The following is an AI summary of the event.
You can download the slide deck for this presentation at the bottom of the summary.
Overview
This session featured a real-world case study from Maria Ghlonti on delivering an AI product under tight constraints. The talk focused on how Agile practices were adapted in a high-pressure environment with fixed scope, budget, and timeline, and what actually made the project succeed.
Project Context and Constraints
Maria described a project for a Berlin-based client with strict limitations:
- 3-month delivery window
- Fixed budget from an innovation fund that could not roll over
- Limited flexibility in contract terms
- Ambitious expectations despite constraints
Despite this setup, the project achieved a 5/5 customer satisfaction score, which framed the discussion around what worked in practice.
Investing in People First
With little room to adjust scope or budget, the main lever was team composition:
- Focus on bringing in highly capable, adaptable people
- Prioritize mindset and collaboration over just technical expertise
- Treat the team as the primary driver of success in a complex AI environment
Maria operated formally as a project manager but applied a Scrum Master mindset, emphasizing transparency, teamwork, and process over hierarchy.
Defining a Clear Business Goal
A major early investment was aligning stakeholders around a single, clear objective:
- Move from ~4% of cases evaluated manually to 100% evaluated by AI with recommendations
- Spend significant time (1 to 2 weeks) in workshops to align expectations
This clarity helped guide prioritization and trade-offs throughout the project.
Identifying the Real User and Decision-Maker
Initially, the “client” was assumed to be a platform team. That turned out to be incorrect.
- The platform team was a supporting function, not the end user
- The actual user and decision-maker had to be identified and engaged
- A true product owner role was established with clear responsibilities
This shift improved decision-making speed and reduced misalignment.
Managing the Client Relationship
Maria highlighted a common issue in service-provider setups:
- Teams tend to over-accommodate clients to avoid conflict
- This can lead to poor decisions and weak outcomes
Instead, she:
- Set clear expectations with the product owner
- Positioned them as part of the team, not above it
- Encouraged shared decision-making and accountability
Agile Execution and Rapid Adaptation
The team adapted its working model as complexity increased:
- Started with 2-week sprints
- Moved to weekly sprints
- Eventually shifted to daily iterations and feedback loops
Key practices:
- Daily review of AI outputs
- Continuous fine-tuning
- Nightly batch runs to generate new results
- Frequent replanning based on real feedback
This created a tight learning loop between users and the AI system.
Using AI as a Support Tool
AI was used in limited, practical ways:
- Refining backlog items and acceptance criteria
- Generating additional ideas or options
However:
- The backlog remained a human-driven thinking tool
- AI supported decisions but did not replace them
Handling Data and Infrastructure Limitations
A major challenge was poor data and testing infrastructure:
- “Garbage in, garbage out” directly impacted results
- Traditional testing environments were not reliable
Workarounds included:
- Using pre-production and production environments for testing
- Running controlled, limited-access “beta” usage in production
This carried risk but allowed progress under constraints.
Key Takeaways
- Clarity beats complexity: A simple, shared goal enabled faster decisions
- People matter most: Strong team dynamics outweighed process limitations
- Real users drive success: Identifying the correct product owner was critical
- Agility must increase with AI: Faster feedback loops were necessary, not optional
- AI exposes weak systems: Data quality and infrastructure issues become immediate blockers
- Adaptation is constant: The team changed not just the product, but how they worked
Bottom Line
This case shows that AI product delivery is less about tools and more about alignment, feedback speed, and decision clarity. Under pressure, the team succeeded by tightening collaboration, shortening feedback cycles, and keeping ownership with people rather than the technology.
