SheetAI
Healthcare & Research

Work the denial, not the spreadsheet

ACA marketplace plans denied 19% of in-network claims in 2023 — the highest rate in nine years — and U.S. hospitals spent roughly $18 billion last year overturning them. SheetAI lives inside the workbook your RCM and research teams already use, reads cell ranges in place, writes the formulas, and produces a paper trail a reviewer can sign.

Try it on your sheetSee real workflows
Audit-ready trailNo data trainingReversible per action
19%
In-network ACA denial rate
KFF analysis of 2023 HealthCare.gov claims (published 2025) — the highest in-network denial rate since KFF began tracking the data in 2015.
12 hr
Per data manager, per study, per week
Manual reconciliation and data cleaning consume roughly 12 hours of a clinical data manager's week — and 90% of the trial data lifecycle is setup, validation, and cleaning, not analysis.
<1%
Of denied claims appealed
Of the denials KFF analyzed, fewer than 1% were appealed by patients — meaning the win-back work falls almost entirely on hospital RCM teams.
Hospitals spent ~$18 billion in 2025 just overturning denied claims.

70% of those denials were eventually paid — but only after multiple costly reviews. Most of the work is structurally a join across payers, codes, and dates, done by hand in a spreadsheet. That is exactly the work an in-sheet AI is good at.

Source: AHA — Costs of Caring 2025

This page is about the administrative and research-statistical work that keeps a health system running — claims reconciliation, denial cohorts by reason code, days-in-AR aging, payer-mix variance, clinical-trial enrollment funnels, and the monthly compliance pack. It is not about clinical decision support. SheetAI does not diagnose, recommend treatment, or replace a certified coder. It reads cell ranges on demand, writes standard Excel and Google Sheets formulas a reviewer can audit, and stops the moment it is unsure. Every action is reversible, every output is explainable, and your file never has to leave your account. For workflows that touch protected health information, our default guidance is to de-identify under HIPAA Safe Harbor or Expert Determination before processing — and we say so plainly in the boundaries section below.

The state of healthcare admin in 2026

Four numbers, sourced from 2024–2026 healthcare benchmarks, that explain why your billing and research teams are tired — and where AI is and is not helping yet.

In-network ACA marketplace claims denied (2023, published 2025)

19%

Insurer-level rates ran 23–35%: Blue Cross Blue Shield of Alabama (35%), UnitedHealth (33%), HCSC (29%), Molina (26%), Elevance (23%). Less than 1% of denials were appealed.

KFF 2025

Physician EHR / admin time per 8 hrs of patient care

~3.4 hr

Physicians spend 13 hours/week on indirect care (orders, notes, results) and 7.3 hours on prior auth and insurance forms — much of it landing in spreadsheets at month-end.

AMA 2025

Clinical-trial data lifecycle spent on cleaning, not analysis

~90%

Setup, validation, cleaning, and review eat 90% of the cycle; only 10% is statistical analysis. 75% of clinical data managers cite manual steps as the primary inefficiency.

IntuitionLabs / IQVIA 2025

Net revenue leakage at U.S. hospitals from denials & uncompensated care

$48.4B

2,300 hospitals analyzed; net revenue leakage rose 25% year-over-year, driven by an increase in clinical-grounds denials. The average hospital now staffs ~64 admin/billing FTEs (~6.5% of headcount).

HealthLeaders / AHA 2025

The pattern: humans are doing structural join work — claim to remittance, enrollment to lab, ICD-10 to CARC — by hand, in spreadsheets, at scale. That is exactly the shape of work an AI that reads your cells (not your screenshots) was built for.

Anatomy of a denial-management cycle

Where the hours actually go on a typical mid-cycle denial-management workflow, before any automation. If your week looks like this, you are not behind — you are the median.

