How to Run a Successful AI Hackathon: A Playbook for Marketing Teams

• Start with business problems worth solving: Identify high-value use cases where AI can meaningfully improve efficiency, scale, or business outcomes.
• Bring the right people together: Involve executive sponsors, practitioners, technical stakeholders, and technology partners who can help turn ideas into working solutions.
• Prioritize end-to-end execution: Build a complete workflow within a manageable scope that demonstrates tangible value.
• Build toward a bigger vision: Put successful workflows into practice, expand their capabilities, and connect them over time to accelerate agentic AI adoption.

Over the past year and a half, I’ve watched AI fundamentally change how our engineering teams at Skai work. Developers who once spent much of their day writing code are now directing AI agents, defining outcomes, reviewing decisions, and managing multiple workflows in parallel.

I believe marketing teams are on the cusp of a similar transformation. One of the most effective ways to accelerate that shift is through an AI hackathon.

At Skai, we’ve run numerous internal hackathons and recently worked alongside clients, including SharkNinja and Haleon, to explore what agentic AI can do for their marketing operations. These experiences have reinforced an important lesson: with the right preparation, people, and technology, a few days of focused collaboration can become the foundation for a fundamentally different way of working.

Here’s what we’ve learned about making that happen.

1. Start With Business Problems Worth Solving

A successful hackathon begins well before the event.

Identify a collection of use cases that, if automated, would meaningfully impact your business. Consider tasks that consume hours of your team’s time, work you pay someone else to perform, or opportunities you lack the capacity to pursue.

Perhaps your team conducts detailed performance analysis for its top product category but rarely applies the same rigor to the other 29. Or a marketer in one region has developed an optimization process that could benefit every market.

For each use case, define four things:

• The trigger: When does the workflow begin?
• The inputs: What data, tools, and information are needed?
• The process: What happens today, and what decisions are made along the way?
• The outcome: What should the finished solution accomplish?

You might arrive with five well-defined projects or a shortlist of 15 ideas to narrow down during the event. Either approach works, provided each candidate has a clear business purpose.

2. Get the Right People in the Room

The people involved can determine the success of a hackathon before anyone starts building.

There are four groups I consider essential:

An executive sponsor. You need someone who owns the business area, has a vision for what comes next, and can remove organizational obstacles. A team can lose an entire day waiting for system access or departmental approval. An engaged sponsor helps keep things moving.

The practitioners. The people doing the work understand the details that leadership often cannot articulate. Where do they find information? What signals do they look for? How do they decide what action to take? Understanding how your best practitioners operate gives you a foundation to improve upon.

Technical stakeholders. IT, security, and data owners hold the keys to the systems your agents need. Make sure they are in the room or available throughout the event to help with authorized access, APIs, and other technical requirements.

Technology partners. Bring in partners with the expertise and infrastructure to support agentic workflows today. Can their systems connect through APIs or Model Context Protocol (MCP)? Can agents access data and execute actions? Can their teams help you build in real time?

Technical connectivity is another critical piece of pre-work. In the days leading up to our SharkNinja hackathon, we provided access to Skai’s MCPs and connected to SharkNinja’s, allowing teams to hit the ground running on day one.

Before the event, test the following:

• Credentials and platform access
• API and MCP connections
• Data feeds and availability
• Read and write permissions
• Network rules and security requirements

The importance of practitioner involvement was evident during our recent hackathon with Haleon in Milan. Local-market leads from across EMEA came together, with Publicis also participating, to explore Skai’s agent-native marketing platform.

In just 20 minutes, each team created a functioning AI agent using its own marketing data, exploring applications ranging from performance analysis to uncovering new business opportunities.

3. Set an Ambitious Vision, Then Prioritize

Start the hackathon by painting a clear picture of success.

What would it mean to complete an analysis across 30 product categories in two hours? Could a process that works exceptionally well in one market become available to every region? Could an insight generated by one system trigger an action in another?

