SaveThePappa
AI-powered decisional tool
Developed with
Timeline
March 2023
June 2023
My Role
Concept Development
Service Design
User Research
User Testing
Concept Presentation
Tools
Figma
Midjourney
Maze
Summary
A project developed in the Envisioning AI course where both technical notion about Artificial Intelligence and design methods were put to test to reach a plausible system that could help solve a problem in the real world.
Brief
Choose a community based stakeholder and help it reach Net-Zero Goals through an Artificial Intelligence based tool that supports decision making.
Outcome
A food supply management system that connects Non-Governmental Organizations to increase the efficiency of the redistribution process and prevent food waste through the use of Artificial Intelligence.
Stakeholder Definition
Preliminary Requirements
While framing the stakeholders, the approach of 'community of practice' was adopted, which refers to a group of individuals who share a common interest or passion and regularly engage with one another. With this perspective in mind, the focus was primarily on non-governmental associations that had the potential to contribute to initiatives aligned with the net zero goal.
In order to ensure the project's success, specific criteria were established to guide the collaboration with the stakeholders:
The stakeholders’ needs should align with the objective of reducing carbon emissions.
The stakeholders’ requirements should be substantial enough to warrant the implementation of A.I.
The stakeholders’ activities should generate social value by enhancing community communication and engagement.
The data collection necessary for the AI model should be feasible or already available.
The Process
There were notable challenges encountered during the process of selecting the target stakeholder. The search for a suitable main stakeholder commenced with Off-Campus NoLo under Polisocial, as it had an association with Politecnico. However, after conducting a comprehensive interview with Mr. Francesco Vergani, the coordinator, it became apparent that none of the design aspects were applicable to our specific goal of achieving the 'net zero goal’.
Subsequently, the focus was turned towards the indoor market in Via Monza, where Off-Campus NoLo is located and maintains a seamless collaboration. The objective was to explore potential needs related to logistics management and food waste reduction among the market merchants. However, the interviews conducted with the merchants revealed that this small market, comprising only seven stalls, did not face any issues pertaining to food waste. Consequently, it was deemed impractical to design solutions for non-existent needs.
Later on, it was noticed that a project named ‘SpesaSospesa’ exists within Off-Campus NoLo that connects food donators and recipients within the neighbourhood, while the requirement of sensitive personal information made it impossible to initiate a data driven tool.
Despite encountering those setbacks, the project team maintained its focus on addressing food waste reduction.
The exploration continued, eventually leading to a connection with RECUP, a non-profit organization that collaborates with Off-Campus NoLo. RECUP is dedicated to collecting surplus food from markets, distributing it to families in need, and promoting a culture of sustainable consumption within the neighborhood.
During an in-depth interview with the head of volunteers at RECUP, it was revealed that a significant amount of food waste still occurred during the distribution process to downstream organizations.
Additionally, RECUP expressed a need for a visual representation of their quantitative impact thus far, which would support their promotional and communication efforts.
About the Stakeholder
RECUP
RECUP is a non-profit organization originated in Milan , dedicated to the transformation of lost economic value in unsold market food into social value. Its primary mission is to recover unsold food from both supplier markets and urban markets, effectively reducing waste and creating a positive impact in the community.
RECUP's operations involve collecting significant quantities of unsold food from supplier markets, which is then allocated and transferred to downstream associations that work closely with families in need.
In urban markets, a smaller amount of food is often taken directly on-site, minimising any surplus. This approach ensures that the collected food reaches those who require assistance most urgently.
The impact of RECUP's work has been remarkable. In addition to its operations in Milan, the organization has successfully expanded its services to Rome. Each week, RECUP collects over 8 tons of unsold food, benefiting more than 3000 individuals in need.
One of the prominent supplier markets RECUP focuses on in Milan is the Ortofruit market, where unsold fruits and vegetables are collected and distributed to collaborating associations such as Banco Alimentare and Croce Rossa.
Through its dedicated efforts, RECUP is not only addressing the issue of food waste but also making a significant contribution to the lives of individuals and families facing food insecurity.
Problem Reframing
Idea Development
Prototyping
During the development process, the team had the opportunity to undertake activities that helped them critically analyze the strengths and weaknesses of the project outside the university context. We imagined the project placed in a real-world context and carefully considered its feasibility.
Landing Page
After creating an outline that detailed the functional features of the service, the team proceeded to create a prototype of the landing page to showcase the project. The page displays the motto, values, and offer of the service, such as AI-powered supply redistribution suggestions, a strong connection between non-governmental organizations and data visualization assets for media presence. At this stage of the idea development the business, which would manage and deliver the service, and the service itself collide and are under the name Smart Food.
