Why are some NGOs more successful in their efforts to use artificial intelligence (AI) efficiently than others?
By Oliver Norkauer
Working with 30+ NGOs
In the last 18 months, Omdena has worked with 30+ NGOs worldwide, creating numerous real-world AI solutions for them. While the projects were very diverse in nature, we found important similarities that went beyond merely delivering an AI-based solution.
In this article, we will look at the differences and similarities of our AI projects for NGOs. We will show the success factors on a journey to become an “AI-enabled NGO” and illustrate them with real-life examples.
Omdena’s projects were very diverse in nature: Some had a clear problem statement, others did not, and in others, the problem description changed during the project. Some projects came with their own data set, others started without any data. In some projects, we had to distill meaningful information out of literally thousands of pieces of text or social media posts, in others we analyzed satellite data, and in others, we had to extract and combine data from multiple sources in various formats.
Omdena’s clients were diverse, too. Some of them wanted to run only one project, while others ran multiple projects with us. Some clients wanted to gain insights from their data, and others wanted a prototype, others a solid working solution for everyday use.
Omdena’s clients have wanted to use a data-driven approach to either improve inefficient processes or to deliver at least one of their services at a much larger scale. They were willing to enrich their expertise and qualitative decision-making processes with insights and predictions generated by data and AI models.
When looking back, we realized two important take-aways:
- Our most successful clients used their first project to start a journey towards becoming an “AI-enabled NGO”, although most of them weren’t aware of it. (In the beginning, we weren’t aware of it either.)
- All clients went through the same phases in their efforts.
The AI-enabled NGO
An AI-enabled NGO is an organization that recognizes the value of data and uses AI-algorithms to deliver services and programs highly efficiently and at a large scale.
The journey starts where an organization realizes that it can use data to improve current processes and takes the first step to discover concrete options. Once the NGO has realized the potential of data-based solutions, it almost naturally continues on this path, and step by step moves towards becoming an AI-enabled NGO.
On the way to becoming an AI-enabled NGO, all organizations went through the same process:
- Phase 1: Discovery – “What can AI do for us”?
- Phase 2: Rapid Prototype – “Demonstrate what AI can do for us!”
- Phase 3: Productionize – “Have AI deliver concrete value to stakeholders”
- Phase 4: Capacity Building – “Use AI-based solutions to build in-house capacity”
We will now look into each of these phases in more detail and illustrate them with examples.
Phase 1: Discovery
In the first phase, organizations need to clarify two main questions:
- Which of the problems we face are suited for an AI-based solution?
- Which data sources are available within our organization?
During the Discover phase (between two weeks and two months), we will usually find new information that might refine the original answers.
Example 1: Change the problem description
Our project with “Impact Hub Istanbul” started with the following problem description:
Istanbul lies within a region of earthquake activities. Almost inevitably, an earthquake will strike the city again. If this happens during the daytime, families will be spread across different quarters of the city. A lot of roads will be blocked or unsafe. How can families be reunited after an earthquake? Which roads will be safe?
The first challenge was to identify what “safe” and “unsafe” mean in the context of an anticipated earthquake. When we started looking into finding secure pathways, we found that the problem of reuniting families means finding safe routes between schools, hospitals, workplaces, and homes, where all of these locations can basically be anywhere in the city.
So, the problem was not limited to families, but to everybody who needs to find a secure path between two locations, including other NGOs that need to deliver e.g. medical supplies.
Thus, the new problem description was: “Calculate the shortest and safest path between two locations in Istanbul after an earthquake.”
Example 2: Identify data sources
In the same project, the NGO itself had no data with which to start. So, we searched for open data, starting with city maps based on satellite images provided by the OpenStreetMap (OSM) Foundation. When looking at these maps, we realized that they showed roads within the city, but did not allow predictions which roads would be safe after an earthquake.
Which pathways would be safe? We decided that broad roads and green areas, i.e., areas without houses, would be safe, as smaller roads might be blocked by debris. As the OSM data does not provide road width information, we identified rooftops from these satellite images and calculated the space between these rooftops as a measure of road width.
With these new data sets, we could develop a prototype of an algorithm to find the safest route between two locations in Istanbul after an earthquake.
In phase 1, we clarified the most important questions. We next need to find potential answers and show them to stakeholders.
Phase 2: Rapid Prototype
After identifying the problem statement and data sources, phase 2 starts. In this phase, the NGO needs to verify the value of an AI-solution both internally and externally, so they get the desired stakeholder support, which in turn can lead to additional funding.
This phase should not take longer than two months.
Example 3: Rapid Prototype
The goal of a project with TrashOut was to “build machine learning models on illegal dumping(s) to see if there are any patterns that can help to understand what causes illegal dumping(s), predict potential dumpsites, and eventually how to avoid them”.
In two months, we have built a prototype based on data from TrashOut and combined this with two other data sets. The prototype not only showed existing dumpsites (image 1), but also predicted the probability of illegal dumpsites in the form of a heatmap (image 2).
Image 1: Show existing locations
Image 2: Predicted probability
Customer quote from Lucia Kelnarová, Project Leader Trashout
“Amazing work done in a super short time. We hope to implement the work and make an impact on the world.”
