I understand why people hate AI support. Everyone has been trapped in a bot that seems designed to keep them away from a person. So when Tyler Denk, the founder of beehiiv, posted last week that he will never understand why companies replace their support teams with AI, I agreed with his instinct.
Handing your customers to a bot and calling it support is a mistake.
But at Tern, we run most of our customer support on AI. Not a little, most of it. And our customers get faster, real-time, detailed answers, tied to the context of their business and our product and don’t need to wait for a human.
The reality at Tern is that AI amplifies the human philosophy underneath it. If your goal is to avoid customers, AI will help you do that faster, but that’s not what we’re trying to do.
Our goal is to protect human judgment for the moments that actually need it, meaning AI makes our support feel more human than it did before. That is the difference that matters.
AI amplifies the support philosophy underneath it
For as long as support has existed, companies have treated it as a cost to reduce. The job of a cost center is to get cheaper. So when AI arrived, most leaders pointed it at the obvious target: deflect more tickets, involve fewer people, lower the cost.
That framing created the version of AI support customers hate: a bot that exists to keep them away from a person and a workflow designed around avoidance.
The backlash makes sense. If the goal is to remove humans from support, customers will feel it.
The better question is not, "Can AI answer this instead of a person?" bur rather, "Which parts of support require human judgment, and which parts are volume, retrieval, routing, or repetition?"
That distinction changes everything. Bad support plus AI becomes faster avoidance. Good support plus AI creates more room for human judgment when it matters and offers our users AI as a high value, trained assistant to get help in real time.
AI only works when experts shape it
I learned this the hard way.
My first real attempt to use AI for support work was a mess. I had just run a 3-day training for agency owners and had hours of questions, feedback, and real examples of how advisors actually run their businesses. I gave it to ChatGPT and asked for a help center article.
What came back was mediocre. It took hours to wrangle and I finished thinking I should have just written it myself.
The problem was not the AI model I was using to synthesize. It was that I handed it raw material with a bad prompt and hoped it would have judgment in procuring the results.
The second attempt worked because I invested the time upfront (probably longer than I would have in doing the task myself). I added what the model could not know on its own: our templates, our voice, our product context, our standards, and the judgment to know what not to say and what not to assume: do not overstate a feature, flag uncertainty for testing, write for an advisor running their business in language they can act on, etc.
The guidance we give to AI to make it personal to Tern became the operating principle for our support team.
AI in support does not replace the expert. It only works when the expert pours their judgment into it. The bot is not the alternative to your team, but it is the output of your team.
What we automate, and what we do not
At Tern, the line is simple. AI can handle volume, retrieval, routing, and repetition. Humans own judgment, nuance, and care.
That is how we can handle nearly 2,000 support conversations a week, with Fin, our AI agent, resolving more than 85% of them on its own at a 96% satisfaction score. The conversations that reach a human are the hard ones, and last week our human CSAT hit 98%, the highest it has ever been.
We are on track to nearly double the advisors we serve this year without doubling the team.
Those numbers only work because Fin is not left alone. Every week, every person on our support team spends dedicated time improving it. We feed in new product releases, real user feedback, use cases surfaced during 1-1 or group trainings, and the gaps we catch in the queue. When the product changes, Fin changes with it, because a person made sure it did.
The loop is simple:
- AI handles the repeatable volume.
- The team reviews what it misses.
- Product changes and user feedback get fed back into the system.
- Human specialists spend more time on the cases that need context and judgment.
We also use AI inside the human workflow. Tern’s support team has built a variety of tools to act as a personal assistant to them in the queue, surfacing product explanations, customer context, and troubleshooting steps prior to them digging in. Our use of AI doesn’t stop at operations and instead is also deeply embedded in role development and feedback. An AI coach reviews every ticket a specialist handles and shows them where their scores are trending and where their product knowledge has gaps, so each person owns their own growth instead of waiting for a manager to read tickets one at a time.
The point is not the tooling itself, but rather what the tooling protects.
When a conversation reaches a person, that person is not buried under every repetitive question in the queue. They have time to review the advisor's actual situation instead of just answering the question at the surface. When an advisor mentions the trip they are stressed about or an urgent issue with an itinerary for a traveler leaving for their vacation tomorrow, Tern’s support team has the knowledge and capacity to thoughtfully respond in detail, share walk throughs, or even hop on a live 1-1 call to walk them through the fix
That is the part a deflection machine can never do.
The customer standard
For our customers, this distinction matters. We are not using AI to make a person harder to reach. We are using it so simple questions get answered faster and in real-time regardless of when our users are working and the moments that need a person get one with more context, more time, and more attention.
This is the same bet we make for our customers. We tell advisors that AI should take the transactional work so they can spend their time on the client relationships only they can build.
We hold our own team to exactly that standard.
In 3 years, the support teams that win will not be the ones that deflected the most tickets and they will not be the ones that proudly refused to use AI either.
They will be the teams that used AI to carry the volume and spent every hour it gave back on the customer. They will be the teams that know what should never be automated and equally, don’t assume we reach a ceiling with AI and continue to innovate on how to deploy it in high value ways to the customer and to our internal teams.
As we build the future of support at Tern, every person on our team deeply believes in our vision for AI-powered customer service. Building that buy-in and culture in a function that historically has been reduced to a cost center transforms support into so much more.
Support stops being the department you contact when something breaks and becomes a reason people stay.
The test is not whether AI answered the ticket. The test is whether the customer felt more understood because of how you used it.