Why GLP-1 Demand Breaks Traditional Forecasting Models

A planner using a stationary time-series method is being asked to forecast a demand system that keeps changing its own rules. That is the practical problem behind demand forecasting for GLP-1 cold chain: the volume is large, but size is not the most difficult feature. The harder issue is that patient starts, payer access, formulation mix, and cold-chain constraints do not move like a stable seasonal pattern with a noisy trend line.

The GLP-1 category has three structural traits that matter more than any single market-size number. First, the patient base compounds: J.P. Morgan projects roughly 10 million Americans on GLP-1s in 2025 and about 25 million by 2030, a 2.5x increase over five years rather than a one-quarter spike that later clears from the system.[1] Second, demand is exposed to regime shifts: Medicare pricing, cash-price moves, oral launches, and patent expirations can reset access and substitution patterns. Third, the cold chain is physically unforgiving. A few percentage points of forecast error do not merely create awkward inventory; they can become validated-packaging shortages, lane capacity misses, excursions, frozen product, or unusable stock.

The Patient Base Does Not Reset

Many conventional demand models are built to be forgiving when demand is noisy but fundamentally stable. They can absorb a promotion, a short allocation period, or a recurring seasonal prescription pattern because the underlying system eventually resembles its past. GLP-1 demand is less cooperative. New users layer onto an already-growing chronic therapy population, while discontinuation sits beside that growth as a separate variable rather than a clean offset.

That layering is the point often lost when GLP-1 growth is described only as a surge. SupplyChainBrain, citing Anaplan’s Emily Nicholls, describes the dynamic plainly: “demand compounds over time as new patients layer on top of a growing base.”[2] For forecasting, that sentence is more useful than a decorative total-addressable-market slide. It says that starts and continuing therapy have to be modeled separately. A moving average that compresses both into one smoothed signal can look calm precisely when operational load is becoming more brittle.

A chronic therapy base changes the meaning of forecast error. If a model underestimates new patient starts, the error is not confined to the launch month. Those patients may continue to require refrigerated fulfillment in later periods, so the miss propagates into future demand, inventory positioning, packaging needs, and service-level exposure. If the model overestimates persistence, the error also persists, but in a different form: finished goods, packaging components, lane bookings, and specialty distribution capacity may be committed against demand that does not materialize.

Affordability is the churn variable that keeps this from becoming a simple adoption curve. PwC’s June 2024 GLP-1 research reported that 45% of respondents taking GLP-1 drugs said the drugs were difficult to afford.[3] KFF’s Fall 2025 survey put a related affordability measure higher, at 56%, which reinforces the same point while showing that the exact percentage is method-dependent. These figures do not prove a discontinuation rate by themselves, and they should not be treated as one. They do say that persistence cannot be assumed from clinical eligibility or interest alone. A cold chain planner needs a forecast that can separate patient acquisition from affordability-driven continuation risk, because each produces a different operational consequence.

This is where a familiar smoothing method becomes misleading. A moving average is not neutral just because it is familiar. It encodes a view that recent demand is the best available summary of near-future demand. That assumption is weak when the installed patient base is expanding, when new starts are driven by social adoption and access changes, and when discontinuation is tied to out-of-pocket burden. The model may report a tolerable error rate while hiding the source of the error from the people who must decide how much refrigerated inventory to stage, where to stage it, and which packaging lanes need validation.

Market Size Is Context, Not the Forecast

Large market projections are useful mainly because they confirm that the category is not a transient niche. They are less useful as planning inputs. Grand View Research projects the GLP-1 receptor agonist market at $185 billion by 2033, while J.P. Morgan discusses a $200 billion incretin market by 2030.[4][1] IQVIA frames obesity market growth in a broad $105 billion to $200 billion range for the later 2020s.[5] Those figures differ in scope, timing, and definitions, which is exactly why they should not be mistaken for a cold chain forecast.

