The economics nobody quotes
Administrative transactions have a measurable unit cost, and the gap between doing them by hand and doing them electronically is not marginal. Per transaction, for the provider side:
- Eligibility and benefit verification: $7.97 manual, $2.18 fully electronic
- Claim status inquiry: $11.37 manual, $4.29 fully electronic
- Prior authorization: $10.97 manual, $5.79 fully electronic — and 22 minutes against 11
- Claim submission: $5.65 against $3.10; remittance advice $4.76 against $2.86
Stacked across a visit, CAQH estimates 70 minutes of staff time available per patient visit from running the full transaction set electronically — a sum across every transaction type a visit can touch rather than one saving — and puts the remaining national savings opportunity at around $20 billion.
For a practice, the useful translation is per-day. A single-physician practice seeing 22 patients has somewhere between two and four hours of front-desk labour tied up in transactions that have a fully electronic equivalent. That is not a productivity abstraction — it is the difference between a receptionist who can answer the phone and one who cannot.
Why the front desk decides the claim
The three most commonly cited causes of denial are missing or inaccurate data (50%), authorisation (35%), and incomplete patient registration (32%). (Experian Health, 2025)
Every one of those originates in the first ten minutes of a patient's visit — or in the phone call that preceded it. By the time a claim is being scrubbed, the error is several days old and the person who could have caught it has seen forty patients since.
The revenue cycle does not start when the claim is submitted. It starts when someone types a member ID off a phone photo while the waiting room fills up.
What actually needs fixing
Insurance as data, not as an image
A photographed card is evidence, not information. What the claim needs is a payer, a member ID, a group number, a subscriber relationship and — for anyone over 65, on workers' comp, or with other coverage — a completed Medicare Secondary Payer questionnaire. When those are captured as structured fields at check-in with the image retained behind them, the retyping step disappears and so does its error rate.
Eligibility on the day, against the plan presented
Coverage lapses between scheduling and arrival. Checking at booking is useful; checking again when the patient is standing there is what prevents the denial.
Intake that reconciles instead of accumulating
Most practices already send digital intake forms. Most practices also then have a staff member read the answers and retype the relevant ones into the chart, because the form and the record are different systems. The value is not in the form. It is in the reconciliation — surfacing what the patient reported next to what the chart already says, so a human confirms differences rather than rekeying agreements.
Self check-in for the patients who want it
A QR code that lets a patient complete arrival, forms and signature from their own phone removes a queue without removing a choice. The practices that do this badly make it mandatory; the ones that do it well leave the desk staffed for everyone who prefers a person.
A signature that actually captures
A small, specific, and embarrassingly common failure: signature pads that silently lose the capture on certain devices. Consent and financial responsibility forms that appear signed and are not are worse than forms that obviously failed. Test this on a real phone, mid-scroll, with an interrupting notification.
Where AI helps, and where it is just software
Worth being clear-eyed: most of what is described above is not artificial intelligence. It is integration and data modelling — the unglamorous work of making two systems agree. Vendors increasingly label it AI because AI sells. It is a good idea regardless of what it is called.
The genuinely model-shaped problems at the front desk are narrower:
- Reading unstructured payer correspondence — a denial letter, a request for information — and extracting the reason, the deadline and the required documents.
- Reconciling patient-reported history against the chart, where the patient writes "the water pill" and the chart says furosemide.
- Drafting the appeal or the authorisation narrative from the note that already exists, for a human to review and send.
Notice that all three end with a person. That is not caution for its own sake — the disclosure rules in several states now turn on whether a licensed person reviewed an AI-generated patient communication before it went out.
The measurement problem
Practice leaders are split on whether AI has helped at all: 44% report no reduction in staff workload, 39% report improvement, and 17% aren't sure. The groups reporting benefit are the ones using it in more than a quarter of visits. (MGMA Stat, 2025) As with ambient documentation, shallow deployment produces indistinguishable-from-zero results.
Sequencing, for a practice with limited attention
If you can only fix one thing this quarter, fix same-day eligibility against structured insurance data. It is the cheapest transaction to automate, it addresses the largest single denial category, and it requires no clinical change management.
If you can fix two, add intake reconciliation — not more forms, but forms that land in the chart.
Self check-in and QR-code arrival are worth doing third. They are the most visible to patients and the least connected to revenue, which is precisely the order in which most practices do them backwards.