SaafRun by Sasse
Saafrun is an AI-powered facility management and ticketing platform developed for Sasse, a major cleaning operations contractor at Munich Airport (FMG). The project was executed through Digital Product School (DPS) over approximately 12 weeks in 2025.
The Project
The core challenge was the absence of a reliable, real-time feedback loop between cleaners, supervisors, dispatchers, and airport management.
The team conducted in-depth research, prototyped multiple solutions, and ultimately delivered a working product.
Product Vision
Be the leading facility-management ticketing platform. We bring the highest operational efficiency by making work intuitive, engaging and rewarding.
Dispatcher
Shift Managers
Cleaners
Client (FMG)
There were multiple user groups involved in the system, and they were all closely interconnected. The activities of each group significantly influenced the workflows of the others. Therefore, mapping the entire system was necessary to gain a comprehensive understanding of the overall flow of information.
Our effort to understand the full network and draft the complex web of calls, paper work and manual checks
Problem
The absence of reliable proof of work and a functional communication loop. The team pivoted to focus on creating a verifiable feedback loop that would benefit every stakeholder simultaneously.
Language Barrier
Multi-language workflow, miscommunications cause errors.
Paper Laufzettel
Physical task sheets cause double-entry work, tracking delays.
20km - Supervisor
Supervisor walks 20km/shift, just to verify cleaners work manually.
No proof of work
Airport client (FMG) has no verifiable evidence of service.
Dispatcher Overload
Managing calls + emails + manual entries + spreadsheets
Many problems were invisible to management — they lived in informal processes, verbal communication habits, and cultural dynamics. The team employed a rigorous mixed-methods research approach combining -
-
Stakeholder interviews with cleaners, supervisors, dispatchers, object managers, and QA managers
-
In-situ shadowing — including cleaning toilets alongside workers for direct experience
-
Prototype testing with real users across all roles
-
Secondary research on facility management digitalization
-
Airport data analysis for prediction feasibility
Post Interview - Core Pain Points Summary
From Left to right : 1) Cleaning Data from the Airport , 2) QM Touch Data from the Washrooms
Stakeholder Pain Points
Cleaner
As a cleaner,
I want to record the bathroom,
so that I can have a proof to send to the foreman.
As a cleaner,
I want to notify the foreman after a task is finished,
so that he can evaluate my work.
As a cleaner,
I want to notify the foreman after a task is finished,
so that he can evaluate my work.
As a cleaner,
I want to open the feedback loop part, so that I can prove the cleanliness.
As a cleaner,
I want to be notified every time I am needed, so that the airport to maintain the cleanliness level.
As a cleaner,
I want to be able to mark the task completed, so that the other employees have an overview.
Object Manager / Supervisor
As an OM,
I want to contact the cleaner,
so that I can ask about a waiting status
As an OM,
i want to be able to click a specific task, so that I can see its status
As an OM,
I want to filter the dashboard,
so that i can tasks by (urgent, in progress, reviewed by client),
As an OM,
I want to see the feedback of a task, so I can approve a task has been done
As an OM
I want to have a dashboard,
so that i can see all current tasks & employees in the current shift
Dispatcher
As a dispatcher
I want to receive calls about a sonderreinigung so that I can create a corresponding laufzettel
As a dispatcher
I want to approve the laufzettel so that I can forward it to the client for further processing.
As a dispatcher
I want to fill out a form for the Laufzettel so that I can fill the information in faster.
Stakeholder Journey Mapping
Cleaners
Dispatcher
Shift Managers
The biggest Human Problem in the current system was between these two totally different realities clashing.
Cleaners - They are dealing with language barriers, hwavy work load, broken communication
Managers and Dispatchers - Dont have time to check work, dont trust the staff
Hypothesis : The Feedback Loop
Introducing video proof for completed cleaning tasks will help supervisors and dispatchers verify work remotely, resulting in fewer ghost tasks, reduced supervisor travel time, and higher client trust.
