Design preview. Adapted from the FoodEdge.dev project for this Ghost demonstration. Editorial review is pending. The cover is an AI-generated conceptual illustration.
Almost every precision-fermentation deck has a slide with a single cost-of-goods number on it. That single number is usually where the pitch quietly falls apart. An investor who has seen a few bioprocess decks knows that a cost of goods sold is not a fact you look up, it is a model you build, and the model is only as good as the assumptions underneath it. A hero number with no assumptions reads as marketing. A defensible range with the drivers shown reads as engineering.
This guide shows you how to build the second kind. You will not need a full techno-economic model to start. You need to understand what actually moves the number, why two drivers dominate the rest, and where honest founders hit a wall at scale.
Start from the molecule, not the spreadsheet
Work in the direction the biology runs. A microbe eats a feedstock, grows, and secretes or accumulates your target molecule in a broth. You then pull that molecule out and purify it. Every cost sits on one of those two phases: upstream (the fermentation) and downstream (the recovery and purification). A first-principles COGS estimate is just an honest accounting of both, expressed per kilogram of finished product.
The reason to start here rather than in Excel is that it forces you to name your performance metrics before you cost anything. Get those wrong and every downstream number is fiction.
The three numbers you must define first
Three fermentation metrics set the whole model. Founders routinely blur them, and the blur is where credibility leaks out.
Titre is how much product you get per litre of broth at harvest, in grams per litre. It is the primary measure of upstream efficiency. Where your strain actually sits, measured, is the single most load-bearing input in the model.
Yield is ambiguous, and that ambiguity trips people up. There are two yields, and you must state both. Metabolic yield is grams of product per gram of substrate consumed, which is bounded by biochemistry and sets your feedstock floor. Recovery yield is the fraction of product that survives downstream processing to become sellable material. A process with a strong titre and a leaky recovery yield can still have a terrible cost of goods.
Productivity, or volumetric rate, is grams per litre per hour. It decides how many batches a tank runs per year, which decides how much product each unit of capital produces. Low productivity means your expensive steel sits idle, and idle steel is one of the most expensive things in a bioprocess plant.
Define all three, with real measured values from your own runs, before you cost a single input. The Good Food Institute's science of fermentation is a solid, vendor-neutral reference for how titre and yield are conventionally defined.
The cost drivers, ranked
A complete estimate has to touch every line below. But they are not equal, and pretending they are is a rookie signal.
- Feedstock and media. Your carbon source (usually a sugar), nitrogen source, salts, and any speciality media components. Governed by metabolic yield: you cannot spend less on sugar than the biochemistry of putting carbon into your molecule demands. This is a hard floor, not a negotiable line.
- Downstream processing (DSP). Cell separation, lysis if the product is intracellular, filtration, chromatography, concentration, and drying. Consumables here, chromatography resins especially, are brutal.
- Capex and depreciation. The fermenters, the DSP train, the facility, spread across the kilograms they produce over their life. Utilisation is everything: the same plant depreciated over a full order book versus a half-empty one gives wildly different per-kilogram numbers.
- Utilities. Steam for sterilisation and clean-in-place, cooling water to remove fermentation heat, and electricity for agitation and aeration. Small per litre, real at scale.
- Labour. Meaningful at pilot, where a batch is hand-held, and a shrinking share of the total as you automate and scale.
Why titre and downstream processing dominate
Two of those drivers move the number far more than the rest, and they are linked.
Titre dominates because it is a denominator. Almost every fixed and semi-fixed cost, the tank time, the utilities, the labour, the depreciation, is spread across the grams of product in a batch. Raise the titre and you spread the same costs over more product. The effect is strongly non-linear. The first improvements, moving off a low base, cut cost of goods hard. Later improvements, from an already high titre, give diminishing returns. This is why early-stage strain and process work pays back so heavily, and why a founder claiming cost parity from a low measured titre is not credible.
Downstream processing dominates for the mirror-image reason. At low titre your product is a dilute needle in a large haystack of broth, and the cost of finding and purifying it is largely fixed per litre processed, not per gram recovered. The cost share varies by molecule, required purity and process design. In other words, the two dominant drivers are the same lever seen from both ends: lift titre and you both spread your fixed costs and shrink your DSP burden per kilogram.
The practical consequence: test titre and recovery yield explicitly, then check whether feedstock, cycle time, utilisation or another constraint dominates your particular process.
The wall many founders hit at scale
Bench economics and plant economics are different animals, and the gap between them is where a lot of precision-fermentation stories stall.
At scale you run into physics that a shake flask hides. Oxygen transfer gets harder as tanks get bigger, so an aerobic strain that thrived in a 10-litre vessel can become oxygen-limited in a 100,000-litre one, dragging titre and productivity down exactly where you needed them up. Fermentation generates heat, and removing it from a large tank costs real cooling capacity. Mixing is slower, gradients appear, and contamination in a batch that size is not a bad afternoon, it is a costly written-off run.
Then there is the media floor. As you optimise everything else, feedstock becomes a larger share of what is left, and metabolic yield caps how low that can go. You cannot engineer your way below the cost of the carbon that has to end up in the molecule. Many founders model their way to cost parity by quietly assuming a titre, a recovery yield, and a plant utilisation that no one has yet demonstrated at scale. That stacked optimism is the wall. Name each assumption and you will see it coming.
How to present a credible range to investors
Do not present a number. Present a range with its assumptions exposed. Three moves make it land.
Separate today from tomorrow. Give your real, measured pilot cost of goods now, and your projected cost at nameplate scale, clearly labelled as two different things. Conflating them is the fastest way to lose a technical investor.
Show the sensitivity. State your base case, an optimistic case, and a conservative case, and tie each to specific values of titre, recovery yield, and plant utilisation. When the investor sees that your bridge to cost parity runs through a titre you have a credible plan to reach, you are having an engineering conversation, not a hope conversation.
Anchor every number to an assumption. Feedstock cost to metabolic yield, DSP cost to titre and recovery yield, depreciation to utilisation. A reviewer should be able to change one assumption and watch the number move. That traceability is what "defensible" actually means.
A model worth reviewing
Start with measured pilot data. State the product specification, the mass basis and the scale of every input. Put uncertain assumptions next to the quantities they affect.
- Record the measured base case.
- Change one uncertain input at a time.
- Build a conservative case with realistic recovery losses and utilisation.
- Show what evidence would justify moving to the projected case.
A useful cost model explains what must become true before a target price can become credible.
Keep process evidence separate from the commercial forecast. A target at full utilisation is a scenario, not a result already achieved.
