What we've learnt from 30 AI-enabled One-Day Sprints
Earlier this year we built our AI website using GPT4. It turned my thinking about technology on its head. In three days we had the website ready. But it took three weeks for all of us to get aligned as a team. From this challenge we started thinking about one-day sprints.
All technology becomes cheaper and more accessible. This tendency has accelerated over time. The printing press took 300 years to generalise. The Internet: 15 years. ChatGPT had a million users in a week. The speed at which we can build things will only increase.
But the point isn’t technology. It’s people. We need to take a human-centred approach to building services and products. That’s hard to do when we’re working with dynamic and ever changing challenges. And artificial intelligence is creating some very complex challenges.
Today, an approach based on experimentation will outperform an analyse-to-predict approach. Planning-based approaches aren’t helpful if everything keeps changing. Enter one-day sprints. They’re the backbone of our experiments. Since July we’ve run 30 of them. This is what we’ve learnt.
Our one-day sprint
Experimentation is the basis of our one-day sprint thinking. Something that can be built cheaply allows us to learn as fast as possible. Something built quickly reduces the chance of falling for the IKEA effect or first-solution thinking. It’s less painful to throw something away that hasn’t used up much time.
There are two groups within a one-day sprint. The client team has three to five people with direct experience of the challenge to explore. There are a pair of us responding, and building on, where the client team pushes us.
Our sprints are based around deep collaboration. The client team joins three times in the day. For an hour in the morning. 20 minutes half way through the day. And finally 40 minutes at the end. Our theory is that it makes active participation from all the client team easier.
In a day we
- Unpack the challenge
- Document how everyone understands the challenge
- Document assumptions around the challenge
- Create a tangible artefact
And then we stop. The day is designed for building. Having the 5pm deadline helps focus on what’s important. From there it’s about learning. We create a cadence that gives time and space for reflection. It allows our client the time to explore what’s been built. It allows time for testing with end users. It allows time for internal conversations and decisions.
If you’re interested in a deep-dive Andy and I ran a webinar in August. The recording is here.
Design sprints aren’t new. They build on the expression, “Work contracts to fit in the time we give it.” We can get a surprising amount done with a tight deadline. Parkinson’s law explains this from the perspective that work will expand if given too much time.
The Google Venture model developed by Jake Knapp, John Zeratsky and Braden Kowitz is the most well known. The promise is that you can go from zero to tested prototype within five days. There are two challenges with the GV model. The first is that it is very dependent on the participants and what is being built. I’ve led GV-based sprints where I’ve felt very conscious that people were being excluded from the process. Because they’re excluded the team is missing out on these individuals’ valuable knowledge. The second challenge is how much resource a GV Sprint requires. It’s five days with four to six people who are focussed 100% on the sprint. That’s an expensive week for any organisation.
Other innovators have iterated on the model. AJ&Smart of Strategyzer both have workshops that are designed to last less than a day. AJ&Smart’s Lightning Decision Jam is particularly good at getting quick collaboration without wasting time.
Our thinking behind sprints builds on these.
‘Doing’ ditch digging
We’ve been through three iterations of the one-day sprint.
Our early sprints were based around ‘doing’. We didn’t bother to think about things. We only wanted to build and explore what the new technologies could build. How quick could we go with AI? How far could a single prompt push things? Was it possible to be done in an hour thanks to AI?
We’d adopted an experimentation-based approach. That meant we could see quickly that ‘doing’ 100% of the time wasn’t creating value.
Currently AI is an enabler. It hasn’t changed the rules. If you dig a hole that’s going in the wrong direction it really doesn’t matter that much if you’re able to dig it more quickly.
Stopping to think
One of the key points of Innovation is to avoid money, time or resource being sunk into ideas that shouldn’t be built. We wanted to be sure we weren’t building the one thing during these one-day sprints.
We made two big changes.
We added a pre-survey. You can see an example here. It helped us to get a sense of the size of the challenge. To see how aligned the client team was. To understand if we needed to do some pre-reading about the challenge’s domain. Presenting the results of the pre-survey became the first part of our morning kick-off.
We changed the kick-off workshop to be about unpacking the challenge. When we were ‘doing’ we focused too much on ideation. Ideation is awesome. But if ideas aren’t underpinned by a real-world problem they’ll move you in the wrong direction. Andy and I would then write down what our understanding of the challenge was. Once we’d done that, we’d write down what we’d focus on and why that would be our focus. It helped us be clear about what we believed, what we’d do and what we hoped we’d learn.
From a dot to a line
We wanted to learn. That was the point of the sprints. Build something fast to give an outsized amount of time to learning. But the way we planned in the work stopped us from learning. Our sprints were initially one-offs. That meant we had lots of energy before the sprint, and on the day of the sprint, but it disappeared soon afterwards.
Running sprints as a one-off event didn’t leave any space for thinking. It didn’t leave any space for learning. It didn’t leave any space for non-participants to get involved.
Our most recent change has been to move from being a dot in time to a line. At a minimum we’ll book in three days of building with enough space around them to let us learn. So far this has created the best of both worlds. We’re moving fast but we retain focus. We’re moving fast but helping everyone in the team keep up with the pace. We’re certain there’ll be future iterations but we’re excited about the value we’re getting to build.
We’d love to work with you on your challenges. AI gives us tools as innovators that we could only dream of a year ago. We’d love to explore how these new tools can enable your experiments. If you want to chat, book a catchup in my Calendly. Look forward to talking!
Uncover the underestimated impact of AI for nonprofits
Join us for our next webinar where Andy and I will showcase some of the most exciting, recent AI use cases across multiple sectors and industries, from Coca-Cola to Dogs Trust, and highlight insights and tools to harness its power for your organisation’s innovation journey.
15th November, 10am. Find out more and register here.