This hypothesis drove the entire product direction.
Value Propositions
Dispatcher
Our Ticketing System helps Dispatchers who want to process incoming and completed cleaning requests, by removing repetitive tasks and multiple data sources/interfaces, and increase time savings (unlike the current manual workflow)
Shift Managers
Our ticketing + excel/mail automation helps supervisors who want to observe correct work of Sasse employees, by reducing trust issues in cleaners and work in supervising — and build direct communication without language barriers and a transparent workflow & feedback system (unlike current fragmented tools).
Cleaners
Our ticketing system + video recording helps the cleaner who wants to clean and gain more trust, by understanding and proving the tasks better — and have a pleasant work environment (unlike manual, unverifiable reporting).
Call, Mail
Bill
+ Video
Ticket
Ticket
+ Video
Ticket
Ticket
+ Video
Complete functioning feedback loop
Prototypes and testing
QR Video App
Goal: Proof of cleaning
Tested with: Cleaners
Ticketing Dashboard
Goal: Task Management
Tested with: Supervisors, OMs
Dispatcher UI
Goal: Emails/call processing
Tested with: Dispatcher
Gamification
(Treasure Hunt)
Goal: Cleaners Engagement
Tested with: Cleaners
Ticketing Dashboard
Goal: Language barrier
Tested with: Supervisors
From Left to right : 1) Experiment with the cleaners , 2) Briefing to supervisors/OM , 3) Treasure Hunt test with the cleaners
Challenges
1
GDPR & Video Privacy
Problem - Face blurring required before upload to comply with German privacy law
Solution - TensorFlow BlazeFace — real-time in-browser blurring before upload
2
Device Adoption
Problem - Cleaners won't use personal phones for work tasks
Solution - Designed for work phones; Sasse to provision devices
3
Airport Data Access
Problem - QM Touch live data requires formal FMG proposal approval
Solution - Scoped for Phase 2; architecture prepared for integration
4
Change Management
Problem - Supervisors hesitant: 'I want to test before explaining to my team'
Solution - Train-the-trainer model; supervisor gets app first, cleaners follow.
5
And some real world challenges on the way
Features - Their Impact
Cleaners select their native language (Romanian, Croatian, Albanian, Italian, German, English) at login. All tasks, locations, supervisor instructions, and notifications are automatically translated in real-time.
Role based Login, ensures effecient use of the limited number of mibile phones.
Context-based AI Video Review powered by Google Gemini API.
Mail-to-Ticket Automation powered by OpenAI API.
Eliminates double work. The system automatically ingests client emails, parses the data, and generates actionable tickets directly on the Kanban board without manual data entry.
Automated GDPR Face-Blurring via TensorFlow.js BlazeFace.
The application automatically detects and blurs all faces in uploaded videos.
Measurable KPIs
Efficiency
• Ticket Creation Time
(manual vs. AI-automated)
• Ticket Processing Time
(creation → completion)
• Revenue Variation
(ghost task losses)
Operational Execution
• Rework Rate
(tasks requiring redo)
• End-of-Day Task
Completion Rate
• Time in Backlog
(avg wait before assignment)
Service Quality
• Client NPS
(FMG satisfaction)
• Employee NPS
(cleaner/supervisor)
• Overall Dispute Rate
(FMG claims of incomplete work)
Key Learnings
- Shadowing generated insights no interview would surface
- Learning curve and Device adoption needed to be handled with care. Time was the most helpfull tool in this, but since we had not much of it, we had to bring in some strategies to neutralize it.
Research
- AI explainability is critical: users must understand why AI flagged something (Importance of explainable Ai and trust)
- Multi-language support should be designed in from day. It improves rate of success.
Product
- Halfway reviews with the client are pivotal recalibration moments
- In-situ testing at the airport revealed constraints no office prototype would catch. They were super helpful.
Process
Batch#23 Final Product Show
Life @DPS Batch#23
Final Product Show Presentation
One with the Team
