Market Access Data and Payer Analytics: The Systems Behind Access Strategy

The Data Assets Behind Access Strategy, and What Each One Cannot Tell You

Ask a market access team where their numbers come from and you will usually get a list of vendors rather than a list of data assets. That is the first problem. A vendor name tells you who invoices you. It does not tell you what population the file covers, how the records were collected, what was excluded, how long it takes for a transaction to appear, or what happens to the record when it is later corrected. Those five properties determine which questions the data can answer, and they are almost never written down in a place an analyst can consult before building a model.

A useful starting inventory for a mid-cap or large pharmaceutical company looks like this. Purchased longitudinal prescription and medical claims, usually from one or two syndicated suppliers. Retail and specialty pharmacy dispense data, sometimes direct from a limited distribution network. Wholesaler transaction feeds, including the sell-through detail that shows where product actually went. Chargeback claims submitted by distributors. Rebate submissions from payers and pharmacy benefit managers. Formulary and plan design files. Payer and plan reference data. Contract terms, held in a contract lifecycle system, a revenue management platform, and often a set of spreadsheets that the contracting team trusts more than either. Government price calculations, held in a separate compliance system. Internal shipment and invoice data from the enterprise resource planning system. And, increasingly, tokenized patient-level data assembled for evidence work that then gets borrowed for access questions it was never scoped for.

Each of these has a specific failure mode. The failure modes are knowable. What makes access analytics fragile is that they are rarely documented alongside the data, so every analyst rediscovers them, and some of them do not.

Purchased does not mean authoritative

There is a habit in commercial analytics of treating a purchased file as ground truth because someone paid for it. The invoice creates a psychological guarantee that the data does not earn. A syndicated claims file is a sample assembled from whichever data contributors the supplier has agreements with in a given period. Those agreements change. When a large pharmacy chain or a clearinghouse enters or leaves a supplier’s network, the apparent market shifts, and unless someone is tracking the supplier’s coverage notices, that shift will be read as a real change in prescribing behavior.

The practical control is unglamorous and it works: maintain a data asset register that records, for every file feeding an access decision, the population covered, the collection method, known exclusions, the lag from event to availability, the restatement policy, and the contractual limits on how the data may be used. Update it when the supplier changes anything. Require an analyst to cite it in any model that goes to a governance committee. This is the same discipline applied to source data in regulated work, moved to a commercial context where nobody has required it yet.

Claims Data: Lag, Capture, and the Questions It Cannot Answer

Claims data is the workhorse of access analytics, and the distinction between open and closed claims is the single most consequential thing a technology leader can understand about it. The two are not different vendors selling the same thing. They are different data products with different physics.

Open claims are captured during processing, before final adjudication, typically from pharmacies, practice management systems, and clearinghouses. They arrive quickly, often within days of the encounter. Because they are not restricted to a single payer’s enrolled population, open claims datasets can be very large; published comparisons describe open sources with sample sizes an order of magnitude or more above closed datasets.³ The trade-off is that the dollar amounts and the final payer disposition are not authoritative, and coverage depends on which data contributors participate in a given geography and channel.

Closed claims are fully paid and adjudicated. They are financially accurate for the lives they cover, and they support enrollment-based denominators, which matters for anything resembling a rate. The trade-off is lag and scope. Published work on health outcomes research notes that most commercial closed claims sources carry a lag of roughly six months, with Medicare and Medicaid research files considerably longer.³ Closed sources are also restricted to the payers who contribute, and commercial insurers see meaningful annual disenrollment, so requiring multi-year continuous enrollment shrinks the usable population sharply.

What the lag does to an access decision

Lag is not just a delay. It changes the shape of what you see. A launch tracking dashboard built on closed claims will show a curve that is six months behind the market and will keep restating recent months upward as claims complete. If the same dashboard is used to judge whether a formulary win produced uptake, the early reading will always look worse than reality, and the correction will arrive after the decision window has closed. Teams that do not model completeness explicitly end up reacting to the fill pattern rather than to the market.

The fix is to publish a completeness curve alongside the data. For each source, estimate what proportion of a month’s eventual volume is visible at 30, 60, 90, and 180 days, and apply it. This is ordinary practice in claims-based research and it is routinely skipped in commercial reporting. Once the curve exists, the reporting layer can present both the raw and the completeness-adjusted series, and the conversation about a launch trend stops being an argument about whose extract is right.

