Modality COGS Benchmark

1. What it sets in the model

The rNPV pro-forma computes operating cash flows as:

Before this disclosure the default cogs_pct was 0 — modality-blind — which silently overstated rNPV for biologics and cell/gene therapies where manufacturing cost is material. The wizard now seeds cogs_pct from the table below. SG&A defaults to 0 and is user-supplied; the COGS prior implies no SG&A assumption. Set SG&A explicitly for a complete operating margin.

2. Constructed ratio, not a 10-K margin

Each value is a constructed, disclosed ratio:

Why not a 10-K gross margin? Two reasons a CMO will raise immediately:

1. 10-Ks report blended portfolio margins, not per-asset COGS. A large-cap’s “cost of sales ÷ revenue” mixes dozens of products across modalities, with royalties, inventory step-ups, write-offs, and idle-capacity absorption baked in. It is not the COGS% of any single asset. (Merck FY2024 gross margin 76.3% is a company number.)
2. Pre-profit cell/gene companies show launch-distorted COGS. A newly-launched autologous or AAV product running a half-empty suite reports COGS at 60–90% of net sales purely from under-absorbed fixed cost and low-yield early batches. That is a capacity-utilization artifact, not steady-state economics. These values deliberately use steady-state manufacturing cost (mature yields, loaded suite) to avoid importing that distortion.

The antibody nuance. Naively dividing bulk drug-substance cost per gram by net price per gram gives every biologic a ~1–3% COGS floor, because marketed mAbs sell for roughly $8,000–12,000/g while drug-substance costs $50–150/g. That ratio is real but misleading: it omits fill-finish, QC/release, stability, scrap and yield loss, freight/cold-chain, royalties, reserves/returns, and idle-capacity absorption — items that dominate biologic product COGS. The central values below are anchored to fully-burdened product economics and cross-checked against mature blended-product gross margins, with the bare drug-substance/price ratio visible as the floor of each band.

3. Benchmark table (7 modalities)

All values as of methodology@2026-05-20. Both the manufacturing cost anchor and the reference net price are disclosed; the arithmetic is explicit. Override per asset with asset-specific cost and price.

Table 1. Per-modality COGS benchmark. Central = cost ÷ reference price (both disclosed above). Band reflects defensible range under variation in cost assumptions and price point. Peptide and bispecific cost figures carry lower confidence (see Limitations). Steady-state; does not reflect launch-year under-absorption. Source: modality COGS research file, 2026-05-20.

4. COGS% is price-dominated — the prior is coarse

The most important single fact about these numbers: COGS% is determined far more by the denominator (price) than the numerator (manufacturing cost). Across the table, manufacturing cost per course varies by roughly 4 orders of magnitude (cents to $100k); net price per course varies by roughly 5 orders (dollars to $3.5M). Most of the cross-modality COGS% spread comes from price, not cost.

Two concrete illustrations of price-dominance:

• CAR-T. Autologous CAR-T manufacturing costs ~$80–100k/dose whether the product is priced at $400k or $1M. At $400k list (possible future commodity-CAR-T), COGS% ≈ 24%. At $1M (hypothetical premium indication), COGS% ≈ 10%. Same factory, same cost, very different ratio.
• Gene therapy. The same ~$50k/dose AAV manufacturing cost is ~2–5% of a $2.1M Zolgensma course but ~12–25% of a $400–850k course. A future gene therapy priced like a specialty biologic ($300–500k) with a $100k dose would be ~20–30% COGS — CAR-T-level — from the same manufacturing platform that produces a "2% COGS" result at ultra-high pricing.

Implication: the correct modelling approach for a specific asset is to anchor to the manufacturing cost for that modality and derive COGS% from the asset’s own projected net price — not to take the central as a fixed input. The defaults here are a starting prior for assets where cost/price is not yet known; replace them when you know either figure.

