The travel problem your project solves, including challenge(s)
The global COVID-19 pandemic and efforts taken to fight it had never been experienced by modern society before. We quickly realised that social distancing is a key factor in our post-covid “new normal”. Travel is not going to be the same and passenger behaviours have changed forever. Transportation companies need to provide an outstanding passenger safety experience not just during the crisis but in years to come.
A few days ago, the Spanish Government published a new regulation for buses: married couples can sit together but the seat next to a single passenger must remain empty. One of our team member's family own a small bus company business based in the south of Spain, focused on recreational trips. This kind of regulations imposed by the government don’t always help companies, nor passengers. Fixed distancing rules do not fit every business and are not scalable.
For shorter trips, the passenger has a myriad of transportation options: bus, bicycle, car, taxi, train, etc. For longer trips, often passengers have no choice but to travel by air: to see family, to see friends, to see the world and to live unique experiences.
The Trave/Scrum Challenges that we want to address with our initiative are:
- Hygiene and Health
- Travel & Experiences
Your Solution and what it does
DaaS (Distancing as a Service) is a service that helps airlines guarantee safe, dynamic seat allocation using Artificial Intelligence.
Our system generates a risk profile per passenger based on data like age, itinerary or booking group. Then, these risk profiles are used to allocate seats for each passenger. The process leverages an AI algorithm that calculates the best distribution, maximising social distancing between passengers. As a result, passengers’ trust is increased and airlines can sell more seats.
How did you build it?
DaaS is based in an REST API model that can be easily integrated with any Airline management system. The system overview is availble in the project gallery.
There are three different communication points between the airline and DaaS:
When the Passenger Name Record (PNR) data flows from the PSS (Passenger Service System) to the Departure Control System (DCS), the airline sends a request to do an initial seat allocation. As part of this request, the following data is provided:
- Anonymised extract of PNR Data including Itinerary, Booking/Group id, Age Details (Optional) and Special Service Request (Optional).
- Aircraft code
- Flight itinerary Using the PNR Data and 3rd party services (Sitata, Governments data…), DaaS calculates the risk factor for each passenger or group of passengers. This calculation, along with the aircraft’s seat map fetched from Amadeus’ API, is provided to our machine-learning model, which generates the optimal seating allocation.
Get Passenger Seat
At check-in time, the airline requests the seat for a given passenger id. The allocation process has been previously completed and stored in a high available geo-replicated database, so this process is blazing fast.
Explainable Allocation Seat Map UI
The airline can visualise how the algorithm has allocated the plane and the risk factor for the different groups. This is useful for internal operations and to inform cabin crew.
What are you proud of?
Making an impact
We are really proud of the deep understanding that we have gained from the airlines industry. With this knowledge, we have been able to create a disruptive solution that solves a real problem.
We have spent a big amount of time to find the right balance between innovation and process disruption, and we have finally been able to design a solution that brings good benefits minimising changes in the airline process and systems. Special thanks to our mentors Daniele and Helene for helpiing us with this.
Apart from the solution, we are really proud of our interactive playground that allows everyone to test our APIs. We couldn't find any open source project fitting our requirements, so we developed our own in record time!!
Additionally, we have been able to connect our system with different 3rd party APIs and deliver it using cutting-edge technology provided by AWS.
What is the next step for your solution and how will you take that step?
Firstly, we will share our value proposition with the target market so we can gather feedback that will help us to define the final product.
Then, we will select the right channel and messages and we will start looking for investment support to develop a Minimum Viable Product.
In terms of features, we have put together a product roadmap based on the feedback received from mentors and the scope that we have decided to leave out of the hackathon.
What tech did you use to build it? (e.g., APIs, tools, other relevant details)
Amazon Web Services to take advantages from a cloud-native application:
- Lambda functions in Node.js & Python. AWS serverless technology offers us great scalability and more flexibility for our deployments.
- DynamoDB. Chosen due to its high availability and easy scalability.
3rd Party integrations
- Python scikit-learn - python modules for machine learning and data mining.
- Gradient boosting for regression and classification.
- Flask to make the model accessible as an API.
- Amadeus Fake PNR dataset
Links to test the solution (e.g.,Github, Website, App, adobe)
Is not easy to showcase and test an API... but we have created an interactive playground (which serves as API documentation too) where you can play with the algorithm.
The code for all the different components of the solution is available in the DaaS Github repo
How are you dealing with GDPR regulations?
DaaS doesn’t require any identifiable information from the passengers, only focuses on qualities/risk factor. The list of passengers airlines send for each flight is anonymised
Does the airline need to wait until the flight is closed to start assigning seats?
Not at all! It is totally up to the airline. However, our suggestion is to start allocating seats between 48h and 72h before the departure, so DaaS has a better view of who will be travelling and can also take advantage of real time risk data for greater optimisation.
What about the revenue generated by seats ancillaries?
This has been one of the main discussion points during these days.From our research, we have found that seat ancillaries constituted 4% of the total revenue in average for legacy carriers, per flight (reference : Jay Sorensen report 2018). The trade-off between customer safety and revenue generated by seats ancillaries seems reasonable in these crisis times where regaining customer trust is a must. In order to leverage an effective social distancing, the airline could adapt to two different models:
- Selling a limited amount of seats, so the rest of the passengers can be allocated accordingly.
- Offer other generic purchase options: window, aisle, front, back, etc… without fixing the actual seat number.
This will really depend on the airline. For this hackathon, we have decided to focus in our value proposition, allowing the airline to leverage social distancing with minimum changes to the current check-in process.
How do you make money?
We are a B2B business, based on a flexible ‘pay per use’ model depending on the volume required by the airline.
Are you considering the aircraft weight & balance?
We don’t directly manage aircraft balance. However, one of the advantages of allocating seats close to the departure time is that the algorithm can provide an evenly distributed allocation, making it easier for the weight & balance process. In the future, weight & balance parameters could influence the seat allocation to optimise fuel consumption and to improve safety.
Could this service be applied to other industries?
Definitely yes. The AI from our product and our concept of ‘Distancing as a Service’ can be adapted to buses, trains, ferries, cinemas, conference rooms, etc… We have had this in mind since the beginning. We decided to focus on airlines because aircrafts are a real pain point for passengers, as it is the only transportation method able to cover long distances effectively.