What the First Months With an AI Receptionist Look Like

The first months with an AI receptionist are a learning curve followed by a fast correction loop. At Arcadia Neuropsychology, real callers raised questions nobody had planned for. The office manager flagged each one, it was fixed within a day, and the next caller with the same question got the right answer. Three months after launch, Dr. Brian Klinck said the system was "honestly working better than I thought it was going to work."

This field note draws on the Arcadia Neuropsychology case study and a call with Dr. Klinck on October 9, 2026.

What did Arcadia need?

Arcadia, based in Phoenix, performs psychological evaluations before surgeries such as spinal cord stimulator implants and bariatric procedures, all by phone. Referrals come from surgeons in many states, and a small office team was fielding scheduling calls, insurance questions, and cancellations. Dr. Klinck wanted a receptionist that could book directly, follow the practice's insurance rules, and hand off to staff when needed.

Grace, the AI receptionist Cream Digital built, went live in July 2026. She matches each caller to the referral on file, checks the insurance plan against the practice's rules, and books, moves, or cancels evaluations directly in Practice Fusion. By mid-September she had handled more than 230 calls.

What happened when real patients started calling?

They asked things nobody had written down. Dr. Klinck described it this way on the call:

"Neuropsych is its own animal. There's no possible way that it was easy to train your actual AI system to do this because we really, none of us knew exactly what pitfalls to expect until people started calling."

That is normal for any practice with its own rules. The scripts and decision rules written before launch cover the calls everyone can imagine. The first weeks show the rest: a request phrased in a way nobody expected, an insurance situation the rules did not cover, a caller who needed something the system did not know how to process. The way to prepare is to turn judgment calls into written rules up front, and to expect the first weeks to add more.

How do wrong answers get fixed?

Through one short loop, run every time. Arcadia's office manager sees how calls go. When something is wrong, she tells Cream Digital what happened and what Grace should have done. According to Dr. Klinck, she told him that every time she reported something, it was fixed within a day, and the next time someone asked that question, Grace answered it correctly.

The loop has four steps:

1. Someone who sees the calls flags the problem, with the example.
2. The rule or the script changes, not just the one answer.
3. The change is checked against the call that went wrong.
4. It goes live for every call after it.

The last step is the difference from training a person. A fix to the AI does not have to be repeated at the next shift or for the next new hire.

Should you expect mistakes?

Yes, and plan for them. Dr. Klinck's view after three months:

"It's going to mess up every once in a while, but not any more than a human."

He compared it to training a new medical assistant who has never worked in neuropsychology, who makes the same kinds of mistakes the AI did. What keeps mistakes from becoming problems is the safety net around the AI. At Arcadia, every booking is checked against the live schedule to prevent double booking, insurance plans the office must review go to staff instead of being booked, and a backstop emails the office every 15 minutes if a caller mentioned cancelling but no cancellation was logged. The same patterns appear in whether an AI receptionist can book into your EHR.

What changed for the office?

Time, mostly. Dr. Klinck said the system has given his office manager more time, and that it is scheduling people correctly. In his words, "the turnaround from you guys stating what you were able to do and going to do and actually doing it has been huge. It's been exactly what you said was going to happen."

What comes next?

Prior authorization. Payers covered by CMS-0057-F must build prior authorization APIs by compliance dates generally beginning January 1, 2027, which lets a practice send a request directly instead of through a portal. Dr. Klinck called the change "a godsend." Arcadia's staff member who handles authorizations has already learned the practice's insurance rules, so a person stays in the loop for anything that needs a closer look. The wider picture is in AI for prior authorization.

What should you plan for in your first 90 days?

Four things made Arcadia's first months work, and they apply to any practice:

• A named owner at the practice who sees the calls and reports problems.
• A fast fix cycle, measured in days, not weeks.
• Safety nets for the costly errors: double bookings, missed cancellations, and insurance exceptions.
• Honest expectations. The first weeks will surface calls nobody predicted, and that is how the system gets good.

That is how Cream Digital runs front desk operations after launch day.