Unscheduled Treatment Backlog: The Dollar Figure Already Sitting in Your Practice Management System

A general practice with 1,500 active patients is typically carrying somewhere between $400,000 and $1.2 million in diagnosed-but-never-scheduled treatment inside its practice management system at this moment. Almost none of that figure surfaces on a production report or a monthly P&L, because Dentrix, Eaglesoft, and Open Dental all treat an unscheduled plan as a clinical document rather than as a receivable.

The number is real in one sense: a licensed clinician diagnosed every line item on it and attached a fee to each one. It is fiction in another sense, because a meaningful share of it was completed at another office, superseded by a later diagnosis, or attached to a patient who stopped being a patient two years ago.

What follows is the operator's version of the problem — how to size the backlog so the figure survives contact with the ledger, why it decays on a predictable curve, and what an automated agent has to verify before it re-surfaces a twenty-four-month-old treatment plan to a live patient.

Why The Figure Stays Invisible

Accounts receivable gets an aging report because a posted charge creates a ledger entry, and a ledger entry demands a follow-up cadence with a named owner. A treatment plan creates neither, so nothing in the standard reporting stack ever ages it, chases it, or writes it off.

Consider how the same practice treats a $1,400 crown depending on a single keystroke. Completed and billed, it enters accounts receivable days and gets worked weekly by a specific person; planned and unscheduled, it enters a report nobody owns.

The result is a category of production with no owner, no aging, and no write-off policy. That combination is why the backlog grows almost monotonically in any practice open more than five years, and why the first honest measurement is usually uncomfortable.

What Counts As Unscheduled Treatment

Every practice management system has a report for this, and each one names it differently. Dentrix exposes it through the Treatment Manager, Open Dental through Treatment Finder and the Unscheduled Treatment Plans report, and Eaglesoft through treatment plan follow-up.

The reports disagree because the underlying category is not one thing. Sizing the backlog honestly starts by separating five populations that most offices collapse into a single total:

• Presented and declined. The patient heard the plan, understood the fee, and said no. These convert at the lowest rate of any band, and they are the most likely to have been completed at a second-opinion office.
• Presented and deferred. The patient agreed in principle and left without scheduling, usually over benefit timing or a January deductible. This is the highest-yield population in the entire backlog.
• Diagnosed but never presented. The doctor charted D2740 on #14, the operatory ran behind, and nobody walked the patient through it. It sits in the system as a plan the patient has never actually seen.
• Partially completed sequences. The patient finished D3330 and never returned for the D2950 buildup and the D2740 crown. An unrestored endodontically treated molar is a clinical liability as much as a revenue line.
• Watch items and conditional plans. A charted watch on an incipient lesion is not a diagnosed procedure, and rolling it into the dollar total inflates the number by a wide margin.

All five live in the same table, and most canned reports return them as one undifferentiated list. Accordingly, the first engineering task here is classification, not outreach.

How To Size The Backlog Without Fooling Yourself

The gross number is a query anyone can run: sum the fee of every treatment-planned procedure with no completion date and no linked future appointment. That query is also the source of nearly every inflated backlog figure ever put on a whiteboard.

A defensible figure requires a discount ladder applied in order, because each step removes rows the next step would otherwise double-count:

• Drop inactive patients first. Anything attached to a patient with no completed visit in 24 months comes out before any other filter runs, which is why an honest active patient count is the prerequisite for this entire exercise.
• Remove procedures completed under a different code. A planned D2391 that came back as D2740 on the same tooth is finished work, and a naive tooth-plus-code match will miss it every time.
• Collapse superseded plan versions. Most systems retain every version of a plan, so one crown can appear three times across three exam dates and triple its own contribution to the total.
• Re-price against the current fee schedule. A 2023 plan carries 2023 fees, and the gap between the planned fee and today's contracted rate is the same leak described in dental fee schedule leakage.
• Strip watch items and hygiene recall. D1110 prophylaxis and D0274 radiograph codes belong to recall, not to the treatment backlog, and mixing them in makes the number look considerably better than it is.

In the practices we have measured, the ladder removes 55% to 75% of the gross figure. That result lands badly the first time an owner sees it, and it is the only version of the number worth building a workflow on top of.

Why The Backlog Ages Out Of Relevance

Treatment plans decay clinically before they decay commercially. A twelve-month-old plan for D2391 on #30 carries a second problem beyond staleness, because the lesion it described may have progressed past a composite, which makes the plan wrong rather than merely old.

Because of that, the aging bands govern more than tone of voice. They govern what verification is mandatory and what the outreach is permitted to say:

The operative rule sits in the last two rows. Past roughly eighteen months, an agent should stop referencing the planned procedure entirely and invite the patient to a re-evaluation, because the practice can no longer stand behind the specific clinical claim in the record.

The Verification Gate An Agent Has To Clear

This is where unscheduled-treatment automation earns or loses its reputation. An agent that reads the treatment plan table and starts sending messages will, inside a week, tell someone to schedule a crown on a tooth that was extracted last spring.