User Personas & User Stories
The team used the open-source AI software ChatGPT to create a user persona, representing a volunteer of the upstream NGO RECUP, and a user story related to their interaction with the so-far developed service. The user persona profile was created using segmentation types such as geographic, demographic, psychographic, and behavioural.
Based on previous desk and field research, the team reconstructed the phases of work in which the volunteer is involved in their daily life, including their motivations, feelings, thoughts and pain points.
The team iterated through a few processes before settling on a convincing user persona profile, thanks to the use of AI. In fact, the AI software integrates and cross-references data from multiple sources to validate and augment user personas in a shorter time.
After outlining the main characteristics of the user persona, a storyboard script was created to explain why the realized platform could be supportive in the organizational and decision-making process of an NGO. To do this, again through the use of ChatGPT, they simulated the dynamics of how the service takes shape.
Next, images were generated through the DALL-E tool, representing the different points in the storyboard, so as to have a visual representation of the context. This is where the first limitations of using artificial intelligence, especially open-source AI, arise, as the images not only had visual errors, but also because in order to make a suitable image, a detailed explanation of the sense of context for which the image is created is needed. Currently, it is the user who decides how to set up the image in its composition and what is within it, and not the AI to figure it out on its own.
Wizard of Oz
Using the storyboard previously elaborated, the team took part in a role-play activity in order to bring the service to life, receive peer-to-peer suggestions and feedback to further improve the work.
The interaction recited was the one happening between the primary user of the service, the upstream NGO volunteer, and the AI powered system; still side characters were added as they were essential so that the flow of interactions and actions reflected the reality as much as possible, thus the merchant at wholesale market and the downstream NGOs' volunteers.
The roles (volunteer, merchant, NGO and AI algorithm) were assigned among team members and some paper-based props were made, like unsold food and the AI dashboard screen.
The role-play mimics the actions performed by an upstream NGO volunteer. These actions include saving food collected from wholesale markets, inputting data into the dashboard, receiving food requests from downstream NGOs and indirectly from families, visualizing the suggested food distribution elaborated by the AI algorithm, and proceeding with the fair redistribution of food to downstream NGOs.
The feedback received from the peer-review in the exercise was useful for the team to summarize the dynamics of the service and to understand whether it was clear to someone not involved in the project.
The feedback was mostly positive and the only questions were about the explanation on some of the steps among the different stakeholders in the service.
Scenario Prototype
The techniques used in the ‘Wizard of Oz’ approach were used to design a short video. The goal was to create in 2 to 3 hours a short film summarising the main dynamics of the service. As in the previous exercise, cardboards, markers, and colours were used to create lo-fi prototypes of the final dashboard screens and the "Smart Scale" that will be used by the operator to upload directly onto the platform the food they will weigh. A short storyboard was created to tell the structure of the video, both as shots and speech. Finally, the team divided the roles to be played and who would be in charge of filming, editing, and dubbing.
In summary, the video shows a scenario in which the distribution of food by the NGO to the partner associations occurs arbitrarily and without the support of a digital database. Because of this, the associations receive quantities of food that will later be wasted by the recipient families or will not be sufficient to meet their needs. With the new service, on the other hand, food redistribution is done in a more controlled and conscious manner, thanks to the support of artificial intelligence in the decision-making process.
Final Concept
Thanks to the idea development exercises previously described the team found confirmation in the starting idea of the product service, refining inaccuracies and hidden details.
The service aims to address food waste in an urban context, such as the Municipality of Milan, and establish a greater network between NGOs to better support families in need. To accomplish this, the service will use digital touchpoints to collect and analyze real-time data from both upstream and downstream NGOs. This data will be combined and converted into decision-making suggestions to efficiently exchange unsold food.
Touchpoints
The touchpoint for upstream NGO volunteers involves weighing, organizing, and redistributing food. This touchpoint comprises two devices: a smart scale and a tablet application. The smart scale serves as the data collection point, while the tablet is used as the management and decision-making tool.
The touchpoint for downstream NGO volunteers is used to make food requests to the Upstream NGO, collect feedback from the family and manage the family numbers and type. For this reason this touchpoint is a mobile application.
AI Goals
The objective of using artificial intelligence in our product-service is to help our stakeholders with decision-making. The algorithms and models of choice were thought with this goal in mind. The collection of data allowed by our interface will feed the algorithms and improve the further suggestions.
Data
Knowing what the concept would be, the team was now in search of a series of plausible datasets that would have helped to make the ground on which to build the AI System and the whole service. At a conceptual level, the company and the system would work with both a series of data that the NGOs already possess and another one that they have to collect on a weekly basis in order to update the system.
