E.l.f. Beauty answers a lot of comments. Every post on TikTok and Instagram brings in questions about launch dates, product comparisons, and whether something works for a certain skin type. The company built a tool called E.l.fluencer to draft replies automatically, trained on its style guide and years of past comments, with a human community manager approving every reply before it posts.
The results are real. Before the tool, the company answered somewhere between 30 and 40 out of every 100 comments. That number is now above 80, and the company says AI drafts about 90 percent of the replies customers see.
For a company that sells low-priced cosmetics through fast-moving social trends, faster answers mean more sales conversations get closed while a product is still trending. But the more important detail is that E.l.f. still puts a person in the loop before anything gets published. That single choice matters more than the technology itself.
Younger shoppers, the exact audience E.l.f. sells to, are quick to call out brands for posting content that feels fake or robotic. Keeping a human as the last check is not just caution, it is a way to avoid the kind of backlash that has hit other brands for using AI-generated content without disclosure.
The bigger story is what is happening inside the company, away from customers. E.l.f. built an internal chatbot used by more than 80 percent of its staff, and an IT help desk assistant the company says does the work of one full-time support employee.
It also built a shared platform that lets any employee, not just engineers, build their own small AI tools using pre-approved building blocks. One team used it to build a dashboard that tracks how people feel about the brand across campaigns.
This build-your-own-tool approach is the part worth copying. Most companies keep AI projects locked inside the IT department, which slows everything down and creates a backlog.
E.l.f. instead gave the people closest to the actual work, like community managers and marketing staff, the ability to build what they need themselves, with guardrails already in place. That is a faster way to find useful applications than waiting for a central team to guess what every department needs.
The company currently has 85 of these AI projects running as pilots, with a goal of moving most into everyday use within six months. That is an aggressive target.
Industry surveys on enterprise AI consistently find that most pilot projects never make it to full production, getting stuck somewhere between a good demo and a tool people actually rely on every day. If E.l.f. hits its six-month goal on even half of those 85 projects, it will be moving faster than most large companies managing similar AI rollouts.
None of this is risk-free. Handing 90 percent of public replies to an algorithm, even with human review, raises the stakes if that review step gets rushed as volume grows.
The company's own leadership has said it is more cautious about using AI for creative content than for operational tasks like answering questions or routing IT tickets, and that is a reasonable line to draw. Businesses looking at this example should note that the safest use of AI right now is still speeding up routine tasks, not replacing the judgment calls that shape how a brand is perceived.