A short guide to what the Biologics COGS Estimator does, who built it, what it can and can't tell you, and how to use it without misleading yourself. If you're here to scrutinize the math first, jump to how the model works.
The Biologics COGS Estimator is a free, browser-based calculator for cost of goods sold across six top-level therapeutic modality groups: antibodies (monoclonal and bispecific), antibody–drug conjugates, recombinant proteins, mRNA therapeutics, oligonucleotide therapeutics (ASO and siRNA), and cell & gene therapy (AAV and autologous CAR-T). It produces order-of-magnitude unit cost estimates ($/g for biologics, mRNA, and oligonucleotides; $/dose for C>) based on user-supplied process and facility parameters, with explicit modeling of manufacturing grade (GMP, Preclinical/Tox, Research) — from research to commercial scale.
The tool is deliberately transparent. All defaults are visible, all assumptions are editable, and the underlying methodology is summarized below with citations to the published sources it draws from. The output is shown alongside a sensitivity matrix exposing the two largest cost levers for each modality, so users can see how robust their estimate is to the most uncertain parameters.
I'm Rob Shaffer, a biotech entrepreneur with a PhD in biochemistry and several years in the biotech industry. I've run a small venture-backed biotech for a few years, and have consulted for large organizations mainly in the tools/CDMO space. My work focuses on technology deep dives, competitive intelligence, and due diligence; all scientific, strategic, and financial.
I set out here to solve a problem I saw other small biotech founders have: when you need quick cost figures that are defensible and you don't want to pay a consultant for an hour of time or bug a busy board member. This may also be helpful for investors or BD folks pressure testing budgets, and for CMO/CDMOs trying to gauge customer expectations or estimate competitive pricing without mystery shopping. It's a side project, made public so others in the same conversations can use it too. Please don't hesitate to reach out with any feedback! See my LinkedIn below.
The tool is built to be useful to four different kinds of reader, each with a different question in mind when they open it:
Sanity-check pitch-deck COGS claims, model competitor unit economics, evaluate CDMO investment theses.
Frame CDMO negotiations, evaluate quotes against an independent baseline, plan capacity decisions at clinical/commercial scale.
Build first cost-of-goods models for fundraising or board materials, identify the process parameters that most affect unit cost.
A defensible, source-anchored reference to share with clients or non-technical stakeholders when internal numbers can't leave the building.
The estimator allocates manufacturing costs across cost categories appropriate to each modality:
Costs are modeled per gram of drug substance, with optional drug product (fill/finish) layered on. The model takes user-supplied bioreactor scale, titer, downstream yield, batch cycle time, success rate, facility type, and resin economics, and computes batches per year, yield per batch, and per-gram costs across upstream (media, feed, single-use consumables), downstream (Protein A, polish resins, buffers and filters), and facility & labor (depreciation, utilities, FTEs, QC/QA). Bispecific antibodies are modeled as a sub-modality of mAb with parameter adjustments reflecting the typical reality: lower expression titer (~45% of equivalent mAb), reduced overall DSP yield (~78%, from additional homodimer/aggregate removal), elevated QC overhead (1.35×), and an extra polish step.
Built on the mAb engine with a separate layer capturing linker–payload cost (typically the dominant line item), conjugation yield, and conjugation overhead. Payload mass per gram of ADC is derived from drug-antibody ratio (DAR) and payload molecular weight.
Built around in vitro transcription on a per-gram basis. Raw-material costs (cap analog, modified nucleotides, template plasmid, enzymes) typically dominate the cost stack. An optional LNP formulation layer adds ionizable lipid, helper lipids, microfluidic mixing, and an encapsulation-loss term, and typically lands at roughly similar magnitude to the DS cost.
Modeled with two sub-modalities reflecting the same underlying solid-phase phosphoramidite chemistry but materially different manufacturing realities.
ASO (Spinraza/Tegsedi-class) is a single-strand synthesis with heavy phosphorothioate backbone and 2′-MOE modifications. Per-step coupling yield compounds across (length − 1) cycles — a 1% shift in per-step yield meaningfully changes economics for a 20-nt oligo because the loss compounds.
siRNA (Alnylam-class) is a duplex of two strands, modeled as two independent solid-phase syntheses (antisense and sense) followed by an annealing step. Each strand's per-step yield compounds across its own (length − 1) cycles independently. Strand lengths are specified separately, supporting both symmetric designs (e.g., 21/21 nt for a classic 19-mer + 2-nt overhang format) and asymmetric designs (e.g., longer antisense with a shorter sense strand for off-target tuning). Material consumption (amidites, support, cycle reagents, purification) is approximately doubled per batch — both strands are purified independently — then duplex formation is limited by the strand in lesser molar quantity, with an annealing yield applied. GalNAc conjugation is an optional add-on for both ASO and siRNA.
mRNA and oligonucleotide defaults are calibrated for clinical to early-commercial scale — the audience the tool is most useful for (small biotechs pricing first programs, BD teams evaluating clinical CDMO quotes, investors sanity-checking early-stage pitch decks). Mature commercial-scale RNA therapeutics cost meaningfully less per unit than these defaults suggest. For reference: published estimates of pandemic-scale mRNA vaccine manufacturing land at roughly $0.50–$3.00 per dose (Light & Lexchin 2021; Kis et al. 2022), and Alnylam-class commercial GalNAc-siRNA is widely estimated at $200–$800 per gram of finished drug substance. To model larger commercial scenarios, increase IVT scale (mRNA) or synthesis scale (oligo), increase batch frequency, and reduce per-batch consumable assumptions to reflect bulk contracting. The default per-gram numbers will drop 5–20× depending on the magnitude of scale-up.
