Ivan Krnić

Thank you, Gene. Hello, everybody. It's great to be here.

My name is Ivan Krnić, and I'm here to inspire you to throw a company-wide generative AI hackathon. We had one at CROZ, and it was great, and I'm here to tell you why we did it, how we did it, and what have we learned from it.

But before I take you on this journey, a bit about myself. I am Director of Engineering at CROZ, but here is my boss at home. As you can see, she is not really practicing generative culture.

I'm also running the 0800-DEVOPS podcast, and I'd like to thank all of my guests. Many of them are here in this room today. Thank you so much for these conversations.

I had also the privilege to write a chapter for O'Reilly's 97 Things Every Cloud Engineer Should Know.

CROZ is a Croatian and German IT consultancy. We strive to help our clients every step of the way, from ideation to operating services in production.

An important part of this support is embracing new technology and helping our clients embrace new technology. Now, with everything that's going on lately in the technology space, this can sometimes feel like drinking from a fire hose. And with all the technology publicly available, I can't help but wonder: has technology stopped being a differentiator today?

This is maybe too harsh to say, since many new business cases have been made possible today precisely because of the advances in the technology. Maybe a better way to put it would be that the real differentiator today is not the technology per se, but our capacity to navigate vast amounts of new technology and embrace what is needed and what is not needed.

And we found out that a good way to navigate vast amounts of technology is by using hackathons.

This year we did one around generative AI, simply because every once in a while, a new technology comes along with the capacity to change the world. And generative AI is definitely one of those technologies.

With generative AI, it's like with any other technology. In the beginning, the uncertainty is super high and our confidence to use the technology is super low. But eventually these two lines cross, and this is where magic starts happening.

But for these two lines to cross, it's not, of course, enough just to sit and wait. We need to practice. We need to learn about the technology. We need to experiment with the technology. Only by practicing will we become true masters of the technology. And we found out that a great way to experiment with the technology is by using hackathons.

Now, hackathons, of course, shouldn't serve their own purpose. They should be part of a wider strategy. So this is the approach that worked for us.

First, we defined strategic goals, meaning: what do we want to achieve with generative AI? Then we threw an internal hackathon, gathered all the ideas, evaluated them, prioritized them, and finally, we developed a roadmap on how to pursue these ideas further.

I wanted to go a bit deeper into this step number two. What does it mean to throw an internal hackathon?

For us, everything started with setting up an organization team. They were later in charge for everything that was going on around the hackathon. They did an introductory workshop. They set the date. They prepared all the logistics. They invited cross-functional teams to submit their ideas. They prepared cloud environments, created API keys. They even put some budget on them to keep things under control.

And when everything was done, they made sure that teams documented everything in their wiki pages. Based on everything that we learned, they defined new policies for using the technology, and we were good to go with new initiatives.

Now, when asking people to submit ideas for a hackathon, there is always this dilemma. Should we give them full autonomy to submit any idea they want? Or do we want to retain full control over this ideation process and get precisely those ideas that we want?

We knew from the beginning that an open call for ideas without any guidelines would overwhelm people. So we gave them some guidelines. We asked them to think about two areas.

How can generative AI help them be more efficient in everything they do every day? And the other area: how can generative AI help them deliver more business value for our users?

So in essence, we gave them enabling constraints so they can focus their ideas instead of spreading them around.

We didn't ask for any upfront business plans or any business case justifications at this point. We just wanted to cast wide enough net to catch as many ideas as possible. And there will be plenty of time later to evaluate these ideas and to talk business.

This is how our event looked like. The energy was amazing, and there were even more people at the event than we had anticipated. In terms of numbers, we had 125 participants working in 36 teams across four geographically distributed offices.

And after 70-something pizzas, and after God knows how much beer, and you need to know that one of these offices was in Munich in Germany, and these guys really know their beer, these teams came up with 36 ideas.

Now, we won't pursue all of these ideas, and I won't be talking about all of these ideas here, but I wanted to mention three of them just to spark a bit your imagination.

One team was working on CRM enrichment and predictions. Basically, they took descriptions from Salesforce opportunities, ran them through ChatGPT, and created a bunch of tags that were attached to the opportunity.

