Co-designing AI with mash-ups
Collaborative workshops that not only resulted in new ideas, but showed how our different teams think.
Doom, gloom and big ideas
Right now we are passing through the year of the Rabbit, zodiac for ‘longevity, peace and prosperity’. If you are in fact a GPT model, that is probably how you ‘feel’. But for the rest of us, there is just a pinch of anxiety.
What does change look like?
I am a product designer. I work in a team building smart software to reduce our energy footprint. AI is just what the doctor ordered for the utilities industry. Legacy systems and archives of bad data have trapped many in a glacial crawl from manual to smart.
But the fizz of excitement is paralysing. While LinkedIn is alive with ChatGPT blogs and Midjourney logos, many companies are struggling to grasp AI’s potential for their use cases. Meanwhile, the fear of falling behind is real. And fear leads to big decisions.
Grasping for control
A couple of months ago, I decided that if we were going to see new technologies and priorities adopted at the company level, I wanted my teams to have a voice from the start. So I set up collaborative workshops with individual teams, so we might spark the beginnings of an opinion on where AI has value to our products. This way, we could bring ideas enhanced by our tactical user and product wisdom to an otherwise strategic conversation.
My main aims
- we explore how AI can directly enhance our products/services
- those with more knowledge on AI can disseminate this to others
- I can compare how different teams view AI and their products
Unexpected benefit
- I found a stark contrast in culture between product teams
Enter the mash-up
This piece was a ‘nice-to-have’ and not a priority for my workload. So I couldn’t spend too long on it. I wanted something easy for my participants to use. And I wanted a pre-existing structure that only needed tweaking.
Hyper Island was perfect. Their toolbox of over a hundred workshops is an Aladdin’s cave of wonders. Looking for the perfect template leaves me with another list of ideas for the future. On this occasion, I chose the ‘Mash Up Innovation’.
What is a mash up?
You have three groups of elements:
- Technologies (e.g. smartphone, laptop, GPS, etc)
- Needs (e.g. accessibility, handling time, metrics, etc)
- Products/services (e.g. Netflix, Google Pay, Amazon, etc).
You can brain storm these, I chose to pre-fill them.
Participants are asked to combine elements from different groups, to create new features or products/services. After ideation, individuals/teams share their ideas and can vote for the favourites.
So from the above examples: GPS + accessibility + Google Pay = detect your location to create a localised receipt with regional tax for your transaction.
Preparation: the groups of elements
I pre-filled all groups. I knew what this was about and I wanted to save time and energy to focus on ideas.
‘Technology’ elements
For technology, I lazily asked Bard to clarify what the subsets of AI were. I made it clear the participants were welcome to add anything that was missing. I nearly merged Deep and Machine Learning, but left that to be considered by the group.
‘Needs’ elements
I have a set of principles I’ve been validating for almost a year now, called the Data Delivery Checklist. It is inspired by The UX Honeycomb, Lou Downe’s 15 Principles of Good Service Design and leaders in API design excellence like Plaid and Stripe. They are also influenced by case studies of when data products go wrong.
If you are building an API or data service, you need to make sure that the journey it supports meets these principles in the experience it provides to end-users. I will definitely explore this in a future article.
Does ‘X’ data delivery service enable information delivery to be..
- available? (e.g. regardless of time, platform, expertise or location…)
- intuitive? (e.g. reactive, abstracting complexity, customisable…)
- informative? (e.g. offering feedback, extensive detail and event history…)
- familiar? (e.g. consistent with convention, data governance and brand experience…)
- accurate? (e.g. no ‘wrong orders’, everything is checked, fix problems, trustworthy…)
‘Products/Services’ elements
For this I used the known products and services for each team. I was open to exploring ideas that could enhance our company and wider platform, but still encouraged teams to think hard about their own areas of expertise.
Running an AI co-design workshop
Circulating invites.
I expected no-one to make time, as this is a future-facing activity. But PMs and developers alike saw the value. In fact, after the first workshops, teams became excited and numbers increased.
Ideation
I always find participants are more effective when co-creating individually and sharing at the end, rather than creating in teams. There are exceptions, but giving each person a chance to design solo squeezes more energy out of a group.
Sharing
I had to push for discussion sometimes, finding a panel of nods and smiles in response to a ideas being presented.
Voting
My favourite. I find watching dots move out of a pile and into clusters across Miro highly satisfying. And this was my foundation to gain insights in post-workshop analysis.
Learnings post-workshop
Grouping the various ideas by their common themes was fun. There were lots of intriguing suggestions. I can’t get specific as it touches on proprietary information. But ideas didn’t all apply generative AI, but other uses such as its ability to decipher new groups and trends in data.
There are some valuable learnings about team and product culture, and the closeness of each team to their various downstream users. One fun question I’ve asked at subsequent discussions has been; “imagine your best ideas without AI. Would this be possible?”
Validation
Some teams were more focused on preventing errors and using AI to enhance their role of a gateway to data quality.
Predict bad data
Other teams explored ways to use AI to enhance the value they add to raw data.
Inform users
Other teams took a reactive approach, using AI to make the error-handling pipeline more automated and efficient.
I was also interested in which of the five needs were the most popular and how this varied among different teams. There was a clear distinction.
Playing these team insights back to leads and HoDs has been one of the most satisfying part of the process.
Conclusions
A great experience. Not just to talk about AI, but to discover how the different teams think.
If there is a sudden adoption of AI at a company level, these teams have already kick-started that work and will have a seat at the table.
P.S. these sessions also leave me with some great ideas for the upcoming Kaluza AI Hackathon this August.
Suggested reading
Good Services by Lou Downe
