AI in GP: Where the Line Belongs

Artificial intelligence (AI) is everywhere right now and Government Pricing (GP) is no exception. We have heard people wondering if there is a way to use it for GP calculations. AI has real potential to improve efficiency in certain areas, but at The Pricing Group, we believe relying on it for the actual calculations and compliance determinations carries significant risk.

The core issue is that GP calculations aren't a pure data function. They're an interpretation of overlapping regulatory and legislative requirements that demands a working understanding of regulatory nuance, product and contract specifics, distribution channels, and, perhaps most importantly, accountability for the result.

Government Pricing Requires Judgment

Many GP calculations depend on nuanced interpretation of regulations, agency guidance, and company-specific facts – things such as how to treat a payment to a customer, whether a sale is bundled, whether a product qualifies as a Medicaid “line extension.” Answering these questions correctly requires company- and product-specific knowledge and reasonable assumptions layered on top of regulatory interpretation, legislative history, and precedent, not a pattern match against prior outputs.

Historical Data Can Reinforce Existing Errors

AI models learn from the data and decisions fed into them, drawn from countless and often unverifiable sources.  Some of it may be accurate but the quality of the data cannot be validated. If that underlying data is outdated or wrong, or reflects rules that are no longer in effect, the model has no inherent way to know that. Without strong governance and continuous updating, an AI tool can just as easily reflect yesterday's regulations as today's. Additionally, legacy misclassifications or outdated assumptions don't stay contained. They compound with every calculation that relies on them. 

AI Lacks Accountability

An auditor's first question is almost always, "Show me how you arrived at this result." Manufacturers need to produce source data, calculation methodologies, the reasonable assumptions applied, and the review and approval trail behind it.

AI, by contrast, is a black box. If you can't explain the logic behind a number, you can't defend it. And the accountability doesn't shift just because a tool was involved. Government agencies hold the manufacturer liable, not the AI vendor. If a system misclassifies a transaction, applies the wrong methodology, or mishandles a price concession, the manufacturer bears the consequences that might include over- or underpayments of rebates, fees/penalties, restatements, and potential False Claims Act exposure.

An AI tool cannot certify a calculation, sign off on a reasonable assumption, or sit across from an auditor and defend a position. A person has to do that which means a person has to actually understand how the number was built.

Confidentiality and Data Security

GP calculations are based on proprietary information – contract terms, customer-specific pricing, etc – and the outputs (AMP, Best Price, etc.) are among the data manufacturers protect most fiercely. Before any of that touches an AI tool, a manufacturer needs to fully understand where the model is hosted, how data is retained, who has access, and whether inputs are used to train future outputs. Without those answers, sensitive data is at risk of exposure well beyond the company's walls, a particular concern for publicly traded manufacturers.

So Where Can AI Actually Help?

Used well, AI is a support tool, not a substitute for judgment, the same way I might use it to help rough out a vacation itinerary. In GP, that looks like using AI to help identify relevant legislation, guidance, or applicable sections of a regulation as a starting point for research. Even there, any company- or product-specific data run through an AI tool should be scrubbed first to keep it confidential, and outputs, especially summaries of guidance or regulations, should be verified, not taken at face value.

The manufacturers who benefit most from AI over the next few years won't be the ones who let it replace their GP expertise. They'll be the ones who use it to make their experienced GP professionals faster and sharper, while keeping the judgment, and the accountability, directly with the people who can defend it.