What Data do NGO have?
Families
Each NGO possesses a database regarding each household that they help, and for each household, they know the registry information of the main person, and the number of people in the household, divided by children, adults, and elderly people.
Direct Input
For all the other data gathered by direct feedback, the team agreed to ask the person in charge to agree to the privacy consent.
The data needed to be collected directly are the ones related to the request from the downstream NGOs to the upstream NGOs, the actual donations (food allocations) from the upstream towards the downstream, the food collected by the upstream NGOs, and, finally, the feedback of each family in terms of food wasted.
These data will be useful to update the AI Model correctly and to keep upstream and downstream NGOs in touch.
A blueprint for the stock
One aspect that the team could not figure out from the interview was the part related to the food stock and collection. The only thing that the president was able to tell us was that the distribution of food is made by volunteers and they don’t keep track of how much food they collect from RECUP, for example.
Also, the making of Pantry Boxes (Pacchi Alimentari) is made by estimating the food based on the number of people per family. To create the food stock database so the team looked for a pre-made dataset to use as a blueprint. On Kaggle we found a dataset used specifically to forecast food donations from food banks, with data coming from food banks in North Carolina.
The most interesting features for us were the description, the storage type, and the gross weight.
Database Structure
Having answered our initial questions about the elements needed to build the back end, the team started developing the structure of the datasets, always making sure that their features were properly declared and interconnected so that they could be eventually merged to create the final big dataset useful to the AI Model.
The food stock database is used to keep track of the food collected by the upstream NGO and available to be donated.
Upstream and downstream NGOs are connected through the donations and request database. Each downstream makes a request to the upstream in terms of kilograms and for each request, there are one or multiple donations (food allocation) divided by the type of food given.
Finally, downstream NGOs and their families are connected through the pacchi alimentari, used to keep track of the precise amount of food given to each family and to evaluate their waste.
From Data to AI
AI Model
The datasets shown in the previous chapter will be used to identify the exact amount of food to allocate to each downstream NGO, in order to minimize the waste and accommodate at best their request depending on the food available and the behaviors of their families.
To achieve this solution the team decided to adopt a Genetic Algorithm, a method that emulates the process of natural selection applied to optimization problems, where the solution is the selection of the best element, with regard to some criterion, from some set of available alternatives.
In a nutshell, the GA creates a series of possible solutions (population), and each solution (chromosome) is composed of individual elements (genes), that in our case would be a specific quantity of food to be allocated to a specific NGO. Each solution is evaluated and ranked based on their fitness. Then the system combines the chromosomes to create the children that will compose the next generation of the population, making an actual evolution. In the creation of the next generation, there will also be processes of crossover and random mutations, to avoid an early convergence of the chromosomes. The process ends when a certain solution will have reached a specific level of satisfaction.
This process can optimize the time in which a solid and affordable solution is produced, even without a large amount of data.
The data will be collected weekly by the volunteers and the datasets will be updated accordingly. So, at least for the first period, the process will run weekly, mostly to compensate for the relatively low amount of available data. Then, when the model will be sufficiently trained, it would be possible to run it every two weeks or even monthly in order to reduce the costs related to the AI computation. e storage type, and the gross weight.
The Simulation
To evaluate the fitness we decided to run each possible solution in a simulated environment, composed of the data gathered from the volunteers and the feedback of the families. Using a Recurrent Neural Network we can establish the potential waste of each solution and use it as a parameter for its evaluation.
So the RNN of the simulation will predict how much food each NGO will give to each family and how much of it each family will waste. Knowing that, we can calculate the fitness (see next page), taking into account both the food wasted by the families and the possible food wasted in the distribution process as a collateral event.
Waste for Fitness
To determine the fitness of the solutions, in order to rank them, combine them, and produce the next generation until satisfaction we decided to use as a main parameter the overall waste to be potentially produced by each solution inside the scenario.
To calculate said waste we use these variables:
food_donated is the overall quantity of food given at the beginning, it represents the sum of all the food allocations that composes a single chromosome;
food_to_family[] is the array that represents how much food has been given from the downstream NGO to each family;
fw_distribution is the collateral food wasted in the distribution process, calculated by subtracting the overall food actually given to the families (ff_total) from food_donated (it is not always present but it is taken into account anyway);
fw_families_kg is the conversion in kilograms of the overall food wasted by the families, the conversion is needed because the feedback from the families will be in percentage;
waste represents the total waste from start to finish;
fitness is the conversion of the waste in % based on the total amount of food given at the beginning, so we have a fitness point that always goes from 0 to 100 and it is proportionally inverse from the waste: the more the waste the lower the score.