Every candidate record has to pass a verification gate before it becomes an outreach. The checks below are ordered by how expensive the failure is when they get skipped:

• Patient status and record flags. Deceased flags, transfer-of-records requests, collections status, and dismissal notes live in different tables than the treatment plan, and none of them are joined by default.
• Tooth-level completion, not code-level. Resolve the plan against the ledger by tooth and surface across all codes, including procedures posted by a specialist and imported through a claim.
• Conflicting later diagnosis. If #19 was subsequently charted for extraction, the crown plan is void, and the agent has to treat the most recent clinical entry as authoritative.
• Plan version currency. Select the newest plan version for a given tooth and suppress every predecessor, or one patient receives three messages about one crown.
• Insurance re-verification. Remaining annual maximum, frequency limitations, missing-tooth clauses, and waiting periods all have to be pulled fresh, because a cached eligibility response from the original plan date is the stale-state problem described in AI insurance verification.
• Current fee and current carrier. Re-price the procedure against today's fee schedule and the patient's present plan before any dollar figure reaches the patient.
• Provider attribution. If the diagnosing dentist has left the practice, the message cannot imply a continuity of care that no longer exists.
• Contact consent and channel. SMS consent and opt-out status govern whether outreach is permitted at all, and the content of the message governs whether it is appropriate on that channel.

Any record that fails a check routes to a dead-letter queue with the failing check named, rather than to a silent retry. A queue nobody reads is how a verification gate quietly becomes a rubber stamp.

Where The Clinician Stays In The Loop

Verification handles the data, and a dentist still has to handle the medicine. The boundary between those two jobs is the single most important design decision in this workflow.

A workable split gives the agent everything mechanical and reserves everything interpretive for the doctor. In practice, the agent assembles the packet — plan, ledger reconciliation, eligibility result, last radiograph date, last completed visit — and the dentist of record approves or kills each item above the eighteen-month line in a batch review that takes minutes.

Below that line, approval can run on a standing protocol the doctor signs once. Above it, per-item human approval is what keeps a recovery program from turning into a board complaint, which is the practical reading of the state dental board rules on AI in patient communication.

What The Pipeline Looks Like In Production

The architecture is unglamorous, and every interesting failure mode is about freshness. A nightly extract from the practice management system into a staging table, a verification pass, a clinical review queue, and a paced outreach step is the whole shape of it.

Four implementation details do most of the work:

• Idempotency keyed on the clinical fact. Key each outreach on patient, tooth, surface, code, and plan version, so a re-run of the extract cannot double-message a patient about the same crown.
• Shadow mode for the first 60 days. Run the full pipeline and write every proposed message to a review table without sending it, then have the office manager mark false positives. The correction rate on tooth-level completion is where you learn whether the ledger join is actually right.
• Freshness windows on every input. Eligibility older than 30 days, a fee schedule older than the last contract update, or a ledger snapshot older than 24 hours should each invalidate the record instead of degrading it silently.
• An audit trail per message. Store the plan version, code, diagnosis date, eligibility response ID, and approving provider alongside every outreach, because the first patient complaint gets answered from that record.

PHI moves through all four steps, so the usual controls apply: a signed BAA with every processor, KMS-encrypted staging, CloudTrail across the data path, and model-version pinning so a silent provider upgrade cannot change how a plan is summarized. The HIPAA requirements for clinical AI in dental practices govern this pipeline exactly as they govern charting or note generation.

Evaluation belongs in the same build rather than in a later phase. A backlog agent needs a labeled set of resolved cases — completed elsewhere, superseded, extracted, still valid — scored the same way any other clinical AI evaluation is scored.

What A Realistic Recapture Number Looks Like

That is the honest shape of the return, and it is smaller than most pitch decks suggest. It is also production that requires no new patients, no additional marketing spend, and no chair time beyond what the schedule already has open, which is why it compares favorably to nearly any acquisition channel on a dental AI ROI basis.

The second-order effects matter as much as the first. Recovered cases are measured against the same denominator as freshly diagnosed ones, so a quarterly backlog program changes what your case acceptance rate actually describes, and the recovered appointments have to land in slots that AI scheduling optimization would otherwise fill with lower-value production.

Where These Programs Break

Three failure modes account for most of the bad outcomes. All three are visible before a patient is ever contacted, provided the team runs shadow mode long enough to see them:

• The ledger join is wrong and nobody checks. When work posted by an outside oral surgeon never reconciles against the plan, the agent confidently pursues finished treatment, and the patient concludes the office does not read its own records.
• Volume without pacing. Dropping 600 messages in a single week produces inbound call volume the front desk cannot absorb, and unanswered callbacks convert worse than no outreach at all.
• Chart fragmentation. Practices that merged locations or migrated systems frequently hold two records for one patient, which defeats every completion check downstream and is why dental chart consolidation has to precede any backlog program.

Each of these is detectable in shadow mode at essentially zero cost. None of them is detectable from a dashboard three months after go-live.

Frequently Asked Questions

Working The Backlog Without Spending Patient Trust

If you are sizing your unscheduled treatment backlog and deciding whether an agent belongs anywhere near the outreach, the team at NexV builds and operates HIPAA-grade clinical AI inside Dentrix, Eaglesoft, and Open Dental environments every week. Reach out for a working session — we will run the discount ladder against your own extract, specify the verification gate your chart structure requires, and hand back a 60-day shadow-mode plan you can execute before a single patient is contacted.