Weeknotes 506

I did:

Purist to pragmatist

This week involved deliberately sliding back and forth along the purist-pragmatist scale, and being as intentional as I could about what that means and doesn’t mean. And accepting that misalignment is a feature, not a bug. I also did:

• Did lots of planning for a roll-out over the next three months.
• Got approval to build an AI feature.
• Ran a session with technology leaders to consider how they manage different types of work, funded in different ways, for different teams and with different reporting requirements.
• Watched the team deal with an issue really well, and thought about how the potential for issues grows exponentially with every change in complex, interconnected tech.
• Worked on paper to explore some ideas, which I haven’t done in ages.
• Started planning a workshop to look at conversion rate optimisation opportunities.
• Ran a product vision exercise (not quite a workshop yet). It asks what needs to be true for the vision to become a reality to identify the conditions required for a product to succeed.

I read/watched:

Reality drift

I read reality drift from Boring Magic. It reminded me of a conversation I had a couple of weeks ago about trust and expectations of AI systems.

We’ve spent decades working with technology that is about providing a single right answer. Your HR system tells you how many days leave you have left, your banking app shows you how much money you have, your washing machine tells you how long the spin cycle will take. If they tell you the wrong answer, we know something is broken and needs to be fixed. That’s the nature of the deterministic technology we’ve all come to know and trust.

Now, with AI, we’re having to learn to deal with technology that doesn’t provide a single right answer, works in nondeterminisitic ways, and where getting something wrong isn’t a failure. With AI, the answers become take some time off, don’t buy those new shoes, and the sun is shining so hang your washing outside. Those answers aren’t the single right answer and they might be wrong or they might be what you’re looking for.

Problems occur when we expect AI technologies to work like other, more traditional, technologies. There are plenty of times where AI is the right tool to use, but my go-to reminder is ‘only use AI if there isn’t a single right answer’.

The purpose of data is knowledge

And the purpose of knowledge is to allow you to predict the outcomes of your business actions. Common Cog’s series on Becoming Data Driven in Business is fantastic. I started at the end with Becoming Data Driven, From First Principles (you’ll get the joke if you read the intro to the series), which talks about spotting the difference between routine variation and when the data is telling you something significant has changed, either because you did something to change it or because you didn’t. It ends with, “Understanding variation leads to the process control worldview, which leads to an org-wide pursuit of knowledge.”

Rory Sutherland’s 2026 Predictions

Rory talks about businesses being in cost-reduction mode, which might not be a very revolutionary prediction given everything that’s going on in the world. He says modern business is an efficiency competition and that AI will be used by business leaders attempting to win that competition. What I think is interesting about this is that it shows how things without a clear definition, e.g., AI, mean whatever we want them to mean given our worldview.

AI interfaces

Interfaces-on-the-fly is one of those interesting ideas that’s been around a little while. Maybe it’s the path to making personal AI devices ubiquitous MacGuffins in our lives. Imagine a blank smart phone screen that becomes a chat window where you ask your AI assistant to compare running shoes and the interface evolves into a comparison site -looking interface to help you choose and then into a shopping experience to buy them, and later a map interface as you track the delivery.

I thought:

Misunderstanding evidence

When we talk about being evidence-led, what do we mean? I think, we shouldn’t mean undisputed proof that something is true. We should consider evidence as how a phenomenon appears in the real word to allow us to test theories against it. It’s more of an academic definition than a criminal investigation definition and it allows for more useful thinking. If we take evidence to mean incontrovertible proof, and for all kind of reasons, we aren’t able to get that evidence, then we’re stuck. There’s no where else to go. But when we treat evidence as not an end in itself but as part of the process of theory-evidence-analysis, then we can use the observable evidence (accepting its incomplete) to test our theories.

The product operating model for non-product companies

One view of the firm says there are three big value-driving functions: customer relationships, new product development, supply chain management. For the university I work at, new product development is for the information goods (courses) that our customers purchase, not the technology products they use to interact with the course material and tutors. Technology products are in the supply chain management function because they are how the organisation distributes it’s goods. In our case, we use a direct distribution strategy which means the only place you can study our courses is with us, using our technology products. That’s why they are essential to our business model, and make for an interesting product operating model.

Marty Cagan recently posted about the difference between internal and commercial products, which may (or may not, I’ll let you decide) fit for this kind of (mostly mythical) modern product organisation where customer relationships, new product development and supply chain management are all wrapped up in the same commercial product and internal products which support the commercial ones. Our products aren’t like that. We don’t have commercial products in the sense that someone pays to use our products (because what we actually sell are information goods, as I explained above), so by Cagan’s definition all our products are internal products, except they aren’t because they are used by our users. Our products are something different.

So when we talk about a product operating model in the context of a university, we have to be quite strict and critical with our thinking to make sure we stay within the supply chain management function, which sounds like it should be about lorries and logistics but for a distance learning organisation that distributes its products via the internet, is actually about data and technology. Our product operating model isn’t like that of a product-led commercial company, but neither is it like that of a non-technology company with a traditional IT function. Our product operating model is something else.

• Category: Weeknotes