Service Design
Smart Food has evolved from being just a service name to now being the name of the company that provides it. This name conveys professionalism and highlights the main feature of our design, which is food waste prevention.
Save the Pappa is a service specifically designed for Non-Governmental Organizations (NGOs) that collect unsold food supplies from food retailers in wholesale markets and are committed to support families in need of food supplies. So far, the service has been able to help two cooperating NGOs, RECUP and Croce Rossa San Donato.
Service Statement
Save the Pappa is a food supply management system that connects smaller and larger NGOs to increase the efficiency of the redistribution process and prevent food waste through the use of an AI-powered system.
Service Description
The AI-powered system behind Save the Pappa aims at reducing uncertainties in food distribution. Through a stock management system, it reduces fatigue and mistakes in a time and energy-consuming process.
Service Mission
Save the Pappa aims at impacting positively on climate change by preventing food waste and empowering social connections to ensure food supplies to people in need.
Service Values
Connection - NGOs and people that were once distant are now connected to reach a common objective and build a trustful community.
Efficiency - Technology is used to improve a system that today is conducted manually and thus manages to reduce fatigue and time.
Scalability - With the help of AI, a digital platform can be scaled to handle large volumes of food redistribution, making it easier to expand and reach more organizations in need.
Fairness - By working with charitable organizations, Save the Pappa would help address issues of poverty and food insecurity, promoting greater social equity.
Offering Map
Save the Pappa offers functionalities ranging from optimizing food supplies distribution, improving media communication, and enhancing communities' networks.
In detail, our service provides downstream NGOs a fast and easy way to make food requests that are visualized in real-time by upstream NGOs; consequently, the redistribution process is made efficient and fair as the Save the Pappa system suggests precise food quantities depending on the nature of the request.
The data regarding the food collected, food requests, and food redistributed is displayed to volunteers so that data is tracked and graphs are made to convey trends of the process in a timeline.
Ecosystem Map
Based on how stakeholders use Save the Pappa, they can be labeled as direct or indirect users.
As displayed in the following scheme, the Ecosystem Map, both upstream and downstream NGOs directly use the service, whether to input food quantities and receive redistribution suggestions or save a food request coming from families.
Indirect users instead financially support the service, e.g. the municipality and PNRR funds, hand out unsold food, e.g. merchants, or are given food supplies corresponding to their demand, e.g. families.
The Service in a Timeline
To better frame the stages of interactions that happen between Save the Pappa service and the upstream NGOs, the team elaborated first a user persona, a volunteer of RECUP, and then a Human Agent Journey Map that is divided into actions, stakeholders involved, and touchpoints.
The upstream NGO volunteer first collects unsold food from merchants at wholesale markets. They use a smart scale to weigh the food, and the data is directly shown on the tablet the volunteer is using. The AI-powered system of Save the Pappa combines the collected food data with the food requests that downstream NGOs are demanding and elaborates a suggestion for fair food distribution. Lastly, the volunteer delivers the food to each downstream NGO.
On the other hand, the downstream NGO volunteer first consults families asking for food supplies, collects their food requests, and enters this data into the mobile platform of Save the Pappa. At the end of the process, they collect the food from the upstream NGO volunteer and distribute it to the families.
Financial Feasibility
To start the Save the Pappa service, the team envisioned receiving help and support from both the Municipality of Milan and the PNRR. This would be essential for creating digital and physical touchpoints, as well as training the AI algorithm. Once the service is in the hands of users, it will be sustained by their necessary enrollment in a subscription plan.
The basic plan allows users to collect data about food saved over time and offers data visualization assets.
To move to the advanced plan, users must use the basic plan for 2 months. This will allow the system to collect the necessary data for the AI algorithm to generate suggestions about the distribution of food to downstream NGOs.
Visual Identity
The visual identity designed for Save the Pappa service aims to convey a sense of simplicity, a welcoming atmosphere and playfulness while emphasizing the strong value of environmental sustainability that it brings.
The chosen color palette revolves around various shades of green, primarily bright and vibrant tones. Green is often associated with nature, freshness, and growth, making it a suitable choice for a service related to food. The brightness of the colors adds a lively and energetic feel, capturing attention and evoking positive emotions.
For the logotype, a rounded and welcoming typography is utilized. Rounded typefaces often evoke feelings of approachability, friendliness, and warmth.
The curves and smoothness of the letters create a softer and more inviting visual experience
Dashboard Interfaces
Once the user journey and functions were established, the team began designing the interfaces for our multi-touchpoint dashboard.
The work was divided into two separate steps based on priority of accomplishment. The first step was to design the Upstream NGO Touchpoint, starting with the information architecture, followed by wireframes, and finally high-fidelity prototypes that could be tested.