1

Phase 1 — Pull the 835s and the worklist

  • ›Export 835 ERA / 837 claim data from the clearinghouse
  • ›Pull the open-AR aging by payer from the PM/EHR
  • ›Cross-walk CARC and RARC codes against the X12 reference list
⚠ Three exports, three layouts, none of them line up. Reconciling them is structurally a join — done by hand in a spreadsheet.
2

Phase 2 — Cohort the denials by reason

  • ›Group denials by CARC (CO-16, CO-97, CO-50, etc.)
  • ›Sub-cohort by payer, provider, and place of service
  • ›Flag the working set against the 60-day timely-filing clock
⚠ There are 358 CARCs and 1,185 RARCs. Manually grouping a month of denials at a 200-bed hospital is a two-day spreadsheet job.
3

Phase 3 — Root cause and appeal queue

  • ›Tag denials as administrative, clinical, prior-auth, or coding
  • ›Build the appeal worklist sorted by dollar value × win probability
  • ›Draft the supporting documentation request list
⚠ 70% of denials are eventually paid — but only after multiple reviews. Prioritizing the queue by expected recovery is the difference between recovering 60% and recovering 95%.
4

Phase 4 — Days-in-AR aging by payer

  • ›Bucket open AR into 0–30 / 31–60 / 61–90 / >90
  • ›Compute days-in-AR per payer against the MGMA <40-day benchmark
  • ›Surface the >90-day cohort for write-off review
⚠ A clean aging takes 4–6 hours of pivot-table work that has to be re-run every Monday. The benchmark is <40 days; most teams sit at 50+.
5

Phase 5 — Monthly RCM pack & narrative

  • ›Build the denial dashboard for the CFO review
  • ›Variance pack: gross charges, contractual adjustments, net collections
  • ›Narrative drafting on the top three denial drivers
⚠ The pack is last month's deck with the numbers updated. The narrative is rewritten from scratch every cycle.
Cautionary tale

Why "AI inside the sheet" matters more than you think

In 2015, Anthem disclosed that attackers had exfiltrated personal information for roughly 78.8 million members — names, dates of birth, Social Security numbers, addresses, employment data — making it one of the largest single compromises of healthcare-related data ever recorded. The initial vector was credentials phished from a database administrator; the data lake was sitting in a place where one stolen login unlocked everything. Anthem paid a $115 million class-action settlement in 2017, a $16 million HIPAA settlement to OCR in 2018, and $39.5 million more to a 44-state coalition in 2020.

The lesson: The lesson for admin and research workflows is the same one Anthem spent nine figures learning: where your data lives matters as much as what your data is. Copy-pasting member or patient identifiers into a chat tool is exactly the failure mode that makes auditors reach for the smelling salts. SheetAI never asks you to paste identifiers into a chat box; it operates on live cell ranges, every change is annotated, and every step is reversible. Default guidance for PHI: de-identify under HIPAA Safe Harbor or Expert Determination before processing.

Source: Anthem Medical Data Breach — Wikipedia (sourced from OCR settlement records)

Who it's for

If your week is a denial pivot rebuilt from scratch, this section is for you.

RCM Directors

Need to cut days-in-AR under 40 and the denial rate under 5% without hiring three more billers.

Clinical Trial Data Managers

Spend 12 hours a week per study on manual reconciliation between EDC, CRFs, and lab feeds.

Hospital CFOs

Watch $48B of industry revenue leakage and want to see exactly where their share of it is going every Monday morning.

Practice Managers

Run reports out of athenahealth or NextGen, paste into Excel, fix the formulas, send it Tuesday.

Health-System Analysts

Re-build the same payer-mix and service-line variance pack every month and rewrite the narrative from scratch.

Real healthcare workflows

The exact prompt, the formula it writes, and the result you'd present to your CFO or compliance committee.

healthcare_workbook.xlsx — SheetAI
You ask
“In sheet Denials!A:N, group the open denials by CARC reason code and payer. Show count, total billed, total denied, and average days-since-denial. Sort descending by dollars at risk. Flag any CARC where the same code repeats >25 times for one payer in a month.”
SheetAI does
  • ✓Reads the denial export and detects the CARC, payer, and billed-amount columns.
  • ✓Builds a pivot using SUMIFS / COUNTIFS keyed on CARC × payer.
  • ✓Computes days-since-denial against today and a write-off threshold.
  • ✓Highlights the systemic-pattern rows (same code × same payer, repeating).
Formula written
=SUMIFS(Denials!H:H, Denials!E:E, A2, Denials!C:C, B2)
Result

Reconciled 4,800 denials across 6 payers in 11 minutes. CO-16 (missing info) and CO-97 (bundling) account for 62% of dollars at risk — three remediation tickets opened against the front-end registration team.

