Upload a PBM claims export. RxAiAnalytics reprices it against the contract you actually signed, scores the guarantees, identifies the opportunities, and publishes the result to the people who need to see it, with a pharmacy team's sign-off on every number a client sees.
Claims, contracts and plan documents, tagged to a client on the way in. Reversals are netted before anything is counted.
AI maps the file's columns and extracts the contract's terms. Every extracted number is checked back against the document's own text.
Terms and plan rules stay proposals until a pharmacy team member confirms them. Until then, every figure measured on them is labelled preliminary.
A deterministic engine does all the claims-versus-contract arithmetic. AI writes the narrative over the numbers, never the numbers.
PDF reports, a book-of-business view, and a client console configured broker by broker, with your talking points pinned to the charts.
Drop the file your PBM already sends you. The platform reads its headers, maps them, nets the reversals, and starts the analysis in the background.
Pharmacy claims, medical claims, or a summary and HCC report, each tagged to a client record at upload time. Pick a client name that does not exist yet and it is created on the spot.
The file's own column headers are read and mapped to canonical fields. No mapping spreadsheet, no PBM-specific template to maintain.
AWP discounts, rebate rates by channel, trend and member counts go in at upload when the file lacks them, and can be revised later without re-uploading. The dashboard recalculates and tells you when nothing moved.
Attach a per-NDC rebate schedule or a signed copay-assistance (MAP) agreement in the same flow. Both are parsed and applied to the matching claims.
Files up to 200 MB upload directly to the cloud and ingest in the background, with named stages so you can see where a file is.
Reversal and adjustment rows are netted before anything is counted, and control totals are captured on every upload so the figures on screen can be traced back to the file.
The Spend Analysis tab shows plan spend the way a plan sponsor experiences it, with the arithmetic behind every figure one click away.
Contract audit measures confirmed contract terms against actual claims, bucket by bucket, and shows what was excluded rather than quietly dropping it.
| Bucket | Guarantee | Realized | Result |
|---|---|---|---|
| Retail 30 · Brand | 18.0% | 18.4% | Pass |
| Retail 30 · Generic | 84.0% | 83.8% | At risk |
| Retail 90 · Brand | 22.0% | 22.3% | Pass |
| Retail 90 · Generic | 86.0% | 83.9% | Fail |
| Mail · Brand | 24.0% | 24.6% | Pass |
| Mail · Generic | 87.0% | 87.2% | Pass |
| Specialty · Brand | 19.0% | 19.1% | Pass |
| Specialty · Generic | 40.0% | — | No data |
Upload a PBM, TPA or stop-loss agreement as a PDF. The terms come back structured and quoted, and nothing reaches an engine until your pharmacy team says so.
Contract Analysis, Contract Comparison, Claims Audit with no contract needed, and Contract + Claims Audit together.
Discount guarantees, rebate rates, exclusions, fees and dates come back as structured terms rather than a summary paragraph.
Every extracted number is checked back against the contract's own text. Anything that does not reconcile is flagged for review instead of being reported.
Figures measured before your team confirms are labelled preliminary. The raw extraction is kept immutable beside the confirmed version, so the review trail is always visible.
Type "Specialty AWP discount is 19%" and it becomes a validated structured term for you to accept, reopen, or amend.
Re-price your actual claims under one or more confirmed contracts to compare proposals on the plan's own utilization.
Earned versus collected, with remittance statements parsed from PDF, Word or Excel. Only receipts a person has accepted are counted.
An executive summary, a dimension-by-dimension breakdown, the top opportunities, and the terms already working in the plan's favor.
Order a fresh AI read of a contract whenever a version changes, and compare the extraction to the terms already confirmed.
Specialty is where the dollars are. The platform identifies biosimilar and lower-cost opportunities, shows the rebates a switch would forfeit, and keeps the clinical guardrails visible.
Specialty plan spend, claim rate, utilizers, share of total, rebates and forecast, on one screen.
Biosimilar utilization and the gross and net-of-rebate savings per opportunity, so a switch is judged on what the plan would actually keep.
The rebates the plan would stop earning by switching are shown next to the gross saving, never left out of the comparison.
When there is no diagnosis in the data, what a drug is treating is inferred from co-fills, dosing cadence and prescriber specialty. A tie reads Unknown.
Opportunities the clinical rules block are shown with the reason and the forgone dollars, rather than silently dropped.
Mark each opportunity Reviewed, Sent for clinical review, or Dismissed. The numbers recompute and alternative paths close together.
Models the assistance a plan could capture, compares it to reported actuals, audits the vendor fee, and flags when your answer at upload disagrees with the claims.
The platform identifies and models. Clinical and plan decisions stay with the plan and its advisors, and the vocabulary is enforced in every screen and every AI reply.
Member-level views stay tokenized, medical context stays in its own lane, and plan documents are read by AI and confirmed by the pharmacy team.
Total members, Rx utilizers, utilization rate, average fills per member and average copay per fill.
Identifies the small share of members driving most of the spend, with identifiers tokenized and masked in the interface.
Every medication in the file, enriched with drug class and typical use.
A separate, clearly labeled panel for medical-benefit drug spend, loss ratio and members with medical claims. Medical data never changes a pharmacy number, by design.
Plan Summary / SPD, Formulary / Drug List, Exclusion List, HDHP Preventive Drug List, ACA Preventive Drug List, Copay Assistance (MAP) Contract, and MAP Savings Report. AI extracts; the pharmacy team confirms or reopens.