Modeled on a per-dose basis with separate engines for AAV (transient transfection) and autologous CAR-T. AAV economics center on plasmid cost, USP yield (vg/L), and overall DSP yield. Autologous CAR-T uses per-patient batch economics with explicit failure-burden accounting — failed manufacturing runs still incur the material and labor cost they consumed, which materially affects unit economics for autologous processes.
Three grade tiers (GMP, Preclinical/Tox, Research) are modeled with category-specific multipliers rather than a flat factor, because the gap between grades is concentrated in QC/QA (8–30% of GMP cost) and facility allocation (30–55%), not raw materials (40–65%) or labor (30–65%). Modality-specific high-cost inputs — payload-linker for ADCs, plasmid for AAV, lentiviral vector for CAR-T, cap analog and modified nucleotides for mRNA, phosphoramidite monomers for oligonucleotides — have their own multipliers reflecting the steep cost gap between GMP-grade and research-grade material for those inputs:
Net effects land roughly at 0.45–0.55× GMP for Tox grade and 0.15–0.27× for Research grade across modalities, in line with the literature ranges for grade-to-grade cost gaps.
Input fields remain GMP-anchored across grades. When the grade toggle is switched, displayed input values (media cost, payload cost, plasmid cost, etc.) do not change — they represent the GMP reference, and the grade multipliers are applied internally to produce the lower-grade output. This is deliberate: it keeps inputs comparable across grades and preserves the interpretability of the sensitivity matrix.
Default values reflect the published industry literature on biopharmaceutical process economics. The most directly relevant sources:
For modalities where the public literature is thinner — AAV transient transfection economics, autologous CAR-T per-patient costs, commercial-scale RNA therapeutics — defaults are drawn from a broader synthesis of published case studies, industry analyst reports, and CDMO public commentary, calibrated against widely-cited reference points (HEK293 transient yields of 1013–1014 vg/L; CAR-T COGS in the $50k–$100k/dose commercial range; pandemic-scale mRNA at $0.50–$3.00/dose; Alnylam-class commercial GalNAc-siRNA at $200–$800/g).
It isn't a substitute for a real process model. Outputs are order-of-magnitude estimates intended for early-stage planning and comparative discussion. For investment decisions, CDMO contracting, or board-level cost commitments, build a process-specific model calibrated to your actual yields, batch records, and supplier quotes.
It isn't calibrated to any specific facility. Defaults reflect industry-typical ranges from published sources, not the unit economics of any particular CDMO or sponsor. Real-world costs vary substantially with construct, modality, geography, vendor relationships, and contracting structure.
C> defaults are noisier than biologics defaults. AAV USP yield, downstream recovery, and LV vector cost all span ranges of 5–10× across published processes. Treat C> outputs as starting points for conversation, not benchmarks.
RNA defaults are calibrated for clinical to early-commercial scale. Mature commercial-scale therapeutics — pandemic-scale mRNA vaccines or Alnylam-class GalNAc-siRNA — cost meaningfully less per unit than these defaults suggest, typically 5–20× lower depending on scale and contracting structure. The model can be reconfigured for larger scenarios by increasing scale, batch frequency, and reducing raw material assumptions.
The siRNA dual-strand model uses simplifying assumptions. Both strands are modeled at the same synthesis scale with shared per-step coupling yield and the same blended phosphoramidite cost. Real-world programs often use different scales, different chemistries, and different cost profiles for the antisense (guide) and sense (passenger) strands — e.g., heavier 2′-F content on the antisense or different end-stabilization chemistries on the sense. The tool captures the structural cost impact of dual-strand synthesis, asymmetric strand lengths, and the annealing step, but doesn't model per-strand chemistry independently. For a production-relevant siRNA model, calibrate the blended amidite cost and step yield to your actual program.
It doesn't model risk-adjusted economics. The model is deterministic. The sensitivity matrix shows how unit cost responds to two key parameters, but does not produce probability distributions, confidence intervals, or risk-weighted COGS. Probabilistic modeling is planned for the Pro tier.
The free version is feature-complete for single-scenario cost modeling. Beyond that:
You can sign up on the tool page to be notified when the Pro tier launches.
Questions about the methodology, corrections to a default I've gotten wrong, suggestions for what to add, or notes about how you're using the tool — all welcome.
The best way to reach me is LinkedIn:
Connect on LinkedIn →I'm available for select engagements — expert network calls, technology and competitive due diligence, manufacturing cost analysis, and related work in biologics and CDMO economics. Reach out via LinkedIn with context on what you're working on.