For a manufacturer, 3PL, or specialty distributor, the planning question is narrower and more demanding: which formulation, which patient cohort, which payer environment, which geography, which season, which validated shipper, and which service level? A top-line market number does not answer whether refrigerated demand shifts from one lane to another, whether a cash-pay channel changes order cadence, or whether an oral product reduces injectable volume in one segment while expanding total treated patients in another.

That distinction also matters when teams evaluate planning technology. Industry-specific demand behavior determines capability requirements; a life-sciences planning problem is not just a retail replenishment problem with colder boxes. The relevant comparison is less “which tool forecasts growth?” and more “which framework can represent patient accumulation, access changes, formulation substitution, and cold-chain execution constraints in the same planning view?” That is the same reason broader discussions of AI demand planning software by industry should be read through the operating model, not through generic forecast-accuracy claims.

The Category Moves Through Regimes, Not Just Trend Lines

IQVIA’s outlook is useful because it does not treat the obesity market as one continuous curve. It describes three broad eras: rapid adoption from 2024 to 2026, a price-and-access-driven period from 2026 to 2028, and a more generic-driven period from 2028 onward.[5] That framing fits the operational problem better than a single growth-rate assumption. Each era changes the rules under which demand is formed.

The 2026 access changes are not minor noise around a stable baseline. J.P. Morgan reports that Medicare Part D coverage for selected obesity drugs is expected with a $50 monthly cap, and it identifies oral launch timing for a Wegovy pill in January 2026 and orforglipron in April 2026.[1] pharmaphorum also reported a $149-per-month cash-price signal for the Wegovy pill.[6] A model trained only on past injectable prescription volume cannot infer those discontinuities from the time series itself.

Oral GLP-1s deserve careful handling in a cold chain forecast. They may reduce refrigerated exposure for some demand if patients shift from injectable to pill formulations. That does not make the planning problem cleaner. It splits the category into cold-chain and non-cold-chain regimes, with substitution, expansion, and channel effects that may not move together. If oral access expands the total treated population while injectable demand remains high in specific patient groups, a forecast that simply subtracts oral volume from cold-chain volume will be too blunt.

Patent timing adds another discontinuity. IQVIA notes that semaglutide goes off patent in China, India, Brazil, and other major markets in 2026, covering markets with roughly 38% to 40% of the global obese population.[5] That is not a small exogenous footnote for global planning. It can change the geography of demand, the economics of access, and the balance between branded, generic, injectable, and oral supply assumptions.

Traditional forecasting approaches fall short here for a specific reason: they are often strongest when the future is a noisy continuation of the past. Event-sensitive methods can be useful, but only when the events are represented explicitly rather than treated as anomalies after the fact. That is also where claims about AI in planning need discipline. Evidence on AI in supply chain planning is relevant only if the model design can expose assumptions about access, persistence, formulation split, and cold-chain capacity. A more complex algorithm that still treats prior shipments as the main signal remains underbuilt.

Cold Chain Turns Forecast Error Into Physical Loss

The demand mechanics would be difficult even without refrigeration. Cold chain makes the consequences sharper. GLP-1 products commonly require 2-8°C handling, and cold chain packaging must be validated by lane and season rather than assumed as a generic parcel configuration.[7] A lane that works in one temperature profile may not be valid in another. A planning miss can therefore show up as a packaging validation problem, not merely a warehouse inventory problem.

The lower bound matters as much as the upper bound. Pelton Shepherd states that GLP-1 drugs have zero freezing tolerance and that temperatures below 0°C can damage peptide structure in ways that may not be visible.[8] That is a materially different risk from a product that can tolerate brief freezing or show obvious damage. If a package freezes in transit, the inventory problem is not just “late” or “less fresh.” It may be product that cannot be safely used.

This is why aggregate demand accuracy can be a poor comfort metric. A national forecast may be close while a specific refrigerated lane is short on validated pack-outs during a heat wave, or while a winter route carries more freeze risk than the plan assumed. Forecasting has to reach the operational grain where product actually moves: lane, season, package, dwell time, carrier performance, order cadence, and service-level commitment.