Where claims and clinical data genuinely fall short

Claims record billed events, not clinical reality. They will not reliably tell you the line of therapy, the biomarker status, the reason for discontinuation, or whether a prior authorization was submitted and denied as opposed to never attempted. Teams reach for electronic health record and lab data to fill those gaps, and it does help, but that data has its own coverage problem. A study of a large research cohort linked to health insurance claims found that a majority of patient-encounter months appeared in only one of the two sources, and that adding claims to participants with existing records contributed substantial additional service dates, diagnosis codes, and medications per person.⁴ Separately, a 2025 review of electronic health record completeness reported that in some settings 30 to 40 percent of variables are missing more than half their expected values, with overall missingness for key clinical variables commonly in the 15 to 30 percent range.⁵

The point for an access leader is not that the data is unusable. It is that the missingness is not random. Fragmented care produces gaps that correlate with the very populations access strategy is often trying to understand: patients who move between systems, patients with unstable coverage, patients treated outside large academic networks. An analysis that quietly drops incomplete records will produce a cleaner answer about a less representative population, and nothing in the output will say so.

Formulary and Plan Data: Many Sources, Different Refresh Rates

Formulary data looks like the simplest asset in the stack and behaves like one of the most treacherous. The question it answers sounds binary. Is the product covered, at what tier, with what utilization management? The reality is that the answer depends on which source you asked, which plan you meant, and when the file was cut.

For Medicare Part D there is at least a public reference point. CMS publishes prescription drug plan formulary, pharmacy network, and pricing files that contain plan information, formulary detail at the National Drug Code level, cost-share tier, and indicators for step therapy, quantity limits, and prior authorization.⁶ Researchers using these files are cautioned that the formulary information is organized by formulary identifier rather than by plan, and that linking formulary detail to a specific beneficiary’s plan requires a crosswalk that is easy to get wrong.⁷ That structural detail matters. A great deal of published and internal analysis silently assumes a one-to-one relationship between plan and formulary that does not exist.

Commercial formulary data has no such public backstop. It comes from syndicated suppliers who assemble it from plan documents, pharmacy benefit manager publications, and their own field research. Different suppliers refresh on different cycles. An effective date on a formulary change and the date the change appears in a purchased file are often weeks apart, and in some cases the file records the publication date rather than the effective date. If the analytics team treats the file date as the effective date, every pull-through measurement will be misaligned by the difference.

The three questions that make formulary data usable

Any formulary table that does not carry effective date, publication date, source, and a revision flag will eventually produce a wrong answer to a question someone cares about. Building those four columns in from the start is cheap. Retrofitting them into three years of history is not.

Gross-to-Net: Reconciliation Dressed as Measurement

Gross-to-net is the number the commercial business actually runs on. It is the bridge between what was invoiced and what the company keeps. And in most organizations it is not measured. It is reconciled, which is a different activity with a different error profile.

The size of the gap has grown to the point where the estimate itself is a material financial statement item. Drug Channels Institute put total gross-to-net reductions for brand-name drugs at $356 billion in 2024, a 7 percent increase and the slowest growth in at least a decade,¹ then at $416 billion for 2025.² Its 2026 review of manufacturer disclosures found that among the small number of companies that publish an average discount from list, the unweighted average was around negative 51.7 percent, meaning those manufacturers retained less than half of list price.² When roughly half of gross revenue is removed by a calculation, the integrity of that calculation is not an accounting detail.

Why it is a reconciliation

The components of gross-to-net arrive on different clocks, in different formats, from different counterparties, and none of them share a primary key with the invoice they relate to.

• Commercial rebates are claimed by payers and pharmacy benefit managers, usually one or two quarters in arrears, aggregated at a level the payer chooses, using the payer’s own plan naming rather than yours.
• Chargebacks are submitted by wholesalers when they sell at a contract price below wholesale acquisition cost. They arrive as transaction claims, are validated against contract eligibility, and are frequently disputed and resubmitted, which restates prior periods.
• Government rebates are invoiced by state Medicaid programs on a quarterly cycle, calculated from a unit rebate amount your own system produced from price data reported months earlier.
• Medicare discount obligations flow through plan sponsors and program contractors and land as invoices you did not originate.
• Patient support and copay assistance come from a hub or card vendor, on their schedule, in their categories.
• Fees, returns, and distribution service charges arrive from yet other counterparties on yet other cycles.

