Amanda Ciarci wrote an article called Content vs Context: If Content Is King, Context Is Queen in Marketing. It talks about how the right content doesn’t work as well if it’s not paired with the right context. We can all think of examples of this from retail advertising. The right time for a delivery pizza ad is during a sports broadcast, and the right place for a software firewall ad is evidently the San Francisco airport.
Context is so much bigger than marketing, though. It’s something that we talked a lot about when we were thinking about customized feature flag delivery. What was the evaluation context? Location, browser, language, version, hardware, preferences… almost anything can be part of the context. Personalization and mass-production have seemed to be on opposite sides of the context question, but that’s changing. Model-T Fords only came in black because “japanning” the metal was cheaper than painting it, and so there was no customization. Now we can build our car’s trim packages from a website and know we will get exactly what we want – our contextual fit.
A big part of what we’re doing with the abundances of compute and storage that we have is making our experiences more tailored and individual. Think about the difference between ICQ (RIP) and Discord. They both serve the same kind of demographic, but Discord lets you have customization at both the personal and community level. Customization and individualizing is a huge part of their revenue model. In ICQ, you could set your status, and that was about it. No custom icons or colors. In order to do that, we need to be able to keep track of each user and which computer they’re logged in from, and what time of day it is, and which community they’re using. It’s much higher-fidelity information, even if it isn’t what we think of as “user data”.
LLMs are going through all the joys and pains of the hype cycle right now, but I bet that we are going to keep RAG – Retrieval-Augmented Generation. It’s going to rifle through all sorts of unstructured data and retrieve things in a hopefully smart way. I don’t want every instance of the word “electricity bill” that appears in my inbox, I am actually looking for the thing I need to pay. Please don’t tell me about the legislative attempts to fix the grid (although that would be a good idea), or anything from my daily newsletters. If you told a human to find the electricity bill, they would use their experience to figure out you probably want the thing that needs paying. That’s context that a human has, and maybe an LLM can learn.
Progressive delivery has a lot of obvious ways that we can introduce customization, but the best kinds of customization are the ones that are contextually aware. If my wifi connection is poor, turn down my camera resolution automatically. If my browser is set to dark mode, open the web app in dark mode. When we think about giving people what they need, when they need it, we can use so much of their context information.
The problem is… do they want us to? Every customization is another way to become more unique and less anonymous. Is that something we should opt people in to automatically? It’s worth thinking about. In the meantime, here’s a fun exercise: ask your marketing operations person how much information they collect automatically. Are you surprised by it?