Emerging-market execution adds another layer of uncertainty. pharmaphorum reports mishandling rates of up to 20% in emerging markets and links that risk to roughly 25% overproduction.[6] Those figures should be treated as reported logistics risk, not as a universal rate for every GLP-1 lane. Still, they make the operational point clearly: when cold chain reliability is uneven, forecast error can drive intentional buffers that become expensive, wasteful, or insufficient depending on where demand actually appears.

Manufacturing capacity also cannot be detached from demand quality. CRB describes the GLP-1 boom as a capital and manufacturing challenge, with companies facing pressure around capacity, speed, and investment decisions.[9] A forecast that overstates durable demand can support overbuilding or overproduction. A forecast that understates demand can leave fill-finish, packaging, distribution, or cold-storage constraints exposed at exactly the point access expands.

What a GLP-1 Cold Chain Forecast Has to Represent

No source in the available research provides a validated, end-to-end GLP-1 cold chain forecasting methodology. That boundary matters. The practical answer is not to pretend that a named model or vendor category has already solved the problem. The better standard is to ask whether the forecasting framework represents the demand system that actually exists.

• Non-stationary patient accumulation: new starts, continuing users, dose progression, and discontinuation should not be collapsed into one smoothed demand signal.
• Affordability-linked churn: out-of-pocket burden and access changes should modify persistence assumptions rather than sit outside the forecast as commentary.
• Event-sensitive demand shifts: Medicare caps, cash-price changes, oral launches, and patent expirations need explicit scenario treatment before they appear in shipment history.
• Formulation split: injectable cold-chain demand and oral non-cold-chain demand should be modeled as interacting regimes, not as one interchangeable volume pool.
• Cold-chain feasibility: lane, season, validated packaging, freeze risk, excursion risk, and service level should constrain the plan rather than being checked only after the forecast is approved.

The last requirement is where many planning reviews become too clean. A demand forecast can look statistically acceptable and still be unusable for the person responsible for execution. If the model says volume will rise but does not say where refrigerated demand shifts, which lanes need validated packaging, which markets face access-driven starts, or which segments may churn because of affordability, the cold chain team inherits ambiguity as operational risk.

A capable framework also needs governance around assumptions. The Medicare cap should be visible as an access assumption. Oral launch timing should be visible as a formulation-mix assumption. Patent expirations should be visible as geography and competition assumptions. Affordability should be visible as a persistence assumption. Cold-chain limits should be visible as feasibility assumptions. If those drivers are buried inside a single adjusted forecast number, planners cannot tell which part of the plan to challenge when reality diverges.

That is the bridge from forecasting to execution. Many organizations are not short on pilots, dashboards, or statistical outputs; they are short on structured ways to turn model signals into accountable operating decisions. The same issue appears in the broader supply chain AI execution gap: capability does not matter much if it does not change how decisions are made, reviewed, and constrained.

For GLP-1s, extrapolating recent prescription volume is structurally underbuilt. The category needs forecasting that treats demand as non-stationary, event-sensitive, churn-aware, formulation-split, and cold-chain-constrained. Anything less may still produce a curve. It just may not produce a plan that survives contact with refrigerated execution.

References

1. How Supply and Demand for Weight Loss Drugs is Playing Out in 2026, J.P. Morgan
2. The GLP-1 Effect: How Weight Loss Drugs are Reshaping Food and Pharma, SupplyChainBrain
3. What is the future of GLP-1 trends, PwC, June 2024
4. GLP-1 Receptor Agonist Market Size, Grand View Research
5. The outlook for obesity from 2026 to 2030, IQVIA, April 2026
6. GLP-1 pricing and supply: Examining the signals behind the numbers, pharmaphorum
7. Shipping GLP-1 Drugs: Cold Chain Requirements, Mercury
8. The Importance of Cold Chain Shipping for GLP-1 Drugs, Pelton Shepherd
9. The GLP-1 boom: Challenges, trends, and CapEx strategies, CRB