How to think about adopting AI in customer service

If you run CX at a premium DTC brand, you’re facing two conflicting pressures right now.

Vendors are in your inbox weekly. LinkedIn is full of deflection case studies. Investors and operators in your network are asking what your AI roadmap looks like. The pressure to ship something is real.

Your customers still want to talk to a human. 79% of Americans now say they strongly prefer it. Half of US consumers say they specifically prefer brands that avoid using GenAI in customer-facing content at all.

Both pressures are legitimate. The responses most teams reach for first tend to make one side worse to relieve the other. There’s a fourth way of thinking about it that starts by reframing what’s actually being decided.

Where the AI pressure is coming from

AI in customer service was the most visible cost-savings story in B2B tech in 2023 and 2024. Klarna’s 700-job announcement got cited in every vendor pitch deck. Decagon, Sierra, Intercom Fin, and Gorgias AI all published deflection-rate case studies. The economics looked clean. Cheaper per resolution, available around the clock, infinite throughput.

That story made it into investor updates, onto LinkedIn, and into every “future of CX” panel. It’s why the pressure is in the room even when it isn’t on the agenda. It’s also a snapshot of where the conversation was eighteen months ago, and the conversation has moved.

Premium brands waited for more signal

The dust has started to settle, and things are starting to move in a different direction.

We never want our customers to feel they are one after another in a ticketing system.

Skincare is so personal and in an industry like this, customers appreciate the human connection. I think backend, behind the scenes, AI is great to use for our team, but I don’t think necessarily it’s great to be customer facing, especially with a luxury brand. People expect that luxury service as well.

None of these leaders are anti-AI. They chose a different path after watching what happened to brands that automated aggressively.

What’s actually happening at scale

The 2026 data shows it plainly.

That 8% number is the real story. Brands aren’t achieving deflection at scale. They’re shipping it, measuring it, and realizing the promise doesn’t match the practice.

The three moves teams are making

Watching teams work through this, three responses come up over and over.

The half-step deploy. Stand up a chatbot, set a conservative deflection target, see what happens. The bot accumulates one-star moments faster than the deflection rate moves. You end up defending a number that isn’t moving the way the deck said it would.

The hold. Tell the room you’re not deploying customer-facing AI yet. This worked in 2024 but gets harder to explain as time goes on. You’re not anti-AI. You just don’t have a precise way to say what kind of AI you’re open to.

The wait-and-see. Watch what peers ship, let them publish the results first, decide once the dust settles. The catch is you’re absorbing pressure every week, with no story to point at.

None of these are mistakes. They’re the moves a thoughtful operator makes when the choice is framed as “deploy AI or don’t.” A fourth path opens up once the framing changes.

A different framing

There are two distinct places AI can sit in a customer service operation, and they don’t have to move together.

Customer-side AI replies to customers. It deflects. It tries to resolve tickets without a human. It’s what your customers are telling surveys they don’t want.

Team-side AI is invisible to your customer. It pulls order history, previous interactions, context. Your support team sees all of it pre-assembled the moment a ticket lands. They reply faster. The customer sees a quicker, more informed human. The AI never speaks to them.

What to ship, based on your priorities

The two AI options solve different problems. Naming the priority you’re optimising for makes the choice easier.

If your priority is brand voice and retention

Premium DTC brands whose customers chose them partly for the human touch pay the most for getting customer-facing AI wrong. Team-side AI is the path. Customer-facing chat stays human. AI sits behind the scenes pulling order history, parsing returns, triaging warranty, all invisible to the customer.

This is where Wild Alaskan, Levain Bakery, True Botanicals, and KREWE all land. None of them are anti-AI. They run AI deliberately on the team side and keep their customer interactions human.

If your priority is cost per ticket

Both options move it. Team-side does it without changing what the customer experiences: a smaller team can handle the volume of a larger one when each rep has full context pre-assembled the moment a ticket lands.

Customer-facing deflection adds further savings on top, on the ticket categories where it works well: order tracking, ETA, return status, simple FAQs. The Gorgias 2026 data shows only 8% of refund and return tickets resolve end-to-end without a human, so the cost story lands on the simpler half of your queue, not the policy-touching half.

If your priority is response time

Both can win here. Customer-facing AI delivers instant first-response, especially valuable for off-hours coverage or peak-load gaps. Team-side gives faster replies with a human still on the other end, because each rep has the context they need the moment the ticket lands.

The trade-off is consistency versus accuracy. Customer-facing AI is consistent and instant but sometimes confidently wrong. Team-side is consistently accurate but only as fast as your team. For brands where every interaction is a brand transaction, the slower-but-right answer wins. For brands where availability matters more than precision, customer-facing AI is the better tool.

If you have the scale and resources to support it

Customer-facing AI gets harder to ignore at real ticket volume. At 5,000+ tickets a week, even a 30% deflection on the categories where AI works (order tracking, ETA, return status, simple FAQs) is a real labour saving. The brands that get the most out of customer-facing AI tend to have three things in place:

• A mature CX team with the bandwidth to monitor outputs, manage escalations, and maintain the KB the AI reads from
• Clean, documented policies that don’t depend on human judgment for the high-volume categories
• An owner for AI ops, someone whose job is to tune prompts, audit hallucinations, and manage the vendor relationship

When all three are in place, customer-facing AI does what it promises: instant responses on the easy tickets, freeing your team to spend more time on the hard ones. The brands burned by it usually deployed without one or more of these in place.

Sources

• SurveyMonkey 79% prefer humans. SurveyMonkey 2025 Customer Service Statistics, n=2,017 US adults, December 2025.
• Gartner 50%-consumers-prefer-non-GenAI. Gartner press release, March 16, 2026. Survey of 1,539 US consumers, October 2025.
• Gorgias 2026 State of Ecommerce CX. Aggregated data from 16,000 Gorgias brands. 8% AI-only resolution rate on refund/return tickets; 62% of brands planning CX headcount growth in next 12 months.
• Klarna 2024 AI announcement. Klarna press release, February 27, 2024.
• Premium-brand CX leader quotes (Avra Wacks, Amanda Tarantino). Public interviews on industry platforms.