While prototyping is an essential step, it does not provide a final solution. So, as the next step, we need to create a reliable, proven product for everyday use.
Phase 3: Productionize
While phases one and two can be regarded as exploratory steps, in phase three, the prototype will be developed into a solution to deliver value for stakeholders. In this phase, the prototypes’ algorithms become more robust and often increase their reliability. To accomplish this, new data sources might be identified, the model will run through new training cycles and will need to be adjusted.
This is a major step in implementing a real-world AI-solution and a lot of organizations underestimate the effort required and experience difficulties in deployment. The “2020 state of enterprise machine learning” report by Algorithmia reports that about half of deployments take between eight and ninety days, while 18-36% take up to one year or longer.
In the Productionize phase, the NGO needs to have some technical skills in-house, but usually also uses external consultants. Ideally, these consultants were already involved in building the prototype.
In a two-month project for the World Resources Institute (WRI), the problem statement was to “create a machine learning algorithm that can be used as a proxy for socio-economic well-being in India”.
We first built a prototype, using both census and satellite data, which had reliability of 60-75%. In the productionizing phase, we added more satellite data sources, improving reliability to 85-90%.
Phase 4: Capacity Building
The first two phases were exploratory, the third phase puts the first AI-based solution into production, starting to provide value to the NGO’s stakeholders.
In phase four, the NGO starts a larger project, not only to create another AI-based solution but to also build up internal skills and therewith building capacity at a larger scale. This is a longer part of the journey, taking 3-6 months.
Quote from Saurav Suman, United Nations World Food Program
“The collaborative approach of Omdena is taking innovation to a whole new level with the idea of leveraging technology to bring in people with different capacities and work on a problem. The driving force behind this approach is the accelerated learning through collaborative spirit, mentoring and spot-on guidance. On top of all that are the humanitarian problems that Omdena is working on. WFP Nepal is proud to have worked together with Omdena on one of the projects addressing zero hunger “crop area identification project”. We believe this is the start of a long journey together.”
In the earlier projects, only a few NGO staff were involved as domain experts. In this new project, there will be more NGO staff and they will now contribute more broadly during the project.
Omdena’s people-, tool-, and process-based approach not only leads to a technical AI-based solution, but also to more collaboration among the NGO’s workforce. Our proven process ensures that everybody is actively involved, communicates, and contributes towards the common goal. In detail:
- Collaboration establishes a trusted, reliable, and non-hierarchical communication structure that can be expanded throughout the organization;
- Collaboration breaks down organizational silos, one of the biggest impediments to innovation and efficiency in complex environments;
- Agile techniques speed up collaboration and yield more creative solutions;
- Diverse teams not only lead to better results, they also lead to more openness;
- Working on projects that achieve goals at a large scale will improve loyalty and dedication among your workforce.
Example 5: An AI-based tool providing value
A project with the WRI aimed to provide a machine-learning-based methodology that identifies land conflict events in several regions in India and matches those events to relevant government policies. The overall objective was to offer a platform where policymakers could be made aware of land conflicts as they unfold and identify existing policies that are relevant to the resolution of those conflicts.
We developed a visualization app as a prototype within two months. WRI showed this app to sponsors and donors, and secured funding for follow-up projects, which expanded the scope of the project from India to other countries. This, in turn, leveraged the visibility of the project on a wider scale.
Quote from John Brandt, WRI:
“We’re really excited about the results of this project. My team currently uses the code and infrastructure on an almost weekly basis. […] We’re very excited that the results from this partnership were very accurate and very useful to us, and we’re currently scaling up the results to develop sub-national indices of environmental conflict for both Brazil and Indonesia, as well as validating the results in India with data collected in the field by our partner organizations. This data can help supply chain professionals mitigate risk in regards to product-sourcing.”
With AI-based solutions, NGOs can provide services highly efficiently and at a much larger scale. Omdena’s experience shows that an organization’s journey to becoming an AI-enabled NGO goes through four phases if they want to fully realize the potential of AI:
- In phase 1 (“Discovery”), they find out what AI can do for them;
- In phase 2 (“Rapid Prototyping”), a prototype product demonstrates what a real solution can do for the stakeholders;
- In phase 3 (“Productionalize”) the prototype is expanded and made more robust. At the end of that phase, the AI solution is available for everyday use;
- In phase 4 (“Capacity Building”), the organization is building up internal skills to fully use the potential of their AI-based solutions
Omdena has worked with 30+ NGOs, and it has supported NGOs on their journey to become an AI-enabled NGO. Omdena offers a project-, process-, and tools-based approach, which may be complemented with consulting and training services when needed.
- Case Study used for “Discovery Phase” (examples 1 and 2): https://omdena.com/projects/ai-earthquake/
- Case Study used for “Rapid Prototype Phase” (example 3): https://omdena.com/projects/ai-illegal-dumping/
- Case Study used for “Productionize Phase” (example 4): https://omdena.com/projects/ai-economy/
- Case Study used for example 5: https://omdena.com/projects/ai-environment/
- Case Study United Nations World Food Program (quote for phase 4): https://omdena.com/projects/cropclassification/
- 2020 state of enterprise machine learning | Algorithmia
- What it really takes to scale artificial intelligence | McKinsey & Company