Should Your Brand Talk Back?

Every brand, it seems, is getting a chat window this year. Retailers launching AI shopping assistants, beauty brands shipping conversational advisors, CPG companies wiring their mascots to language models. The board asks about AI strategy, and a branded agent is the most visible possible answer.

We have now advised several clients through this decision, including one we talked out of it entirely, and a pattern is emerging worth sharing. The question is not whether the technology is ready. It mostly is. The question is whether conversation is a load-bearing part of your customer relationship, or whether you are about to spend seven figures building a very sophisticated FAQ that also occasionally embarrasses you.

The strategic test: does dialogue add information?

Strip the novelty away and a brand agent is a channel. Channels earn their existence by carrying something the other channels cannot. So the first filter we apply is simple: is there a moment in your customer journey where a genuine back-and-forth changes the outcome?

For some categories the answer is clearly yes. Complex, considered purchases where customers arrive with idiosyncratic constraints: home renovation, skincare regimens, financial products, B2B software. Aperture asked us about this in the spring, and creative software is a genuine fit, because "what tool do I need for this weird project" is a real dialogue with real stakes. A well-built assistant there replaces a support queue and a sales call.

For low-consideration categories, the answer is usually no. Nobody wants a conversation with their sparkling water. When dialogue adds no information, an agent is a slower interface to your website, and customers punish slower interfaces regardless of how charming they are.

Voice is the asset at risk

Here is the piece most implementations get wrong, and where strategy has to lead engineering. A brand agent speaks in the first person, as you, thousands of times a day, unsupervised. That makes it the largest brand voice exposure your company has ever created. Most teams spend ninety percent of the budget on retrieval accuracy and ten percent on voice, and it shows: agents that are factually correct and sound like everyone else's agent, because they all inherited the same polite, exclamation-heavy default register.

Our rule: if you cannot articulate your brand voice precisely enough to write a spec for it, you are not ready to automate it. The voice work has to come first, and it is harder than the plumbing. Tone under complaint. Tone under refusal. What the brand never says. Whether it apologizes, and how. These decisions compound across a million conversations.

The governance questions to answer before launch

• Failure posture: when the agent does not know, does it guess, defer, or hand off to a human? The wrong answer here is where public embarrassments come from.
• Boundaries: which topics are simply out of scope, and does the refusal itself feel on-brand?
• Memory and privacy: what does the agent retain, and would customers be comfortable seeing that retention described plainly?
• Measurement: define success before launch. Deflected tickets, assisted conversion, and satisfaction deltas are real metrics. "Engagement" alone is how these projects hide their failure.

Our honest recommendation

If your category passes the dialogue test: pilot narrow. One journey, one use case, exhaustive voice work, human escape hatches everywhere. Expand only when the transcripts read like your brand on a good day.

If it does not pass: skip the agent and spend the money making sure the platform agents, the ones your customers already use, describe you accurately. Because here is the twist of 2025: whether or not your brand talks back, your customers are already having conversations about you with an AI. The strategic priority is not owning a chat window. It is being represented well in all of them.