The difference between know-how and patent comes down to one trade you make on day one: a patent gives you a 20-year legal monopoly in exchange for telling the world exactly how your invention works. Know-how gives you no monopoly and no expiry date, as long as you keep it secret. File, and the clock starts and the recipe goes public. Stay silent, and you keep the recipe forever but anyone who reinvents it owes you nothing.
That single choice decides more value than most founders realise. Get it wrong and you either hand a working blueprint to your competitors for free, or you sit on an asset a buyer cannot price. At Beyond Elevation we have watched both mistakes wipe seven figures off a valuation in a single diligence call.
Know-how vs patent: the core difference in one table
A patent is a registered right. You file it, an examiner grants it, it publishes 18 months after filing, and it protects you for 20 years from the filing date. Know-how is unregistered. It is the practical knowledge that makes the invention actually work: the parameters, the failure modes, the data-cleaning steps, the tuning that took your team two years to find. It is protected only by secrecy and the contracts around it.
Here is the practical split:
- Disclosure: A patent is fully public. Know-how is private by definition. The moment know-how is published, it stops being know-how.
- Duration: A patent expires in 20 years. Know-how lasts as long as you keep the secret. Coca-Cola's formula has outlived more than a century of patents that would have expired in 1937.
- Enforcement: A patent lets you sue anyone who practises the claim, even an independent inventor who never saw your work. Know-how only protects you against people who took it improperly. Independent reinvention is legal.
- Cost: A patent runs into tens of thousands per jurisdiction across filing, prosecution, and renewals. Know-how costs almost nothing to hold, but the cost of a leak is total and permanent.
When to file a patent and when to stay silent
The rule operators use: file when the invention can be reverse-engineered from the product, stay silent when it cannot. If a competitor can buy your device, take it apart, and see how it works, a trade secret protects nothing because they will figure it out legally. A patent is the only thing that stops them. But if the secret lives in a process or a model that never ships to the customer, disclosure just hands rivals a head start they could not otherwise get.
Three questions settle most cases:
- Is it visible in the shipped product? Yes, lean toward a patent. No, lean toward know-how.
- Can you detect infringement? A patent you cannot police is an expensive certificate. If you could never prove a competitor copied a back-end method, secrecy is stronger.
- How long is the useful life? If the technology is obsolete in five years, a 20-year monopoly is overkill and disclosure is a needless gift. If it compounds for decades, secrecy can outlast any patent.
What AI companies actually choose in 2026
The foundation model companies have already made this call, and they split it. They patent the architecture and the application layer, the parts a user or a teardown could infer, and they keep the model weights and the training data as trade secrets under the Defend Trade Secrets Act, which has no 20-year clock. Following the 2025 to 2026 USPTO eligibility reset, more machine-learning method claims now survive examination, so the patent half of that split is wider than it was two years ago. But the crown jewels, the weights and the curated data, almost never get filed. Publishing them would be commercial suicide.
This is the answer to the most common question we get from AI founders: you do not pick one. You build a portfolio where patents fence off the visible perimeter and know-how guards the core. The valuation premium comes from documenting both so a buyer can see the whole asset. Unpatented know-how routinely accounts for the majority of an AI startup's value, which is exactly why it has to be written down, access-controlled, and contractually locked rather than left in a few engineers' heads. We cover the mechanics of pricing that hidden value in our breakdown of valuing unpatented AI know-how.
The mistake that costs the most
The expensive error is treating this as a one-time legal filing instead of an ongoing financial decision. A patent attorney is paid to file patents, so the default advice is to file. But filing the wrong invention publishes your secret sauce and starts a clock you did not need. The right owner of this decision is whoever owns the company's value, usually the founder working with a finance lead, not the person paid by the document. Companies that turn dormant know-how into licensable, documented assets open a revenue line most never touch, as we show in know-how licensing. And founders navigating the eligibility rules around what is even patentable should read our guide to AI patent eligibility in 2026 before they file anything.
Beyond Elevation helps tech CEOs make this exact call across their whole portfolio: what to file, what to keep silent, and how to document the silent half so it shows up as value in a raise or an exit. We have turned many patents and trade secrets into billions in IP value, and our clients rate us 4.5 on Trustpilot. If you are about to file, talk to us first at beyondelevation.com.
FAQ
What is the difference between know-how and a patent?
A patent is a registered, public right that grants a 20-year monopoly in exchange for full disclosure of how the invention works. Know-how is unregistered private knowledge protected only by secrecy. A patent stops even independent inventors. Know-how only stops people who take it improperly.
Is know-how the same as a trade secret?
Know-how becomes a trade secret when it has commercial value, is not generally known, and is protected by reasonable secrecy measures such as access controls and confidentiality agreements. All trade secrets are know-how, but loose know-how with no protection in place does not yet qualify as a legally enforceable trade secret.
Can you license know-how without a patent?
Yes. Know-how is licensed every day through agreements that bind the licensee to confidentiality and pay royalties for access to the method or process. Because there is no 20-year expiry, a well-protected know-how license can generate revenue indefinitely.
Why do AI companies keep model weights as know-how instead of patenting them?
Patenting requires publishing how the asset works. Model weights and training data are the core value of an AI company and cannot be inferred from the shipped product, so publishing them would hand competitors the crown jewels for nothing. Secrecy under the Defend Trade Secrets Act protects them with no expiry date.
Which is more valuable to an acquirer, patents or know-how?
Both, when documented. Patents are easy to count and price, so buyers credit them quickly. Know-how often holds the majority of the value but only if it is written down, access-controlled, and contractually secured. Undocumented know-how that lives in employees' heads is treated as a risk, not an asset.