5. Gene therapy: bimodal band and re-centered default

Gene therapy has a bimodal COGS% distribution because price varies by more than 4× across today’s approved products ($850k–$3.5M per course) while manufacturing cost per dose spans only ~10× ($10k–$100k):

• Ultra-priced (Zolgensma $2.1M, Hemgenix $3.5M): COGS% ~1.4–5% — below monoclonal antibody.
• Mid-priced (Luxturna ~$850k bilateral): COGS% ~6–12%.
• Lower-priced (~$400–500k hypothetical, $100k dose): COGS% ~20–25% — CAR-T-level.

The band 2–25% is genuinely bimodal, not a continuous distribution. Reporting a single central is an acknowledged simplification.

The auto-fill default (13%) is deliberately re-centered from the price-weighted central (~8%, which weights toward the marquee ultra-priced products) to a mid-reference price (~$600k) to avoid understating COGS for a typically-priced gene therapy. Gene therapy must never be ranked above monoclonal antibody on COGS% at today’s prevailing price points. If the asset being valued is expected to price above $1.5M per course, reduce COGS% accordingly from the band floor.

6. This is a prior, not an estimate

These values are a starting prior for use when asset-specific manufacturing cost and net price are not yet known. They are not an estimate of, or a substitute for, asset-specific cost accounting. Override them as soon as better data is available. A CMO reviewing a valuation model built on these defaults will immediately ask:

• “Is this steady-state or launch-year? For an autologous CAR-T or first-year AAV product, launch COGS% is far higher than the steady-state default.”
• “Your mAb 18%: does that include royalties? Royalties alone can be 5–15 points.” (Disclosed in band; royalties sit inside 12–25%.)
• “Your bispecific 1.3–2× cost premium: source?” (Derived from yield data, not a published per-gram figure; flagged as lowest-confidence below.)
• “Net or list price in the denominator?” (Reference net price stated per row; for CGT products price is roughly list; for small molecule/mAb, net can be materially below list.)

7. SG&A disclosure

SG&A defaults to 0 in the wizard and is user-supplied. The COGS prior does not imply a value for SG&A. For a complete operating-margin assumption, set SG&A explicitly. Typical mature-product SG&A ranges are 20–35% of net revenue for specialty pharmaceutical products; early-stage assets often model lower SG&A pending commercial infrastructure build-out. SG&A is out of scope for this COGS section and is not benchmarked here.

8. Limitations and confidence flags

• Bispecific cost premium (lowest confidence). The 1.3–2× mAb premium is derived from published yield data (12.5% theoretical heterodimer ceiling; 60–75% observed downstream yield), not from a published per-gram bispecific COGS figure. No clean peer-reviewed $/g bispecific number was located. Treated as derived/assumed.
• Peptide $/g at commercial GMP scale (lower confidence). Anchored to a CDMO/industry ~$100/g figure scaled for peptide length; few peer-reviewed per-gram commercial-scale numbers exist. Partly assumed.
• Fully-burdened central for every modality. The step from drug-substance cost to fully-burdened product COGS uses industry burden ratios cross-checked against blended gross margins, not asset-level cost accounting. Defensible as a prior; not an audited figure.
• Net vs list price. Reference prices are stated as net where data is available. For CGT products, list and net are approximately equal (outcomes-based structures rather than volume rebates). For small molecule and mAb, true net is typically below list (raising COGS% above the central). The bands absorb a reasonable net-vs-list spread.
• Steady-state assumption. All values model steady-state mature manufacturing. Launch-year COGS% for autologous cell therapy or AAV gene therapy is materially higher (60–90%) from under-absorbed suites and low first-batch yields. A valuation for a pre-scale asset in its first 1–2 years of launch should use a much higher COGS% assumption for those years.

9. Versioning

This benchmark is carried by methodology@2026-05-20. It is a methodology disclosure — the version it introduces is a methodology version, not a signed-export dataset stamp (compare: backtest cohort data has its own dataset version). Updates to the per-modality values or the construction methodology require a new methodology version. The prior values and this disclosure are permanent and version-resolvable at /verify.