We see and hear a lot of the buzz of AI coming from other platforms. Some platforms are deep in chatbot-land, with the belief that AI is just ChatGPT bolted onto their product. To us, this lacks direction, and results in a lot of "AI Slop" or just chatting your way through things that your mouse would easily manage - basically missing tangible benefits.
Instead, we believe in "AI with purpose" - things that help you "punch above your weight-class" by doing more, moving faster, being smarter, catching problems before they're problems - the types of things that we as genuine practitioners of CRO Experiments have felt the pain of in our own programmes. And so the very varied AI features that we bring to the table aren't just "here's a chatbot, enjoy". They're looking at how we can bring the world's best AI tech and put it in your hands.
And unlike other platforms, we're not shackled to any vendor or supplier. Where it's the right thing to do, we use frontier models from OpenAI, Google Gemini, Anthropic Claude, etc. but also run many open-source models ourselves. Each in the right place for the right reason. And, very importantly, we are strong believers in Sovereign AI - doing our bit to subvert what we think will be the next car-crash moment in tech. So wherever we can, your data is run on WTO-owned hardware and not just shipped to someone else's API where it's used to train their models. It runs on our hardware, and so can be safely discarded once run to generate whatever we need. In this series of blog posts, we talk about the various applications of AI we're adding to the Optimize product, and most importantly why.
AI with Purpose - how to ideate.
Here's a problem, a conflict really, that I've seen over several years for teams of all sizes. Good wins come from personalisation - not treating everyone the same, and instead carving up slices of people with particular needs and building an experience that actually solves the problem. The conflict - while the industry now has a lot of people who can run simple experimentation programmes or feature-flag their product rollouts, good personalisation frameworks and research are both difficult and take experience. So how can AI help us? With Synthetic User Testing.
THE PRINCIPLE:
We critique a website. We come up with memorable audiences of users, and full personality trait profiles - brand loyalty, technical knowledge, time constraints, price sensitivity, etc. Once we know who they are, we can go and study a particular web page (or several), looking for strengths and weaknesses. The net result is a very structured system with summaries, matrixes, full details of how to actually build the profiles, and then rich studies of each page provided. Put together, this forms a kind of test ideation engine. But not just a "hey, we should add a sticky add to bag and change the button colour" - but addressing real problems for that persona. And the next, and the next.
Beyond AI Slop
So how do we turn this into something other than AI slop? First - we have a solid framework. Deep, detailed, meaningfuf. If we're making statements or decisions, we're explaining why. And we're giving people the ability to make adjustments if they don't agree. We then go and scan particular pages with this persona in mind, and build a 360 map. Good, bad, ugly. Where the opportunities are. You go and study your competition, especially for people where brand loyalty isn't so high, and figure out how they're doing it. Where can you win, and where are they beating you?
GOING FROM LONG GENERATIONS TO A BREATHING SYSTEM
We've then been thinking about how you can feed the machine. We have customers who go and run very rich, very detailed studies where people are going and talking to real customers and learning about their experience. They're going instore and looking at what's interesting there vs. how online works. There's a Customer insights team running surveys and post-purchase feedback loops. All of that is really awesome info. And if it stopped at that point, that's a massive shame and waste. But if you feed it back into the machine - you can re-evaluate everything. Extract your own personas, bolster ours with key insights from your research. Everything can flex and grow. And a real win is when we consider the data itself - mixed modality.
3rd party customer insights research often comes back to you as a PDF.
Customer Surveys might live in your Snowflake database.
You might have video recordings of interviews. Ai is really good at churning through this kinda stuff. Trying to manage this level of personalisation in any context would take a team of at least 10. And some companies do manage - they have massive BI teams that are connected with CRM and Web and all their 3rd parties and they're feeding them useful data. But most organisations i've worked with don't have this.
BACK TO THE GOAL
My requirements for home internet are around stable connections, high uploads for Teams calls with screen shares and camera on etc, data engineering means downloading many gigabytes of data many times a day. That need, or use-case at least, is entirely different from a family where someone's on Youtube or Tiktok, someone's playing games online and you're streaming Netflix on a 4k TV. And that differs from the Twitch streamer, the sports-fans, etc. So the principle here is recognition that these personas exist, and quick, painless research into what we do with all of this information. And if we can do this, even smaller teams can start solving real problems by treating people with nuance instead of one homogenous audience. Pretty neat, we think.