Everything healthcare admin and research teams need, in one chat box

Turn EHR exports, payer remits, and trial data into reviewable summaries without leaving the workbook. Automate joins, aging, and variance work; keep the audit trail your compliance team needs.

✓Denial cohorting by CARC / RARC
✓Days-in-AR aging by payer
✓Clinical-trial enrollment reconciliation
✓Payer-mix and service-line variance
✓Audit-trail logging per action
✓De-identification helper formulas

Plays well with your stack

  • Epic Clarity / Caboodle exports (CSV / XLSX)
  • Cerner / Oracle Health PowerInsight extracts
  • Athenahealth report exports
  • NextGen / eClinicalWorks billing exports
  • REDCap data exports for clinical research
  • CMS public-use files (Medicare claims, MDS, OASIS)

What a denial-management cycle looks like

A representative mid-size medical group on a typical month-end, before and after SheetAI lands in the workflow. The "before" mirrors the AHA / KFF picture of admin overhead; the "after" is what RCM customers report after their second cycle on the platform.

✕

Before SheetAI

~42 hours
  • Mon
    Pull 835/837 exports, fix layout drift, rebuild the denial pivot
    ~8h
  • Tue
    Cohort denials by CARC × payer; tag administrative vs. clinical
    ~9h
  • Wed
    Days-in-AR aging across 6 payers; chase >90 bucket
    ~7h
  • Thu
    Variance pack: gross charges, adjustments, net collections
    ~8h
  • Fri
    Compliance audit summary; rebuild the E/M distribution
    ~6h
  • Sat
    Late finds, appeal-queue prioritization, narrative rewrites
    ~4h
✓

With SheetAI

~9 hours
  • Mon
    AI cohorts 4,800 denials by CARC × payer in 11 minutes
    ~2h
  • Tue
    Reviewer pass on flagged appeals; AI ranks by recovery odds
    ~3h
  • Wed
    Aging auto-built; >90 cohort surfaced with statement-run list
    ~1h
  • Thu
    Variance narratives auto-drafted; you edit, not write
    ~2h
  • Fri
    Compliance pack generated with audit trail attached
    ~1h

~80% reduction in RCM cycle hours, on a representative mid-size medical group.

What SheetAI will not do

A healthcare tool that is honest about its limits is the only kind worth installing. Some decisions and some data belong in a different system entirely.

✕

Process PHI without a Business Associate Agreement

HIPAA / HITECH require a BAA for vendors handling protected health information. SheetAI does not currently sign BAAs at the free tier. Default guidance: de-identify under Safe Harbor (45 CFR § 164.514(b)(2)) or Expert Determination before processing. For BAA-covered enterprise workflows, contact us — do not assume coverage.

✕

Replace a 21 CFR Part 11 system

For FDA-regulated clinical-trial data, Part 11 mandates secure, computer-generated, time-stamped audit trails for record creation, modification, and deletion. SheetAI logs every action it takes with prompt, range, and reversible diff — that helps your audit story, but it is not a turnkey Part 11 EDC. Pair it with a validated EDC for the regulatory record-of-truth.

✕

Provide clinical decision support

SheetAI does not diagnose, recommend treatment, triage acuity, or interpret labs against clinical guidelines. Every workflow on this page is administrative or research-statistical. If your question is "what should this patient be on," that is a question for a licensed clinician using a validated CDS tool — not a spreadsheet AI.

✕

Replace certified medical coders

It surfaces patterns and flags outliers — a provider's E/M distribution skewing 15 points off baseline, a CARC repeating across 25 claims for one payer. Translating that into the right ICD-10, CPT, or HCPCS code is still the job of a certified coder. We make their queue shorter, not their credential redundant.

