A marketing team can stand up an AI tool in an afternoon now. That used to be the hard part. It isn't anymore, and it's quietly changing what separates one team from another.
E.l.f. Beauty gave me the clearest recent example. Fast Company reported on the brand's AI reply tool, which it calls E.l.f.luencer. It answers comments on TikTok and Instagram, and it now writes about 90 percent of them. The interesting part is what sits underneath it. The tool was trained on E.l.f.'s own stylebook and on years of replies its community managers had already written, so it isn't a generic model doing a generic voice. It's the brand's voice, learned from the brand's own record. Every reply still passes a person before it posts. E.l.f. says its response rate went from 30 to 40 percent up past 80 since launch. Those are one company's numbers, so I read them as direction rather than a benchmark, but the direction is the point.
The lesson isn't "use AI to answer comments." It's that the model was the cheap ingredient. The stylebook and the archive of human-written replies were the expensive ones, and E.l.f. already owned them. Anyone can license the same model. Nobody else has ten years of E.l.f.'s community managers writing in E.l.f.'s voice.
Matthew Henderson made a related point this week about build versus buy for marketing tools. His rule of thumb is to build when your own data or a real customization gives you an edge, and to buy when you're mostly paying a vendor to run infrastructure that's a pain to maintain. He also names a trap I've watched teams walk into, the tool one person vibe-codes, that only that person understands, and that the team ends up leaning on until the day it breaks. His test is a good one. Ask what would actually stop working if you cancelled a piece of software, and whether you're buying the product or just a tidy way to organize data you already have.
Both come back to the same place. When the tools are commodities, the advantage moves to the inputs. Your proprietary data. Your house style, written down where a model can learn it. The people who know why the work is good, not only how to make more of it. None of that arrives with a subscription.
This is easy to miss right now because the tools are the loud part. There's a new one every week, and each one promises to replace something. But the teams pulling ahead don't seem to be the ones with the newest tool. They're the ones who had something worth feeding it.
I want to test that against real practice. I'm talking with a group of Vancouver marketing practitioners this fall about where AI has actually changed how they work, and I expect the question of what a team owns to come up more often than the question of which tool it bought. I'll write about what they tell me.
Sources
- How E.l.f. Beauty is building AI into its community management: fastcompany.com
- Build vs. buy: marketing tools in the vibe code era: hendersonmatthew.substack.com