The many dimensions of 'AI inside' product development
Where to start and what to ponder
Last weekend I had the delightful experience of joining 200 enthusiastic and very cheerful people at Product Camp Melbourne, an in-person unconference for product managers and adjacent folks. Delightful because it gave me the opportunity to do two things I love - get down into the nitty gritty commercial realities of building useful ‘AI inside’ products and teach others how to avoid getting caught out by the often overlooked said nitty gritty bits (beyond the model hype and the PoCs). Data and AI literacy and education is a long standing passion of mine (child of two teachers, what can I say?). And Australia needs far more AI aware product managers!
Over the next couple of newsletters, I’m going to convey in written form some of what we spoke about. Sadly I can’t distil the buzz of the day, the great venue or the standout organisational skills of the volunteer crew who ran the show with such professionalism and joy.
What do you need to know to work in and around AI as a product manager?
More than you might think. (Hence why my keynote ran long 🙂)
Whatever you do, don’t hide behind the ‘I’m not technical’ line. I’m not suggesting you need to learn linear algebra or set up vector databases. But to operate successfully in this space today, you do need to having a working knowledge of what a GPU is, be able to explain supervised learning and be able to estimate how much it would cost to store 100TB of data. There are whole new areas of commerciality involved in AI products before you ever get to the algorithm.
Of course, give it two years and my specific examples above will be out of date. So most of all, you need a life time commitment to learning. It’s super fair to want to job that is a 9-5 with all required learning included in those working hours. But this isn’t that job.
Data may be cheap but labels are coveted
Our world today is awash with data. It’s pretty difficult to exist in modern society and not leave a digital contrail behind that marks out where you’ve been and what you did. However data is definitely not all created of equal value, and that’s why, even in a market very heavily dominated by a few gigantic players, even a sprightly startup might have data that is of significant value, at least at a given point in time.
Labels are the hottest commodity that a smaller company might find itself having sole access to. For instance if you’re a SaaS company with a substantial number of users completing small tasks within process workflows … then you have the potential to have a lot of labels representing the correct outcome of a given small task. Those labels are what is needed to train the vast majority of commercially useful algorithms in use today.
(Note that I do say that you ‘have the potential to’. Often times, unless your company was founded with data in mind, when you dig into this you’ll find that there are one or two key data elements that you’re not collecting yet that are needed to complete the value puzzle.)
And high quality labels can make the difference between ‘meh’ performance and performance good enough to remove toil and make workflows within your product just that little less time consuming.
So go look for label creating factories inside your product workflows! Maybe you’ll uncover value you didn’t even know existed!
And be aware that even if this stuff is new to you, it absolutely isn’t new to the 800 pound gorillas who dominate the online world today. When signing contracts that involve data exchange with any large vendor, read very carefully for clauses that limit your use of the data, including behavioural and process outcome data of mutual customers that might be exchanged, grant the other partner the right to use any label data that you might be contributing, etc.
To be continued
We covered around 14 dimensions at Product Camp so more to come in coming weeks. Next time, I looking to cover
AI regulation is going to be like quicksand,
creepy lines and how they are going to move around
doing all of this is YOUR responsibility even if you use third party vendors