The second step involved designing
the Downstream NGO Touchpoints.
Following the same phases as the first step,
the team developed it while performing user tests for the first touchpoint.
Upstream NGO
To provide a more streamlined experience for upstream volunteers, the team decided to create a touchpoint consisting of two connected devices.
For this project, only the tablet part of the touchpoint was designed.
This decision was made to allow for the completion of the interface's complexity within the available time.
Information Architecture
The starting point of our dashboard design is the information architecture. The interface is divided in five main pages.
Home
The “Home” is a really simple screen were the user has an overview of the most important data: The current amount of food, the latest addiction in the stock and the most recent food distribution. These widget are also links to the full page.
Profile
The "Profilo" page is where our users can manage their personal information, choose notification settings, and connect to the smart scale.
This is also the screen where users can update their subscription plan for Save the Pappa.
Raccolta
The "Raccolta" page has two sub-sections: the "Magazzino" and "Ultimi arrivi" tabs.
In the "Magazzino" tab, users can view the entire stock of available food for donation, divided
by type. They can also manually add or edit any food entry.
In the "Ultimi arrivi" tab, volunteers can view the latest food added by the smart scale in chronological order.
Distribuzione
In the "Distribuzione" page, which is essential for the user's primary job, volunteers can preview all available food items. In the "Widget View" tab, they can allocate the preferred amount of food to each downstream NGO. Additionally, the A.I. suggestion will be displayed in these widgets.
In the "List View" tab, a list of all connected organizations will be shown, along with some data related to their food requests.
One of the most useful features of this page is that if a new request arrives from a downstream NGO, a notification will appear to alert the user.
The “Widget View” tab will update automatically, which means that they don't have to refresh the page to get the latest updates. This feature saves time and improves efficiency, as volunteers can respond quickly to any new requests.
Impatto
The "Impatto" page provides media communication services envisioned by the Save the Pappa team. It consists of two sections: "Conversione Risorse" and "Tendenze".
The first section displays important data related to the climate change impact of the Upstream NGO, which is a crucial feature that these organizations use to explain their value to the government to obtain further funding.
The "Tendenze" section also aids in visualizing their impact, but in relation to downstream NGOs. It keeps track of the number of families they help and the amount of food waste they manage to save.
Downstream NGO
After completing the first touchpoint, the team focused on designing the mobile application for Downstream NGO volunteers.
Information Architecture
Great attention has been paid to the design of the information architecture and the structure of this app's functions because they are meant to be used on mobile devices, often in busy situations (such as while donating food to a family). The application structure includes a home page, a request page, a feedback page, and a family management page.
Home
The home page is the first page the user will encounter, it contains information about the latest requests, the number of families and the food received from the upstream NGO. In this page the user can also modify the organization’s profile page.
Richiedi
The request page, or "Richiedi" in Italian, fulfills the primary user task of asking the upstream NGO for a specific amount of food.
Feedback
The feedback page collects feedback from families on food waste from the last distribution. This allows the training algorithm to calculate the fitness of the scenario that then gets added in the AI model, resulting in a more precise distribution suggestion for the next time.
Famiglie
The "Famiglie" page displays a list of all families that are part of the organization. Volunteers can add a new family or modify any aspect of an existing one.
User Testing
After creating the dashboard and app frames, the respective prototypes were created. To check the usability and understandability of the interfaces, user testing was carried out.
To do this, the Maze browser tool was used, on which two separate tests were created in which users were asked to perform five different tasks each.
These ranged from those more related to the service tools, such as adding food to the database or selecting quantities to distribute (with the dashboard) or notifying the amount of food wasted by households (with the app), to more generic functionalities such as accessing settings or editing the profile.
For each task was then asked what was the level of difficulty and/or suggestions or problems encountered.
Main Insights
Dashboard (12 testers)
Create a more explanatory introduction to give context to users who are not aware of the service.
At first glance, it is thought that the food dialers are for information purposes and are not controllable.
It was difficult for many to find the request history section, probably because of the choice of words used.
Application (8 testers)
There were no major issues with the usability
It was suggested to change the service naming from “request” to “receive”
In conclusion, we can consider the interfaces intuitive, although the dashboard needs some changes of naming and affordances related to the controls for the quantities to be distributed.
Fixes
To resolve issues with section titling that led to misunderstandings and not finishing required tasks, some dashboard titles were changed.
"Widget View" and "List View," which users said seemed to be different ways of displaying the same content, were renamed to "Distribuisci" and "Storico Donazioni," while "Tendenze" to "Tendenze Sprechi"