Finance closes the books monthly. None of these components are complete monthly. So the organization books an accrual based on a model, then trues it up when the actual claims arrive, then explains the variance. That process is legitimate and unavoidable. What is not unavoidable is doing it without lineage. In many companies the accrual model lives in a spreadsheet maintained by two people, its assumptions are carried forward without review, and the true-up variance is treated as noise rather than as evidence that an assumption is wrong.

What would have to be true for the number to be trustworthy

This is the question worth putting to the team that produces gross-to-net, and it is answerable. A gross-to-net figure deserves trust when all of the following hold. Not most of them.

Most organizations can satisfy two or three of these. The gap between two and six is the real state of gross-to-net governance, and it is measurable. A useful first exercise is simply to attempt the trace on one product for one quarter and record where it breaks. That artifact tends to be more persuasive to an executive committee than any maturity assessment.

Government Pricing: Where a Data Defect Becomes Legal Exposure

Everything above concerns internal accuracy. Government pricing is where the same underlying transaction data acquires legal weight, because the outputs are certified submissions to federal programs.

Under the Medicaid Drug Rebate Program, manufacturers report average manufacturer price and best price, and the basic rebate for a single-source or innovator multiple-source drug is the greater of 23.1 percent of AMP or the difference between AMP and best price, with an additional inflation-based rebate when AMP rises faster than the Consumer Price Index for urban consumers.⁸ The regulations define in detail which sales are included in AMP and which are excluded, and which prices are and are not taken into account for best price.⁹¹⁰ Those inclusion and exclusion rules are, in data terms, filters applied to transaction-level records, driven by class of trade, customer eligibility, and the treatment of fees and discounts.

That is the whole point. AMP and best price are not finance opinions. They are queries over transaction data, and the correctness of the answer depends on the correctness of the customer classification, the completeness of the discount capture, and the fidelity of the contract terms in the system. A defect in any of those propagates into a certified federal submission.

The enforcement record is a data quality record

The government pricing enforcement history reads, from a technology perspective, like a catalog of master data and calculation-logic failures. Government price reporting has produced repeated False Claims Act litigation and settlements involving how prices were calculated and how products were classified.¹¹ In one widely reported matter, a manufacturer agreed to pay $260 million to resolve allegations that it underpaid Medicaid rebates for a long-marketed product by reporting a base date AMP as though the drug had first been marketed in 2013 rather than decades earlier, and agreed to correct the base date AMP as part of the resolution.¹² Whatever the intent behind such decisions, the mechanism is a stored reference value that determines every subsequent inflation rebate calculation.

Oversight has long recognized the verification problem. A Government Accountability Office review of the Medicaid Drug Rebate Program found that program oversight did not ensure that manufacturer-reported prices or the methods used to determine them were consistent with program criteria, and that only limited checks for reporting errors were performed.¹³ The practical consequence is that the primary control over these figures is the manufacturer’s own internal control environment. Nobody else is going to catch the defect first.

340B and the duplicate discount problem

The 340B program adds a second calculation off the same base and a well-known reconciliation problem. Manufacturers are not required to provide both a 340B discount and a Medicaid rebate on the same unit, but preventing that duplicate depends on identifying which claims were 340B-eligible, which is information the manufacturer does not originate. The Health Resources and Services Administration maintains program integrity mechanisms including the Medicaid Exclusion File and audits of covered entities and manufacturers.¹⁴ A 2025 Government Accountability Office review found that agency oversight had improved but that audits did not adequately assess whether duplicate discounts were being prevented, among other unresolved weaknesses.¹⁵

The policy environment is also moving. HRSA has been working through a rebate model approach for certain drugs, with an initial pilot that was withdrawn, a Request for Information in February 2026, and then a Notice issued on 31 July 2026 establishing a rebate model under which qualifying manufacturers may provide the 340B price for selected drugs through retrospective rebate for the 2026 and 2027 initial price applicability years.¹⁶ A rebate model changes the data flow materially: instead of a discount applied at purchase, the manufacturer would process claims after the fact, which means claim-level validation, dispute handling, and a new reconciliation stream. Whatever the final shape, the technology leader’s planning assumption should be that 340B becomes more claim-driven and more data-intensive, not less.

