The real innovation lab cost isn't cost-per-venture. It's cost-per-validated-venture, and the failure rate moves the number 10x. The line-item teardown.
The headline cost is the wrong number. The denominator is where the real money hides.
Most CFOs can tell you what the innovation lab costs per year. Far fewer can tell you what it costs to produce one venture that survives contact with the market. Those two numbers diverge hard, and the distance between them is the entire argument for the studio model. So we will build the number from the bottom up, showing every assumption, nothing rounded for effect.
One thing up front, because it changes how to read what follows. This is an illustrative comparison with stated assumptions, not a published benchmark. The cost and rate figures below are the ones we use in our own model, sized to be realistic against what is publicly discussed about corporate innovation costs. Treat the structure of the argument as the point, and the specific numbers as inputs you should pressure-test against your own.
Take an internal venture attempt at a large FI. The all-in figure, salaries, technology, overhead, and the opportunity cost of senior people not doing their day jobs, runs somewhere in the range of $2 million to $5 million per venture on the assumptions we use. That is the cost of trying once. Build the team, stand up the tooling, run the project, take it to a pilot.
Now the studio attempt. A studio runs a venture attempt at roughly $250,000 to $500,000, including its share of shared services. The difference is not magic. It is structure. Product, engineering, growth, legal, and compliance are amortized across a portfolio rather than rebuilt per venture, which is where most of the per-venture overhead comes out. One bench, many ventures, instead of one team, one venture.
So far the gap is real but unremarkable: a few hundred thousand against a few million. The number that decides the question is the next one.
Most internal cost analyses quietly cheat at this step. They report cost-per-attempt and stop. But not every attempt produces a venture worth having. To compare two models honestly, you divide the cost of an attempt by the share of attempts that clear validation. By validation we mean something specific: the venture reached real, paying market demand or a fundable milestone, not just a finished build. That gives you cost-per-validated-venture, which puts a lab and a studio on the same footing for the first time.
Run it both ways, on the rates we assume in our model.
The internal lab: $2 million to $5 million per attempt, and assume roughly one in seven to one in ten attempts clears that validation bar. Divide. A validated venture costs somewhere between $13 million and $50 million.
The studio: $250,000 to $500,000 per attempt, and assume close to half of attempts clear the bar, because the model is built to find the signal early and stop the rest. Divide. A validated venture costs roughly $420,000 to $1.1 million.
Set those two ranges side by side. $13 to $50 million against $420,000 to $1.1 million. That is not a 2x difference. On a per-validated-venture basis it is roughly 10x or more, and the reason is not that the studio's people are cheaper. It is that the studio's failure rate is lower and its failures are cheaper. We have written about why a too-wide, low-discipline funnel destroys economics rather than improving them.
A 90% failure rate at $3M an attempt is a budget. A 50% failure rate at $400K an attempt is a strategy.
The teardown above understates the internal number, because it assumes the lab stops failing ventures at the validation gate. They often do not. Internal efforts tend to run well past the point the market signal justified, because stopping a project costs someone politically and nobody volunteers for that. The venture lives on. The spend keeps accruing.
Quantify it on our model. A venture stopped at week six costs a studio under $50,000 (AOS operating model). The same idea, carried inside a lab until it finally collapses, can run back up toward that $2 million to $5 million attempt cost before anyone calls it. That delta of several million dollars measures something specific: not the cost of building the wrong thing, but the cost of being unable to stop building it. The lab can build perfectly well. What it struggles to do is stop, and an inability to stop is expensive in a way that never shows up as a line item, because the spend gets filed under "ongoing project," not "loss." We covered the mechanism in how, and how fast, a studio stops a venture.
Make it concrete with a hypothetical. An FI runs two ventures through a studio in Year 1 (illustrative, AOS operating model).
Venture A finds a real signal. Fifteen customer interviews, a thin product, five pilot customers, a clean advance at the gate, then a small seed round and early revenue by month 12. It works.
Venture B does not. Ten interviews, two smoke tests, and by week eight the signal is plainly absent and the regulatory path is murky. It is stopped. Cost: roughly $35,000 (AOS operating model). The same idea, run as an internal project, plausibly burns into the millions over 12 to 18 months before the org admits it.
Net Year 1 result: one funded venture, plus several million dollars in avoided waste, against a studio engagement that costs a fraction of a single internal venture attempt. The stopped venture is not the failure in that story. The stopped venture is where the savings came from.
This math is not an argument against ever building in-house. There is a real category where an internal lab is the better model, and an honest comparison has to say so. Deep-tech and long-horizon R&D that is tightly coupled to your core IP belongs inside. If the work is a multi-year bet on proprietary technology that has to live next to your most sensitive systems and feed directly back into the core, the amortized, fast-cadence studio model is the wrong tool. You want that knowledge held in-house, the timeline measured in years rather than gates, and the IP under your roof from day one. The cost-per-validated-venture lens is built for new, customer-facing ventures on a months-long clock. For core R&D on a different clock entirely, the inputs and the answer both change.
Stop quoting cost-per-venture. Quote cost-per-validated-venture, show the division on the slide, and the conversation ends quickly. On the assumptions laid out above, the internal column lands at $13 to $50 million and the studio column lands near $420,000 to $1.1 million. Swap in your own cost and clearance rates and the gap narrows or widens, but the shape holds. The money is already being spent on innovation either way. The only question the math is really asking is what that money buys, and how much of it goes to ventures that were never going to work. Take that one slide into the next budget review and watch which column the room defends.
Read next: Why bank innovation labs stall: structure, not talent
Nothing here is an offer to sell a security or investment advice.
Field notes on venture building, AI, and capital. No spam, unsubscribe anytime.
By subscribing you agree to receive AOS Insights e-mails. We use your address only for this newsletter - see our Privacy Policy.
