For people like me who peddle ideas, New Years is our key season. It’s the time for us to declare what’s important, what to do about those important things, and why people should use our ideas to do those things. I’m not that good at hot takes, and I don’t normally do these kinds of posts. But this year is different.
AI has created big, historic opportunities, and also a kind of excited murkiness about how to pursue them. If you’re paying attention and ready to do the work, this is a great time for you as a PM. Smart people have invented AI for real, but the market has only operationalized a tiny part of its economic value. How do you find your next big product win with AI?
While execution is never easy, these three questions will put you on the scent:
- What underlying jobs (problems, habits, desires) is your user ‘hiring’ you to handle for them right now?
- What has that user been doing with AI so far?
- Even if you have some AI in play today, how might you do those jobs 5x to 10x better with an even better AI-powered solution?
The old chestnut of “Is your product an aspirin or a vitamin?” works pretty well as a way to answer these questions. The basic idea is that an ‘aspirin’ product solves an acute need right away, while a ‘vitamin’ product requires more effort but offers bigger benefits over time. So, you get a headache, you take an aspirin, and you get relief. But why are you getting those headaches? Do you need a better diet, exercise, or, yes, maybe- vitamins?
Let’s say the job you want done isn’t managing your health, but instead maximizing the value of your company’s data. An ‘aspirin’ solution is online backup so you don’t lose data or have downtime because a computer goes bad. A ‘vitamin’ solution is getting all your user data in a shared repository where an AI can ingest it and answer questions like: “Where should our customer success teams focus next month?”. Or, “If we make a [certain change in strategic focus], which new features should we prioritize next month?”.
We’re right in the middle of this transition from aspirin to vitamin AI solutions at one of the startups I help with, Jedburgh Technologies. Jedburgh develops gamified learning solutions for professional language learners, mostly in the US Department of Defense. One of their products is a VR game where you go around and practice listening and speaking with non-player characters. Before chatGPT, Jedburgh had to maintain complicated, error-prone dialog trees (basically- if the player says something like [x], say [y]). With chatGPT, those trees are history and building new VR games is massively easier and better. That’s the aspirin solution, and it was a massive relief.
At their core, however, the intensive language-learning programs these learners are in are about two things: efficiency and predictability. In a year-long program, how much progress does the average learner make? And, post-program, as they go into maintenance mode with their new language skills, how visible is it to command the number and fluency of foreign language speakers they have on staff? While it’s a much bigger endeavor, we’re now working with our customers and partners on AI-powered solutions to put the exact right practice in front of each learner at the exact right moment, to optimize pace for individual learners, and to predict fluency based on that pace. These are vitamin-oriented solutions that require more change and investment, but they have a bigger payoff over time.
AI is a kind of magic tonic that can be made into aspirin-type or vitamin type solutions and, not surprisingly, when it stormed onto the scene in 2023 we mostly used it as aspirin: having it do our rote, repetitive work for us in essentially the same way we’d been doing it, but automatically. So, for example, an individual can get themselves a chatGPT account and without changing the way they or their company do things immediately have it:
- Answer questions better than Google.
- Synthesize meetings notes from a transcript
- Write code that’s using well understood patterns to solve well defined problems
- Inspect a codebase and look for obsolete coding or user interface patterns
Gen AI is basically a calculator for words— from a given context, it can tell you what other people are saying. If you think about the work we do on a scale from primary science (fundamentally new discoveries) to purely humanistic (connecting with human needs in a new and novel way), then AI is great at the stuff in the middle: applying well understood recipes to well understood problems.
In the short to medium term, AI is hollowing out ‘technical’ work, work where you’re applying known techniques to known problems:
But where’s your next big win as a PM? Certainly, there’s still plenty of upside with aspirin-type solutions: just look at the persistent clumsiness of voice UI’s like Alexa and Siri. However, given how much AI has already changed the way we work and play, there’s little disagreement that a lot of the big wins are still to come.
Looking beyond aspirin-type solutions to vitamin-type solutions is a great way for PM’s to go beyond the obvious while staying anchored in user problems that really exist and ways that users really behave. For example, chatGPT and Copilot are great at helping us write and debug code (aspirin), but what about code that writes and fixes itself (vitamin)? AI’s great at helping us sort through gigs of logs to identify possible security issues on our systems (aspirin), but what about a security application that intervenes directly to stop those threats? AI’s great at making chatbots a little more useful, but what about a bot that knows your offer and your customer better than you know yourselves?
There are two things you, as a PM, can give AI to help it outperform what it’s able to do today:
- data
and - context
Custody of proprietary data has been an economic goldmine for the last two decades: as easy as it is to try a new search engine, no one knows more about what you’re after than Google and no one can compete with them on search. If you have customers or employees, you have proprietary data of your own right now. While there are multiple tools that allow you to layer your models and data to keep your data proprietary, private data is now an available feature on both the openAI (chatGPT) and Microsoft products.
Adobe Stock, a marketplace for photos and illustrations, is a recent example of a data-driven AI win. One the one side of the market, they allow their sellers (of images) to opt-in to train Adobe’s generative AI. That AI then generates and posts additional photos and images related to the observed needs of buyers. When a seller’s image is used to create an AI-generated image, Adobe is able to attribute that and compensate the relevant sellers as that AI-generated image sells.
Along with data, context is the other fundamental driver for AI performance. Context for an AI is a lot like context for a human: you can’t ask it to fix your Javascript code if you don’t both show it the code and tell it what you want it to do differently. This extends to any number of situations where you might want an AI to intervene: PM’s are learning how to use the context their application has about a user to improve the prompts an AI is getting. One great example of how this looks and works in practice is the markup language Khan Academy (online education) created to power their ‘Khanmigo’ tutor. They essentially use context about their user, their course of study, and the material at hand to make secondary refinements to what their student’s ask the AI tutor, which then uses chatGPT to generate responses.
In summary, don’t neglect the easy wins with AI, but 2024 is the year PM’s will start to go from ‘aspirin’ to ‘vitamin’ type solutions. They’ll do this by taking the understanding they have about their particular users and pairing it with relevant data and context. Over the next few weeks, I’ll be extending this post to a series on how PM’s are going from aspirin to vitamin across these domains: design, coding, sales & marketing, DevOps, internal IT, and AI itself.
Happy New Year!
