations how they know a campaign worked and you’ll get a model. Elharar’s problem with the models is not that they’re wrong so much as that they’re accommodating.
A lot of what gets called attribution is really correlation wearing a nicer outfit: a deal closed while a campaign happened to be running, and we draw a line between them because it’s convenient. What makes an outcome measurable enough to stand behind is deciding, before the campaign runs, what evidence would actually prove it worked and what evidence would prove it didn’t. Not the metric you’re hoping to hit, the test that could fail.
Incrementality discipline, arriving through the side door of a CMO’s own habit rather than through a stats team. The practical version at Atera runs on first-party data: every SDR asks in discovery how the prospect heard about them, and post-conversion questionnaires offer multiple-choice source options instead of a single guess. Elharar is careful that this doesn’t replace the attribution model — it either validates the model or contradicts it, and disagreement is a signal to go investigate before spending more, not a reason to default back to the model. His closing warning on the subject is the sharpest line in the exchange: “If the only evidence a campaign worked is that a number went up while it happened to be running, that’s not attribution, that’s timing.” Pipeline moves for all sorts of reasons. A competitor stumbles. Sales has a good quarter. Write the failing condition down before launch, because reverse-engineering a flattering story from numbers already on the screen is nearly effortless.
Marketing to a Room Instead of a Person
Atera grew on a self-serve motion where a buyer could sign up and get value without ever meeting a salesperson. Robin’s traction is now pulling the company into security reviews and procurement committees. Elharar is blunt about how much of the marketing job changes.
In self-serve, the product does the convincing: someone signs up, gets value in minutes, and the marketing job is mostly getting the right person to that moment quickly. In enterprise, you’re marketing to a room, not a person. A CISO wants to see the certifications, procurement wants the audit trail, and a champion needs ammunition for their own internal pitch.
Anyone who has run a product-led growth motion and then tried to sell upmarket will recognize the gap, but Elharar’s handling of Atera’s ISO/IEC 42001 certification, announced in early August, shows the translation work concretely. It’s a governance standard for managing AI systems over their lifecycle rather than a security certification, a distinction almost no buyer outside procurement will parse unprompted. His marketing case wasn’t the badge. It was, in his words, “here’s proof we built the guardrails before anyone made us.” For everyone else, the cert gets translated into things a non-security person can picture — configurable guardrails, approval workflows, a record of what an agent did and why. The credential opens the door. Plain language does the actual storytelling. What he says he’d fight hardest to protect through all of this is the self-serve instinct to let the product prove itself fast rather than lean on a deck, which is a reasonable thing to worry about losing, since longer enterprise cycles create a lot of room to over-promise.
The bar Elharar describes for AI vendors — show me what got resolved, not what could be — is the same bar arriving for marketing budgets, and it will not arrive politely. The defense isn’t a better dashboard. It’s deciding in advance what evidence would embarrass you, then keeping that definition where the whole team can see it.
Pick the next campaign on your calendar and write down two things before it ships: the result that would prove it worked, and the result that would prove it didn’t. If you can’t describe the second one, you haven’t defined the first.