Checks what the plan document says against what the claims actually did, including first-dollar cost share on preventive drugs.
A CSV linking tokenized member identifiers across pharmacy and medical files, for the analyses that need both.
Program type, fee structure, savings definition, member cost share, audit rights and HDHP eligibility, read straight off the signed agreement.
A browsable reference of specialty molecules, brands, biosimilars and generics, with the switch rules that connect them and the benchmark cost of each rung.
The assistant answers about your own uploads, counts rather than estimates, and runs under a HIPAA Business Associate Addendum with customer content excluded from training Anthropic’s models.
"Where are the biggest savings opportunities?" "Which specialty drugs have biosimilars?" Answered from the claims you loaded, not from a generic model of a plan.
The assistant queries your claims by drug, condition, tier, channel or month to get real numbers before it answers. Where its detailed breakdowns are enabled it can also cut by plan, prescriber specialty, age band, gender and more, reporting exact counts for every group, however small. Member-level detail is a separate opt-in, off by default.
FDA approval status, safety alerts and biosimilar availability are pulled into the answer when the question needs them. Web search is off in conversations that include member-level detail.
Savings opportunities, specialty spend drivers, biosimilar savings, high-cost members, NADAC benchmarking and an executive summary. When contracts are selected, a contracted-discount check takes the first slot.
Each answer lists the claims queries it ran, the file-summary figures it read and any web sources it cited, so the numbers behind it can be checked.
All AI traffic runs through a HIPAA-configured organization in the United States. Customer content is contractually excluded from training Anthropic’s models, and every AI disclosure is logged for accounting-of-disclosures reporting. Optional member-level detail, where allowed, is a limited PHI disclosure under that Addendum: members are identified only by their masked MBR ID, never by name or the member IDs in your files, and every member included is logged.
Deliverables leave the platform as PDFs and as a live console configured broker by broker, with your guidance attached to the numbers.
Point an SFTP feed at the platform and every pulled file runs the full pipeline without anyone logging in. Silence is treated as a signal.
Connection testing, run history, and filename de-duplication so a re-sent file is never double counted.
Ingest, enrichment, audit and alert run end to end on every pulled file.
"Nothing arrived" is an alert, not silence. Overdue feeds surface in the pharmacy team's view.
Human-in-the-loop review, amendment and confirmation of extracted terms in one back-office screen.
Rolls the audit across the contract year rather than a single file, so a quarter's shortfall is seen against the year.
Findings and failures go to the pharmacy team by email, including audit red flags on a freshly loaded file.
The arithmetic is deterministic, the AI is kept out of it, and every completed upload is re-checked automatically before anyone relies on it.
All claims-versus-contract arithmetic lives in a single pure module with no AI in the path. AI writes narrative over numbers that are already computed, never the numbers themselves.
Ingest totals are reconciled against the database, pinned invariants are re-asserted on the live figures, and the figure set is fingerprinted so identical inputs must produce identical figures.
Unreadable columns, exclusion rules that cannot be compiled, and unpublished benchmarks are disclosed and excluded from measurement rather than assumed.
Every comparison says which way is good. The contracted figure is a target, so realized above it is the plan winning, and only the figure is tinted.
What is in place today and what is still in progress, stated the same way on our public security page.
TLS 1.2+ in transit with HSTS. At rest, AWS KMS customer-managed keys, with separate keys isolating PHI from audit logs. Sensitive fields encrypted AES-256-GCM under a per-batch key that exists only in memory during upload.
Member identifiers are tokenized with keyed HMAC-SHA256 so analytics can group and count without the raw identifier, and PHI is masked in the interface.
Mandatory MFA with an authenticator app, role-based least-privilege access re-checked on every request, and deactivation that takes effect on the account's very next request.
Every access to PHI is written to a tamper-resistant log that blocks updates and deletes, dual-written off-box, and supports HIPAA accounting-of-disclosures reporting, including disclosures to the AI subprocessor.
Multi-region CloudTrail, GuardDuty, Security Hub, AWS Config and a web application firewall, with alarms that page the Security Officer on bulk PHI access or authorization denials.
Private subnets, no public IPs on application containers, a PHI database unreachable from the internet, Multi-AZ, and production deploys only from the main branch through keyless CI/CD.
Doing and approving are separate seats, your pharmacy team sets what each field broker can see and do, and RxAiAnalytics staff hold only the platform-level super admin role. Pricing counts only the two working seats.
Every new account starts here and sees nothing until it is approved: by an admin of its organization, or by RxAiAnalytics staff, who place a self-registered account in its organization and approve each new organization’s first admin.
Works the uploads assigned to them with the full toolset and the assistant, and uploads contracts and plan documents. Does not upload claims files, create clients or confirm terms.
Sees every upload in its own organization, confirms what becomes client-facing math, configures every broker's view, and invites and approves its analysts and field brokers.
RxAiAnalytics staff only, never a customer seat: creates each customer organization, approves its first admin, and alone deletes accounts. Its access to your data is our workforce access under the BAA.
Uploads claims (on unless the pharmacy team turns it off) and creates clients, sees its own uploads plus any the practice shares, and works within the console limits the practice sets.
Send a single employer group's pharmacy claims export and see what the platform finds, from the guarantee board to the console your broker would present.