Internal working document · P1–P12
Each candidate process is scored on the same axes so they can be compared directly. Only one will become a live engagement, and this register records the reasoning behind that choice.
Illustrative and internal
Verdict is a curated aggregate that weights measurable cash and sales-cycle fit, and penalises competitive density, capital needs and data-access difficulty. It is not a raw average.
| ID | Pilot | Readiness | Competition | Effort | Cash | Verdict |
|---|---|---|---|---|---|---|
| P1 | Denial & underpayment recovery | Recommended | Open wedge | Medium | High | |
| P8 | Work prioritisation by expected recovery | Ready | Open wedge | Low | High | |
| P7 | Appeal drafting (copilot) | Ready | Open wedge | Medium | High | |
| P9 | Payment / remittance reconciliation & underpayment detection | Ready | Partial | Medium | High | |
| P6 | Denial classification & root-cause | Ready | Partial | Low | Indirect | |
| P2 | Claim / invoice document ingestion & extraction | Ready | Crowded | Low | Indirect | |
| P3 | Coding consistency & required-field checks | Ready | Crowded | Medium | Medium | |
| P4 | Eligibility / prior-auth failure triage | Ready | Crowded | Medium | Medium | |
| P10 | Aged AR / slow payment chase | Ready | Crowded | Medium | Medium | |
| P5 | Payer rule validation (eClaimLink-oriented) | Ready | Crowded | High | Medium | |
| P11 | Denial prevention (pre-submit) — later phase | Phase 3+ | Crowded | High | Medium | |
| P12 | Full RCM takeover | Locked | Crowded | High | High |
Ordered by recommended run sequence. Positioning is shown by default; problem statement, current market solution, process steps and scoring sit behind disclosures.
Recovery SWAT overlay: historical denial audit → score expected AED → draft appeals → human approve → book cash vs written baseline. Offering: onboarding fee + % of incremental collections. Pricing angle: aligned with hard recoveries, not hours or factoring discounts.
Estimate the probability a denied or underpaid claim is recovered if worked, and how much comes back.
Running in this prototype on synthetic data. Held-out AUC 0.663, expected-amount MAE AED 383. Needs claim-level rows under a signed DPA; it cannot run on the v0 anonymised extract.
Denied and underpaid claims sit in AR while billing teams chase volume work and FIFO queues.
Who feels it: clinic-group CFO (cash & AR days), billing lead (appeal backlog), site managers (write-off pressure).
Failure mode: denials age past appeal windows; underpayments are posted as “paid” without variance chase; no written baseline means nobody can prove incremental recovery.
Cash / ops impact: at-risk AED compounds across payers; scarce billing minutes are spent on low-yield items; outcome economics cannot be sold without a measurable recovery loop.
Labour BPOs staff AR / denial follow-up at ~hourly directory rates. Full-cycle AI platforms bury recovery inside opaque SaaS+services. Klaim Recover overlaps economically but sits inside a revenue-treasury / Accelerate liquidity story. Few sell a crisp clinic-group overlay priced as % of incremental recovery with no payment custody.
Competitive: Klaim Recover overlaps economically; FinPro/Escrow/Focus do AR follow-up as labour. Few sell % of incremental recovery as a crisp clinic-group overlay with no payment custody.
Shared document spine for all pilots: ingest → simulated extract → confidence + evidence → human field override. Not differentiated alone; priced as enabling capability inside the recovery overlay, not a separate SaaS seat pitch.
Read claim and denial documents into structured fields.
Not a tabular problem. Needs OCR and a vision/language model, which under UAE residency must also run in-country. This is the expensive, unsolved half and TabPFN does not touch it.
Claims, remittances, denial letters and invoices arrive as PDFs, scans, and CSVs. Without structured fields + confidence + source evidence, every downstream pilot is blind.
Who feels it: RCM ops (manual re-key), analysts (broken joins), compliance (weak audit trail).
Failure mode: wrong claim IDs, missed variance lines, no snippet linking an extracted denial code to the letter.
Cash / ops impact: indirect but foundational — bad extraction multiplies false appeals and missed underpayments.
Every AI RCM platform and e-claim gateway does variants of OCR/NLP extraction. Sold as platform plumbing or EMR-embedded capture, rarely as a standalone clinic-group product with explicit confidence/evidence UX.
Competitive: Every AI platform and e-claim gateway does variants. Still required plumbing for all pilots — not differentiated alone.
Copilot drafts from denial + claim evidence; human edits/approves. Throughput lever for the recovery wedge. Pricing stays on incremental cash, not draft seats.
Draft a payer-specific appeal from the denial reason and claim evidence.
Language model territory, not tabular. Nothing is filed without authorised human approval.
Writing appeals is slow. Juniors paste templates; seniors rewrite; evidence packs are incomplete.