They also made this functionality for salespeople in motion so they can record voice notes. And these voice notes, through Whisper, were transcribed to text notes, which were also appended to the opportunities. And these texts and metadata were regenerated.

Once we had these enriched opportunities, we could do all sorts of things. For example, we used Einstein AI to forecast the probability of closing the deal. We were also forecasting the deal size. And based on these tags, we generated a number of tag-based insights that were later used by CTO, sales, and HR to make everyday informed decisions in the company.

So this was pretty neat.

Another team was working on an AI support agent. Imagine how cool would it be if a person could pick up the phone and ask a question, and then an AI support agent would respond in equally human voice.

Our team tackled this problem from the middle. They fed FAQ database in ChatGPT and got some very promising results. And right now, we are working on finalizing this end-to-end process with one telco in Croatia, including speech-to-text and text-to-speech part of this process.

The third idea is something that our marketing team was working on to make their life easier when submitting posts on social media.

They took one episode of the 0800-DEVOPS podcast. They transcribed it with Whisper in text form. They fed it into ChatGPT to get a blog post summary. And they also fed it in the Midjourney to get the visuals. And at that point, they had everything prepared for posting on social media.

So the whole event was pretty neat.

A couple of months later, we were invited by one of our clients, a large European bank group, Intesa Sanpaolo, to facilitate their generative AI hackathon. And this one also turned well. It was even larger.

In terms of numbers, we had three group companies participating with 65 teams. They managed to organize everything in a single location, which was excellent. And we had 21 of our coaches helping with the event.

But what struck me the most with this is the number and the list of the departments that participated. As you can see, this is basically the whole bank. And the main stakeholder of the whole event was board member in charge for technology. So we had the highest possible level of sponsorship of this event.

Now, the fact that the client organization, in this case Intesa Sanpaolo, invited a service organization, in this case CROZ, to facilitate their generative AI hackathon speaks volumes about the relationship, or better said, partnership, between client and service organizations.

Much has been said here about this collaboration. Ben Grinnell led some amazing workshops that I participated in. And my favorite talk on this topic is from Courtney Kissler and Suzette Johnson from 2021. I think they managed to perfectly summarize this collaboration from client organization point of view.

I always felt compelled to do something similar from service organization point of view, and I managed to distill my thoughts into a manifesto for sustainable service organizations.

This manifesto has four values that I believe every service organization should live by. First two values deal with how service organizations should operate internally, but other two values deal with how service organizations should collaborate with the rest of the world.

Now, this value number three, missionaries over mercenaries, is of course borrowed from Marty Cagan, and it describes the mindset and the attitude that service organization should bring to the table.

Value number four, community over zero-sum approach, underlines that we should all collaborate in this IT landscape. This is not a zero-sum game, and when the tide comes, it lifts all the boats.

And today, looking in hindsight, I strongly feel that good results of Intesa Sanpaolo hackathon were strongly influenced by us living by these values in this collaboration.

So what did we learn from this?

From technical point of view, we learned: you get what you give. So it's not a New Radicals song. It's just that the quality of generative AI results is proportional to the quality of inputs.

And one important thing that we learned six months now after the hackathon is that it is less about the specific AI technology that we use. It's more about integrating end-to-end this process in the organization. It's more about finding the right test data. It's more about anonymization. It's more about retraining.

In other words, it's more about connecting organizational islands in a company.

From organizational point of view, we learned that to evolve, we need to adapt, and that takes time. So we need to be patient. The change feels natural only if it's unanimous and organic. And that also takes time, so we again need to be very patient.

We also learned not to wait, because by waiting, the uncertainty will definitely not drop, and our competitors will only gain more advantage.

And for this last one, I can almost personally guarantee: whatever you do, you will be positively surprised with the results.

In terms of help that I'm looking for, I would very much like to hear the ways in which you have jump-started AI, generative AI initiatives in your organizations. And I would of course like to hear your thoughts and your ideas on how can we all together elevate this collaboration between client and service organizations.

Happy hatching, everybody.

Thank you.