How Meridian Health Cut Front-Desk Admin Time by 38%

Problem

Meridian Health operates 12 clinics across the region, each staffed by two to three front-desk employees juggling phone intake, walk-ins, and paperwork simultaneously. Internal time-tracking showed roughly a third of every shift went to repetitive intake tasks: insurance verification, symptom pre-screening, and appointment confirmation — almost all of it collected once by phone and then re-typed into their EHR by hand.

The COO's brief was direct: reduce that administrative load without adding headcount, and without compromising the patient experience or HIPAA compliance.

Research

We spent two weeks shadowing staff across three representative clinics — a large urban location, a mid-size suburban clinic, and a small rural office — because intake patterns varied meaningfully by size and patient demographics. We also pulled six months of call logs to quantify exactly where time went, rather than relying on staff's general sense of the problem.

The data confirmed the COO's instinct but sharpened it: symptom pre-screening and insurance verification accounted for 70% of intake time, while appointment confirmation was comparatively minor. That reprioritized our build order.

Strategy

We recommended a phased rollout rather than a company-wide launch: one pilot clinic for four weeks to validate the approach, then a staged expansion to the remaining 11 clinics gated on the pilot's escalation rate staying under 8% — our working definition of "the assistant is handling this reliably."

Design

Two design decisions mattered most. First, conversation flows were written in plain language rather than mimicking a web form — patients described symptoms the way they would to a person, not by selecting from a dropdown. Second, every interaction had a visible, easy escalation path to a human, both because some cases genuinely need a person and because patient trust in the system depended on knowing that option was always there.

Implementation

The system was built on Next.js with a PostgreSQL-backed conversation state machine and Twilio for SMS delivery. Protected health information stays inside Meridian's existing infrastructure rather than a third-party store, and access follows the same least-privilege model as their EHR.

Challenges

The hardest problem wasn't technical — it was ambiguous symptom descriptions. Patients describe the same condition wildly differently, and early in shadow-mode testing, the assistant occasionally under- or over-escalated based on phrasing alone. We solved this with a broader, more conservative escalation ruleset rather than trying to perfect symptom classification, accepting a slightly higher human-review rate in exchange for safety margin.

Solutions

Two weeks of shadow-mode testing — where the assistant ran in parallel with staff intake but its outputs weren't yet acted on — surfaced most of the ambiguous cases before go-live. We tuned escalation thresholds twice post-launch based on real staff feedback, which cut false escalations roughly in half without missing genuine cases.

Results

Six weeks after full rollout across all 12 clinics, front-desk admin time was down 38%, and the escalation rate had settled at 6% — comfortably under the 8% threshold that gated expansion. Patient satisfaction scores, tracked separately, held steady at 4.7 out of 5, indicating the automation didn't come at the cost of patient experience.

Lessons Learned

Shadow-mode testing was the single highest-value part of the process — it caught issues with zero patient-facing risk, at the cost of two extra weeks in the timeline. We'd make that tradeoff again on any healthcare engagement without hesitation.

Future Improvements

Meridian and SmartFlow AI are scoping a second phase extending the assistant to post-visit follow-up scheduling, which surfaced as a related pain point during the original shadowing sessions but was intentionally out of scope for this engagement.