Medicare adds another reconciliation stream

The Part D redesign that took effect in 2025 replaced the coverage gap discount with the Manufacturer Discount Program, changing both the obligation and the invoicing path.¹⁷ From January 2026, negotiated maximum fair prices took effect for selected drugs, with CMS operating a Medicare Transaction Facilitator to move dispense data between plans, the agency, manufacturers, and their processing partners, and an optional payment module to distribute refunds to dispensing entities.¹⁸¹⁹ Independent analysis of the effectuation design highlighted the operational questions this raises around data exchange, timing, and dispute handling across the pharmacy, plan, and manufacturer chain.²⁰

The architectural reading is straightforward. Each of these programs adds a claim-level or dispense-level data stream that must be received, validated, matched to internal transactions, reconciled, and retained. None of them replace an existing stream. The number of reconciliations a manufacturer performs is going up, and each one needs an owner, a service level, and an exception process. Treating any of them as a temporary spreadsheet is how organizations end up with a compliance obligation running on a personal file share.

Contract Analytics and the Assumptions Nobody Writes Down

The recurring commercial question is whether a proposed contract is worth signing. Stated plainly, it is whether the additional access purchased by a rebate produces enough incremental net revenue to more than offset the rebate, across the whole book of business affected by the terms, including the business that would have been retained anyway.

That is a modeling question, and the model is usually built. The problem is not the absence of a model. The problem is that the model’s assumptions about volume, mix, and payer behavior are rarely documented alongside it, so the model cannot be evaluated later against what actually happened, and the next model repeats the same assumptions because nobody knows they were assumptions.

The assumptions that decide the answer

A contract valuation model typically rests on a small number of judgments that dominate the output. Published guidance on payer contract analytics describes the same set: eligible lives, the expected change in utilization management and how strictly the payer enforces it, comparable past events used to size the lift, the uncertainty range around that lift, and the break-even lift required for the deal to pay for itself.

Making the model reviewable

The engineering answer here is not a better algorithm. It is provenance. A contract model becomes reviewable when four things are stored with it and survive after the decision.

1. The input snapshot. The exact extract of claims, formulary status, and lives used, with the extract date, so the model can be rerun on the same inputs. Rerunning a model against a refreshed source and getting a different number teaches nothing about the model.
2. The named assumptions. Each judgment stated as a value with a rationale and an owner, in a structured field rather than a comment in a cell. Assumptions in comments do not survive a copy of the workbook.
3. The scenario set. A small number of distinct, plausible states rather than a single point estimate, with the break-even lift shown explicitly so reviewers can judge whether it is achievable rather than debating a forecast.
4. The commitment to measure. A defined post-deal read: which metric, measured on which source, at which lag, compared against which scenario. Written before signing, not after the results are known.

One caution on post-deal measurement. Attributing an observed change in share to a formulary event requires a counterfactual, and the honest version of that analysis will produce a range rather than a number. Reporting a single attribution figure with no interval implies a precision the data cannot support, and it trains the organization to expect certainty from an analysis that cannot deliver it.

Master Data: Payer, Plan, Product, and Relationships That Keep Moving

Every problem described so far reduces, at some point, to identity. Which customer is this. Which plan does this claim belong to. Which contract governs this transaction. Which product configuration is this NDC. Market access is unusually hard on master data because the entities themselves are unstable and because the relationships between them change more often than the entities do.

Four domains, and the joins between them

Payer, plan, product, and customer are usually treated as four master data domains. In access analytics the relationships carry most of the meaning and get the least governance.

The relationship row is the one that gets missed. Master data programs tend to be scoped as entity resolution: deduplicate the customer file, build a golden record. That work is necessary and it is not sufficient here. If a plan moves from one formulary to another mid-year and the model has no effective-dated relationship, the historical view will be silently rewritten to reflect the current state, and every prior period comparison becomes wrong without any error appearing.

Effective dating is the requirement, not a refinement

Access master data must be bitemporal in the places that matter. Two dates: when a fact was true in the world, and when the system learned it. That sounds like an architectural luxury until the first time someone asks why last quarter’s board number changed. With effective dating and knowledge dating, the answer is a query. Without them, it is a forensic exercise across extract files.

Practically, this means the payer, plan, and customer hierarchies need history tables rather than overwrite-in-place updates, and reporting needs to declare which date basis it is using. It also means resisting the common shortcut of resolving hierarchy at report time from the current state. The current state is the right answer to exactly one question: what is true now.

The Governed Layer: Architecture, Ownership, and Privacy Constraints

The last question is the practical one. What actually belongs in a governed layer for market access data, and what can safely stay in the hands of analysts working quickly?