Who feels it: billing leads (throughput), CFOs (appeal lag → write-offs).
Failure mode: weak narratives, missed attachments, no human gate before resubmit.
Cash / ops impact: high when paired with P1 — minutes/claim drop and recovery rate rises.
Often buried in BPO labour. AI platforms may draft inside denial modules but public packaging of “appeal copilot + human approve + outcome fee” is thin.
Competitive: Thinner public packaging; often buried in BPO labour. Copilot + human gate is a clear rehearsal for recovery economics.
Rank by expected AED × probability − effort; assign top-N. Core differentiator inside P1. Makes outcome pricing credible (“we work the EV queue, not everything”).
Rank open recoveries by what they are worth pursuing.
Expected net = probability x expected AED - intervention cost. The effort and cost terms are a deterministic assumption, not a prediction, and are labelled as such.
Teams work FIFO or loudest payer. High-AED recoverable items age out while low-yield noise fills the day.
Who feels it: billing leads (queue design), CFOs (AED per hour).
Failure mode: no expected-value sort; effort ignored; gain-share pilots cannot prove focus.
Cash / ops impact: high — same headcount, more incremental cash when ranking is real.
Weakly packaged publicly. Some platforms have worklists; BPOs use tribal priority. Differentiator if transparent and tied to gain-share pilots.
Competitive: Weakly packaged publicly; differentiator if real and tied to gain-share pilots.
Lightweight taxonomy → frequency dashboard → feed P1 scoring. Sold as part of recovery overlay insight, not a BI seat. Makes gain-share conversations sharper (“we’ll work DOC+COD first”).
Infer the root-cause family of a denial when the denial code is missing, unmapped or untrustworthy.
Running here. Accuracy 0.360 against a 0.100 chance rate over 10 classes. Deliberately excludes the denial code from its own features, which would leak the target. Where a usable denial code exists the lookup is used instead, because it is exact.
Without a taxonomy, denial work is reactive case-by-case. Patterns (eligibility, coding, docs, fee) never become ops fixes.
Who feels it: RCM managers, quality leads, CFOs asking “why is denial rate sticky?”.
Failure mode: every denial treated equal; prevention backlog empty; recovery queue polluted.
Cash / ops impact: indirect alone; high leverage when linked to P1 prioritisation and P11 later.
SAMCO denials practice; Humaein/SISGAIN denial modules. Taxonomy exists inside platforms/BPO SOPs but is weakly packaged as a clinic-group product tied to recovery economics.
Competitive: SAMCO denials practice; Humaein/SISGAIN denial modules. Partial — taxonomy + recovery link is still weakly packaged for clinic groups.
Match remit → billed, flag short-pay, estimate recoverable gap, route into P1 queue. Same gain-share economics as denials. Distinguish from Klaim Accelerate (liquidity).
Estimate the expected reimbursement for a claim and flag remittances that fall materially short.
The best day-one candidate. Labels are self-generating: billed versus paid is observable from remittances without waiting for any appeal outcome. Population-level variance can be detected from aggregated extracts before a DPA exists.
Remittance advice does not always match billed/contracted amounts. Short pays post as closed; variance dies quietly.
Who feels it: finance (understated revenue), billing (no chase queue), auditors.
Failure mode: REM lines with positive variance never become recovery cases.
Cash / ops impact: high — underpayments are often easier than full denials when fee schedules are clear.
Humaein remittance/bank gaps; Klaim underpayments; FinPro “pending amount recovery”. Idea exists; packaging & outcome pricing still fragmented.
Competitive: Humaein remittance/bank gaps; Klaim underpayments; FinPro pending-amount recovery. Idea exists; packaging & pricing still fragmented.
Use coding checks as a recovery amplifier (flag after denial + before resubmit), not as the first product. Human gate on flags. Pricing remains attached to recovery outcomes, not scrubber seats.
Flag claims whose coding is inconsistent with comparable historical claims.
Deterministic pairing rules already cover the obvious cases. A model only earns its place on the combinations that are impractical to write down as rules, which is a narrower claim than it first appears.
Inconsistent CPT/ICD pairing and missing required fields become denials after submit — or silent underpayments.
Who feels it: coders, billing leads, clinicians whose notes don’t support codes.
Failure mode: high-ticket cardio/ortho lines sail with mismatched ICD; PA-required fields blank.
Cash / ops impact: preventable denials (e.g. DEN-COD-07) and rework minutes; measurable if tied to avoided write-offs.
Axora, Humaein, SISGAIN, Pure Validator and similar sell pre-submit scrubbing heavily. Crowded prevention / coding lane with opaque SaaS packaging.
Competitive: Axora, Humaein, SISGAIN, Pure Validator are heavy here (pre-submit scrubbing). Crowded prevention lane.
