SaaS Says: Above Average

Above Average

Kieran Flanagan pulled a full agentic marketing team off GitHub, stripped out the brand context so it would run cold, pointed it at a fast-growing AI company and graded the output. HubSpot's SVP of marketing was not trying to be reassuring. He was checking whether he could be replaced.

The verdict came back calibrated. The audit was useful. The cold email was good enough that he flinched: "This is very good. It's better than most emails I get. I actually kind of I'm worried about how good this is." The positioning work beat what most product marketers would produce. Then it proposed the wrong category, and its search visibility audit quietly ran a web search instead of querying the models it claimed to be testing, and never said so. "I think what we're proven here is that AI can do an above average job, it cannot do a world-class job." Two rules came out of it. Systems should declare when they could not do what was asked. And "The agent is not reviewing its own work."

Ross Rich reached the same finding from the sales seat and then priced it. The co-founder and CEO of Accord talks to hundreds of revenue leaders, and his starting number is blunt: "if you're selling, your top 20%, honestly, maybe even closer to sub 10% of the team is driving most of the revenue." Every leader says the goal is lifting the middle 60%. His read: "I don't think organically it's moving that middle." They make people who already have taste faster, which widens the gap. The uncounted cost is what happens to the method. "you have your top sellers that maybe previously were sharing with their SE and other team members, what they were doing in more collaborative and shared workspaces on the application layer. Now it's more siloed." The best work stopped being visible to the people meant to copy it.

Clare Corriveau is the counterfactual, and she is hiring nine people. The VP of marketing at Tekmetric sells into 250,000 US auto repair shops, a buyer unreachable by every default a B2B marketer owns. "Our buyers are not on LinkedIn. They're literally never gonna download anything, ever." So the engine is paid search on Bing, a Facebook user group the company refuses to move onto its own platform, peer coaching groups, more than a hundred trade shows a year, and a rule that every new hire visits a working shop before touching a campaign. One customer introduced himself by number before he gave his name: "I'm shop number 1792. That's how they identified themselves." Against a timeline of people announcing that Claude Code replaced their marketing team, her position is unfashionable and specific. "But we still need people to think and we still need people to look at things and say, does this make sense?"

Casey Muratori has the version from an industry that already ran this experiment. Cheap licensable engines removed the barrier to shipping a game, then the storefront filled past the point where quality alone got you found. "the licensable engine thing kind of was our AI transition already, unfortunately. And I regret to inform you that the news is not probably that positive." Being good became table stakes. Distribution became the differentiator. He also has the reason the aggregate studies keep coming back ambiguous: "if you have a tool that can give everyone 10% off, that's great. Almost no one would know, right?"

The average went up. That was never the number that paid.

Sources: Interviews from Marketing Against the Grain (Aug 27, Kieran Flanagan of HubSpot), GTMnow (Aug 25, Ross Rich of Accord), Exit Five (Aug 24, Clare Corriveau of Tekmetric), and The Pragmatic Engineer (Aug 26, Casey Muratori).