What belongs in the governed layer

The test is consequence. A dataset belongs in the governed layer when a defect in it produces a financial statement error, a certified regulatory submission error, a contractual obligation error, or a privacy exposure. That yields a shorter list than most data platform programs assume, which is a feature. A governed layer that tries to contain everything gets bypassed.

• The transaction store that feeds both commercial reporting and government price calculations. Change controlled, lineage tracked, retained.
• Contract terms as executed, with amendment history, and the mapping from terms to the eligibility and calculation logic that uses them.
• Customer master with class of trade and affiliation, effective dated, because it drives both chargeback eligibility and statutory inclusion.
• Product master including NDC and package configuration history.
• Payer and plan master with effective-dated relationships to formulary and to contract.
• Gross-to-net component ledger: accrual assumptions, actuals, and variance by component and contract, versioned.
• Government price calculation inputs and outputs, with the ability to reproduce a prior submitted value from the data as it stood at submission.

Everything else, including exploratory cohort work, scenario modeling, and field-facing analytics, can live in a faster layer that reads from the governed one. The critical rule is directional: the fast layer consumes governed data, and results that will drive a contract, an accrual, or a submission must be reproduced from governed inputs before they count. This distinction, sometimes framed as separating the system of record from the system of analysis, is the difference between a platform people use and a platform people route around.

Ownership: the part that is not technical

Market access data has an ownership problem more than a tooling problem. Finance owns the accrual. Contracting owns the terms. Commercial analytics owns the demand view. Government pricing compliance owns the statutory calculations. Legal owns the exposure. IT owns none of the content and all of the plumbing. In that arrangement, the reconciliation between components is unowned by default, which is precisely why it degrades.

The workable pattern is to name an accountable owner per data object rather than per system, and to make the reconciliation itself a named deliverable with a named owner. Someone owns the customer master. Someone owns the mapping from contract terms to calculation logic. Someone owns the monthly reconciliation between the commercial net revenue view and the general ledger, and reports the variance whether or not it is comfortable. Governance in this domain is mostly a matter of assigning those names and then not quietly reassigning them when the person leaves.

Privacy constraints shape what analysis is permissible

Patient-level data in access analytics is subject to constraints that are frequently misunderstood inside commercial teams, and misunderstanding them is how an organization acquires a problem it did not know it was buying.

Under the HIPAA Privacy Rule there are two recognized routes to de-identification: the Safe Harbor method, which requires removal of a specified list of identifiers, and Expert Determination, in which a qualified person applies statistical principles and documents that the risk of re-identification is very small.²¹ Both are documented processes with artifacts. Neither is achieved simply by removing names, and tokenization on its own does not constitute de-identification: tokenization enables linkage across sources, and the resulting linked dataset still has to satisfy one of the two methods.

Three constraints follow for architecture.

These constraints are not obstacles to access analytics. They are design inputs. An architecture that encodes permitted use and join eligibility as policy lets analysts move faster inside the boundary, because the boundary is machine-checked instead of relitigated in every project. The organizations that struggle are the ones where the rules exist only in a legal reviewer’s memory.

Conclusion

Market access is the part of a pharmaceutical business where the data is least governed and the consequences are most concrete. The gross-to-net figure is a reconciliation across systems that were never built to agree, and its credibility depends entirely on whether the assumptions inside it are named, versioned, and challenged. The government pricing calculations that sit on the same transaction data are certified submissions, which means a master data defect or an untested change to class-of-trade logic is a compliance matter rather than a reporting inconvenience. The contract models that shape commercial strategy are usually sound arithmetic resting on undocumented judgments about volume and mix, which is why the organization rarely learns whether its assumptions were any good. And underneath all of it sit payer, plan, product, and customer relationships that change constantly and are usually stored as though they do not.

None of this requires a transformation program. It requires a small number of specific commitments: a data asset register that records what each source can and cannot support, effective-dated relationships in the master data, a traceable path from executed contract terms to calculation logic, an accrual-to-actual variance tracked by component rather than blended away, provenance stored with every contract model, and permitted use encoded as policy rather than remembered. Each of those is achievable in a quarter or two. Together they change gross-to-net from a number the business hopes is right into a number it can defend.

Sakara Digital works with pharma and biotech organizations building the data foundations underneath commercial and access analytics. If you are working through gross-to-net traceability, government pricing data controls, or the master data model behind payer and plan reporting, and you want an independent perspective on where to start, we are happy to have that conversation.

For Further Reading