“
We were rebuilding the same denial pivot every Monday morning for two and a half years. With SheetAI, the cohort runs while my team is still pouring coffee — and the appeal queue is sorted by recovery odds, not by who shouted loudest. The narrative writes itself. The auditors get a cleaner trail.
— Director of Revenue Cycle, regional health system (~$340M net patient revenue)

Frequently asked

Things RCM, research, and compliance teams ask before they switch.

Is SheetAI HIPAA-compliant?

›
SheetAI is configurable for HIPAA-compliant workflows when used with de-identified data — the right phrasing matters here. We do not currently sign Business Associate Agreements at the free tier, so the default guidance for any workflow that touches PHI is to de-identify under HIPAA Safe Harbor (45 CFR § 164.514(b)(2)) or Expert Determination first. For enterprise customers who require a BAA, contact us — but do not assume one is in place by default.

Do you sign Business Associate Agreements?

›
Not on free or standard plans. We are honest about this because the worst HIPAA failure mode is a vendor that lets a customer believe coverage exists when it does not. Enterprise plans can include a BAA on request — talk to sales before sending PHI through any AI tool, including this one.

Can SheetAI be used for clinical decision support?

›
No. Every workflow on this page is administrative or research-statistical: billing, denials, RCM, payer mix, trial enrollment, compliance auditing. SheetAI does not diagnose, recommend treatment, or interpret patient-level clinical data against guidelines. If your question is clinical, the answer belongs with a licensed clinician using a validated CDS tool — not a spreadsheet AI.

What about 21 CFR Part 11 for clinical-trial data?

›
Part 11 requires a secure, computer-generated, time-stamped audit trail for the creation, modification, and deletion of electronic records. SheetAI logs every action — prompt, affected range, reversible diff — which strengthens your audit story. It is not a turnkey Part 11 EDC, however. Pair it with a validated EDC (Medidata Rave, Veeva Vault EDC, OpenClinica, etc.) for the regulatory record-of-truth, and use SheetAI for the data-cleaning and reconciliation work that sits adjacent to it.

How does it handle de-identification?

›
SheetAI can write Safe Harbor-style helper formulas in your sheet — hashing MRNs, truncating zip codes to three digits, age-banding patients over 89, removing names and direct identifiers — but the de-identification decision is yours, not ours. We surface patterns; you decide what data leaves the original sheet and what stays.

Can it handle a year of claims data?

›
Yes. SheetAI reads ranges on demand rather than loading the whole file into context, so it scales to hundreds of thousands of rows. For very large reconciliations — a year of 837s across multiple payers — the Python tool runs the join server-side and writes the result back to your sheet.

Will the formulas work in Excel and Google Sheets?

›
Every formula SheetAI writes is standard Excel / Google Sheets syntax. There are no proprietary functions to install, and your file remains portable. The Excel add-in and Google Sheets add-on let you run the same prompts inside Microsoft 365 and Google Workspace.

Is there an audit trail my compliance officer can review?

›
Yes. Every action is logged with the prompt, the affected ranges, and a reversible diff. You can export the trail as a working paper alongside the trial balance, denial pack, or compliance summary. For Part 11 contexts, treat it as supporting evidence — the system of record stays in your validated EDC.

Does it replace our certified coders?

›
No. It surfaces patterns and flags outliers — a provider whose 99215 distribution skews 18 points off practice baseline, a CARC repeating across 30 claims for one payer, a missing modifier on a high-frequency CPT. Translating those flags into the right ICD-10, CPT, or HCPCS edit is still your certified coder's job. We make their queue shorter and more prioritized; we do not make their credential redundant.

Sources & further reading

Every benchmark and statistic on this page is drawn from publicly available research. We cite our sources because we read theirs.

Related teams using SheetAI

Adjacent functions that hit the same kinds of problems.

Finance & Accounting

Close in three days, not three nights

Read guide →

Operations & Logistics

Streamline operational data

Read guide →

Education & Research

Statistical analysis without the dread

Read guide →

Stop reworking denials.
Start collecting cleanly.

Open SheetAI, drop in a denial export, and watch the cohort build itself. Free forever for the first 20 credits a day — no card required to find out whether it works on your data. For PHI workflows: de-identify first, or contact us about a BAA.

Start freeAll use cases

No credit card required — Free forever tier available