Draft for Tamir's review. Not published.
Customer value comes before the AI label
Ask target customers which task the feature improves, and separate curiosity from willingness to pay. Compare it with a simpler alternative without AI. Count the operating cost and the failure handling it adds.
If an investor says your operational data could be a product, test three things first. Is the data usable? Would someone pay for what it predicts? Do you have the right to use it that way? Then claim only what the evidence supports, and present the rest as a plan.
What you keep if a provider ships the same feature
Write down, in plain words, what the product does beyond a single call to a provider. List your own data, integrations and domain evaluation sets, and how long each took to build. Ask your best customers why they chose you and what would make them switch.
The same question decides releases and partnerships. Before releasing a model openly, know what customers would still pay for afterward, and have counsel check your rights. Before selling another company's AI under your brand, check which changes the partner can make without your approval.
Depending on your seat
If you're on the board and management proposes pivoting to AI with the remaining funding, ask for customer evidence with names. Ask what the team has built toward it and what happens to current customers. Approve a bounded first phase with a measure, a budget and a decision date.
If you're the CEO, decide which of your holdings to invest in so the answer is stronger by the next round. If you sell AI-assisted work at fixed fees, first compare delivered work with all its review and correction effort.
What to check before you decide
- Ask target customers which task the AI feature improves and whether they would pay for it.
- Compare the feature with a simpler alternative that doesn't use AI.
- List what the product does beyond a single call to a model provider.
- List your own data, integrations and domain evaluation sets, and how long each took to build.
- Have counsel check your rights to the data and models you would use or release.
- Check which changes a partner can make to a white-labeled AI product without your approval.
- Approve any pivot in a bounded first phase with a measure, a budget and a decision date.
Questions people ask
Management wants to pivot our product to AI with the remaining funding, what should the board require before approving?
Require evidence that the problem exists for paying customers, that the team can build the product with what it has, and that the current business can be wound down or kept without collapse. Approve a bounded first phase with a measure and a decision point rather than the whole pivot. It depends on how much runway remains and on what the team has already built toward the new product.
Investors ask what our AI startup keeps if the model providers copy our feature, how do I assess our real advantage?
List what you hold that a provider does not: proprietary data, workflow integration into customers' systems, evaluation and quality on a domain, distribution and trust. Then test each one against how hard it is to copy in a year. It depends on how much of your product's value sits in the model call versus around it.
We have years of operational data and an investor says it could be an AI product, how do I check that before raising on it?
Test three things: whether the data is usable, whether someone would pay for what it predicts, and whether you have the right to use it that way. A few weeks on a sample answers the first and the third; a few customer conversations answer the second. It depends on the data's quality and ownership terms and on whether a buyer exists outside your current business.
Should our startup add AI before customers will pay for the feature?
Start with a customer task and a buying or retention decision. Building is justified only if evidence suggests the capability matters enough to outweigh its operating and support burden.
Should our agency promise fixed fees for work delivered with AI?
Commit around bounded deliverables and evidenced variability rather than tool enthusiasm. The commercial model depends on acceptance, exception work, and who carries changes in scope or input quality.
Should we white-label another company's AI product as our own?
Align customer promises with the control and evidence the partnership provides. Signing depends on support boundaries, change rights, failure handling, and terms reviewed with commercial and legal owners.
Should we release our AI model openly or retain commercial control?
Assess the release against the business model and rights you actually hold. It depends on what customers buy, defensible assets beyond the model, license conditions, support costs, and the reversibility of disclosure.
How I can help with this decision
- Ask or talk (Free)
- I give my view on which of your holdings investors usually credit, and which test of customer value or data usually fails first.
- Review (Pay if it was worth it)
- I write an independent assessment of the product's advantages, their durability and the evidence for the AI plan. I recommend where to invest, what to claim, and whether to proceed, bound or decline.
- Retain (When it makes sense)
- I stay close through the round or the first phase to review the technical claims against what the product actually does.