These are the kinds of challenges that give teams something meaningful to work toward.

At SharkNinja’s Jailbreak 2.0 Hack Week, 800 participants came together over three days, with Skai and Palantir joining SharkNinja to explore how AI could connect consumer intelligence with strategy and marketing activation.

Our collaboration with SharkNinja included building workflows to:

• Connect Amazon Vendor Central information with actions, enabling AI to distinguish genuine issues from false alarms and carry out approved responses.
• Prioritize approximately 200 media alerts each week to identify opportunities that warrant action.
• Apply SharkNinja’s business rules to establish when AI can act autonomously and when human approval is required.

Once you establish the broader vision, narrow the scope to what teams can realistically achieve during the hackathon.

The challenge should be ambitious enough to generate excitement and achievable enough to produce a meaningful result.

4. Protect the Time to Build

Most of the event should be dedicated to building, testing, troubleshooting, and iterating.

Use a few well-placed checkpoints to maintain momentum. Depending on the size of the hackathon, you might check in every few hours or at the end of each day.

Ask three questions:

1. Are you on track to reach your goal?
2. What obstacles are preventing progress?
3. What will your final demonstration look like?

That last question helps teams stay focused on a complete, working outcome. It also creates an opportunity to adjust the scope while there’s still time.

For larger hackathons, smaller group check-ins can help minimize interruptions and give participants more time to build.

5. Build Something That Works From Start to Finish

Every hackathon should end with a demonstration of what the teams have built.

My preference is to see a smaller workflow that works end to end. Show me that an agent can retrieve information, analyze it, make an appropriate decision, and produce a meaningful result for one product category.

You can expand it to 30 categories later.

A hackathon gives you permission to simplify. Work with fewer markets, a narrower dataset, or a single campaign type. The important thing is to reach the finish line and demonstrate that the individual components work together.

That gives you something concrete to improve, expand, and eventually put into production.

6. Build a Village Before You Build a City

One of the biggest mistakes organizations can make when approaching agentic AI is trying to design their entire future operating model before building anything.

Think about how a city develops. You start with a village. Build a house, pave a street, add a hotel. Over time, the village becomes a town, and the town becomes a city.

I think the same approach applies to building your agentic marketing stack.

Start with a few use cases that deliver immediate value. Build them, learn from them, and connect them. Then develop additional capabilities around those initial workflows. Some components may eventually need to be rebuilt as your understanding evolves. That’s a natural part of the process.

You can have an ambitious long-term vision without knowing every detail of what your organization will look like 18 months from now. Each successful workflow informs the next one.

7. Turn the Hackathon Into Lasting Business Value

The most important question comes after the final demonstration: What happens next?

A hackathon delivers lasting value when people begin using what they built.

Spend the following days or weeks refining the solution, adding appropriate security measures, and testing additional scenarios. Get it into the hands of practitioners as soon as it’s ready.

Agree on success metrics before the hackathon begins and establish a process for measuring the impact of each action afterward. At SharkNinja, some of our first post-hackathon action items included measuring marketing results at 12- and 24-hour intervals and capturing why marketers rejected certain agent recommendations. These feedback loops help teams understand what works, refine agent behavior, and improve future recommendations.

From there, expand it to more markets or categories. Develop complementary workflows and connect them where it makes sense. A quarter later, bring the team together for another hackathon to make the next significant leap.

We’ve already experienced this transformation in engineering at Skai. Our developers increasingly define what good looks like, guide agents, and review outcomes. I believe marketers will experience a similar evolution.

An AI hackathon gives you a place to start. Bring the right people together, choose a challenge worth solving, and build something your team can use. Then build the next thing.

Ready to explore what agentic AI can do for your marketing organization? Connect with Skai to learn how our agentic AI capabilities can help your team move from experimentation to execution.

This article includes contributions from Gal Zohar, Chief Data Officer at Skai, and Tzach Zohar, Chief Architect at Skai.

---

Frequently Asked Questions