5 questions to help productize AI

I met two AI engineers building this week — one at ZERO Founders Network’s AI & Climate event and the other at an after-party hosted by Oxbridge AI Challenge — and they got me thinking.

Both AI engineers were chewing over the same problem: how to productize AI. One had scraped open-source data and was using AI to provide insights to businesses as a lead generator using that data; the other had sold himself to a client and was using AI to provide insights or automate processes in the business. The first one was pitching me for investment from Oxford Seed Fund, the other was just thinking out loud on how to scale. In both cases, they were offering a service — not a product. I went looking for lessons from the software boom to see what transfers over to this moment of AI.

The founders were chewing over this transition — from service to product — and this is probably a common journey for AI tech founders, moving from providing bespoke services to offering scalable, repeatable products and hitting an air-gap along the way. How do they move from a service-oriented business model (customized, high-touch solutions) to a product-oriented model (scalable, standardized offerings) ?

What’s the problem? Well, it’s not really a ‘problem’ — there’s nothing wrong with a service-oriented business model. But rather, from a ‘venture backable’ view capital is looking for scale and the founders are looking for growth, so the airgap between the service model and the product model is a ‘problem’ from that perspective.

Founders have built their tool kit — they can build (or even just access) models and improve outcomes in a data rich setting. Early-stage AI founders can then start by providing custom solutions to specific customer problems. These might involve the founder manually building models, analyzing data, and tailoring insights based on the client’s needs. While this builds early revenue, it probably isn’t scalable.

This is the ‘problem’ with the service-based model — it lacks the ability to scale because each project is highly customized. The founder and team have to tweak algorithms, datasets, and outputs to fit individual clients’ needs, which means the solution doesn’t have the same reusable, plug-and-play quality as a productized SaaS offering. AI founders might be focused on solving specific client questions, delivering the results in reports or via API calls on a one-off basis. This is labor-intensive and may not provide the clients with ongoing or autonomous use of the AI solution.

Even from the client standpoint, they’re dependent on the founder for outcomes. There’s no “product” for them to interact with, just results delivered by the team. Without a user-friendly interface or platform, scaling across multiple customers or industries becomes difficult.

(1) What’s the core value proposition?

Ask yourself what is the part of the algorithm or AI service that’s repeatable and valuable across many customers? And ask yourself what specific pain points or workflows the AI solves effectively and consistently that keep popping up from the client? As the founder you have to define the core capabilities that can be automated and standardized & overlay that with the pain from the client. Draw a Venn diagram of the capabilities you draw from the models you’re using (automations, insights, data types) vs. the frequent types of questions the client is having you answer. In the middle is the core value prop. For a nice case study on this, check out the story of Slack — how they moved from being a video game company to unlocking comms. for teams.

(2) Will modularity open up new customer types?

Ask yourself whether your use case of the AI model be modularized to serve a wider array of clients without customizations? Does the data structure of your customer look similar to the data structure of a customer in another industry? If yes, then you might be able to build generalized modules that can be adapted by different customers through parameter adjustments rather than custom code. One of the founders I spoke to zoomed out and realized that the lead gen tool he’d build could be modularized (from the customers’ perspective) and be useful outside of the industry he started with, as long as the structure of the data was the same (which in many cases it was). Twilio’s origin story is a great ref for this — modularizing their capabilities into APIs that could be easily adapted by businesses to fit use cases (customer service, two-factor authentication, marketing communications, etc.).

(3) Can you build a self-service platform?

If the insights you’re generating for the client are pretty similar each time, can you build a platform for them to access the front-end of those insights without having to dig under the hood? Depending on the tech-savviness of the client (assuming on the lower end), then this could be a GUI/dashboard for whoever the role is in the organization that’s to take action from your insights — or if they already have an internal system (e.g., a CRM, an ERP) then consider plugging into that with an API and offer your product as a ‘plug in’. (Likely that non-techy customers won’t understand the ‘API’ bit, but you can position as ‘we’re a plug in to the CRM you use now’ etc.). If a front-end isn’t something you can pull together aesthetically, go out and find someone who can — or maybe just think about what is the lowest-rest version of this that a customer would use… if the MVP is too pretty, it’s too late.

(4) How to charge for the product?

Charging for a service — either in work package blocks, on reward/commission, or in a flat fee — is straightforward but doesn’t necessarily bear much similarity to how to charge for a product. Is a subscription model suitable? Charge a monthly or annual fee for platform access, with different pricing tiers based on feature sets or data usage limits. Or a pay-per-use? For AI models that are resource-intensive, charge based on the number of API calls or the volume of data processed. Does the number of users matter? If yes, then charge per user or role that accesses the platform, especially if different teams within a company will interact with it.

(5) What does the customer think?

Don’t step away from the customer — step closer. Ask them how to improve the product, and then build mechanisms into the SaaS platform that allow you to gather feedback on model performance, user behavior, and pain points. Use this data to iterate on and improve what you’ve built.

To me these two founders are facing something that a lot of tech founders face in their entrepreneurship journey: transitioning from service-based engagements to a productized SaaS model — standardizing their solution, creating a self-service platform, and automating processes while maintaining enough flexibility for customers to feel they are still getting a tailored experience. Hopefully these questions help AI engineers think through what a product offering could be from the services they’re currently selling to clients.