Triage overlay on denial feed: classify ELIG/PA/other → score recoverability → route appeal vs front-office fix. Supports P1 queue hygiene. Outcome fees only where appeal path recovers cash.
Separate eligibility and prior-auth denials that are appealable from those needing a front-office fix.
The value here is mostly process-fix rather than cash: much of what it separates out should never have been appealed. Worth building after P1, not before.
Many denials are eligibility or prior-auth failures. Treating all denials as “appeal everything” wastes minutes and burns payer goodwill.
Who feels it: front office (eligibility on DOS), billing (PA gaps), patients (unexpected balances).
Failure mode: appeals filed when auth was prospectively required; or recoverable coverage-proof cases written off.
Cash / ops impact: mixed — some AED recoverable, much value is process-fix prevention.
Embedded in full-cycle RCM & EMR gateways (eligibility checks, PA workflows). Rarely sold as a standalone clinic-group triage product with recover-vs-fix disposition.
Competitive: Embedded in full-cycle RCM & EMR gateways. Hard to own as a standalone first pilot.
Demo rule pack only (not a live rail). Use to speak eClaimLink fluently in buyer demos and to annotate recovery evidence — not as first commercial wedge. Avoid competing head-on with gateway vendors.
Validate claims against payer and e-claim submission rules.
Deterministic on purpose. Do not use a model where ordinary software is more reliable and auditable.
Dubai claims ride eClaimLink/DHPO. Rejectable patterns (segments, code pairs, auth rules) create hard rejects and rework.
Who feels it: billing submitters, IT/gateway admins, CFOs seeing reject spikes.
Failure mode: teams learn rules only after rejects; demo confusion with live rail claims.
Cash / ops impact: medium — prevention of rejects; crowded and integration-heavy.
e-claim gateways (ClinicGateway, QC Messenger) + AI scrubbers already sell payer/DHA-style validation heavily as clean-submit infrastructure.
Competitive: e-claim gateways (ClinicGateway, QC Messenger) + AI scrubbers already sell this heavily.
Age buckets + chase packs + timing metrics as an extension of recovery overlay. Explicitly not factoring. Price chase outcomes only where incremental cash is attributable; else keep as ops hygiene for P1.
Estimate time to payment and which aged receivables are still economically worth chasing.
Weakest fit. Time-series extensions of tabular models have documented failure modes on trend extrapolation and recency.
Aged open balances drain cash. Without age buckets and chase packs, ops cannot decide chase vs write-off.
Who feels it: CFO (AR days), collectors, site admins.
Failure mode: 90+ day balances ignored; chase letters generic; confusion with liquidity products.
Cash / ops impact: medium — some recoveries, but overlaps classic BPO AR.
Classic BPO AR follow-up. Klaim Accelerate is liquidity (receivables purchase), not chase ops — buyers often conflate the two.
Competitive: Classic BPO AR. Klaim Accelerate is liquidity, not chase ops — distinguish carefully in buyer talks.
Mark Phase 3+. After P1 proves cash, reuse denial taxonomy to score pre-submit risk. Still overlay; still not a scrubber price war. Outcome story remains recovery-first.
Score denial risk before submission rather than after adjudication.
Technically the same model run earlier. Parked: crowded lane, long sales cycle, and it needs the recovery model to exist first.
Intervene before submit using learned denial patterns. Valuable long-term, but crowded and not the first outcome-priced pilot.
Who feels it: coding leads, submitters, CFOs wanting denial rate ↓.
Failure mode: buying prevention SaaS without proving recovery economics first; long sales vs hospital platforms.
Cash / ops impact: medium avoided denials — hard to attribute without baseline; Phase 3+.
Axora, Humaein, Pure Validator and peers dominate pre-submit prevention / scrubbing with platform retainers.
Competitive: Axora, Humaein, Pure Validator dominate prevention. Mark as Phase 3+, not first pilot.
Explicitly locked. Prove recovery economics on P1 first. Future expand only as overlay modules — not rip-replace EMR/BPO. No full-takeover pricing in phase 1.
Not in scope.
Locked. Full RCM takeover is the land-and-expand end state, not a pilot.
End-to-end claim management — eligibility through cash posting. Attractive as land-and-expand fantasy; disastrous as first pilot.
Who feels it: buyers wanting “one throat to choke”; vendors selling full-cycle retainers.
Failure mode: multi-year replacement, hospital-bias platforms, lost focus on measurable recovery.
Cash / ops impact: theoretically high, but capital, density, and sales cycle make it wrong first move.
ACCUMED, Escrow, Humaein, SISGAIN and others sell full-cycle RCM (labour and/or AI platforms). Dominant commercial language in UAE clinic outsourcing.
Competitive: ACCUMED, Escrow, Humaein, SISGAIN and others sell full-cycle. We stay overlay until recovery economics are proven.