Artificial intelligence meets the heart of fundraising - an unexplored terrain with undiscovered potential. My master's thesis explores the symbiosis between advanced technology and non-profit fundraising. Supported by interviews with experts from the AI, fundraising and NPO sectors, I shed light on practical applications that could revolutionize non-profit organizations. This article focuses on the most promising approaches and invites you to take a closer look. If you would like to read more of my work, you can access my master's thesis here.
There is currently no standard definition of the term AI. In terms of practical relevance, I defined AI for the thesis as a subfield of computer science that deals with the research and development of intelligent agents that can independently solve certain problems using various methods. The survey of the experts showed that there are various possible applications of artificial intelligence in fundraising for organizations. The experts agreed that the most promising areas are communication and the planning and analysis of fundraising measures. However, it also became clear in the interviews that most organizations are not yet ready to use AI and achieve a sustainable improvement in communication and fundraising, which is why a look at the prerequisites for the use of artificial intelligence in fundraising should be given in advance.
Fundamentally, there must be an awareness of digital topics in general and AI in particular within the respective organization. This requires a management culture at the highest level within the organization that understands digital topics and AI as an opportunity and lives this across the board and carries it into the organization. An initial starting point here could be the recruitment of CTOs into the organizations' management teams. Furthermore, organizations must define the goals to be achieved through the use of AI, integrate them into strategic planning and provide financial and human resources.
On the technical side, an appropriate IT infrastructure is required on which AI systems can be set up and, on the other hand, access to the technical solutions themselves must be provided, which must then be tailored to the specific use case. Access to data is also fundamental at the technical level in order to be able to underpin AI systems. It is important to have clarity about which data is relevant and to collect it. Fundamentally, organizations need to understand what massive treasures lie in their data and that they should use them sensibly.
The planning and analysis of fundraising measures offers great potential for the use of artificial intelligence. An elementary component of this is target group segmentation, which asks the question "Who do I best address when, how and via which channel? "
This is where AI can be used to support fundraisers in designing their measures efficiently, as making predictions based on existing data, known as predictive analysis, is precisely one of the strengths of current AI systems. This involves using analyses, statistics and machine learning techniques to make predictions about the future based on existing data. Predictive models attempt to describe the relationship between a dependent and an independent variable. In the case of NPOs, for example, the dependent variable, the response variable, could be the probability of donations. The independent variable, the predictor, is used to make the prediction.
Which data can be relevant for NPOs is explained in more detail below. The data is used to understand the relationship between the variables, i.e. to recognize a pattern in which predictor variable or combination of predictor variables leads to the response variable.
"By uncovering patterns and trends hiding within your datasets, you can easily identify donors who may be willing to give more, or determine which programs offer the best return for your investment. And that means you can focus your limited resources and efforts on those areas where they're likely to do the most good." So says the IBM website. This fact shows that the topic of predictive analytics is already well established in practice and that there are organizations, albeit few, that use this technique. One such organization is UNICEF Netherlands, which, according to IBM, has more than doubled its response rate to door-to-door campaigns by using SPSS (IBM's predictive analytics tool). In addition to IBM's SPSS, other software solutions are also available for predictive analyses in fundraising. Examples include the Veera tool from Rapid Insight and the cloud-based LityxIQ solution from Lityx.
As a process, predictive analyses for fundraising measures take place in four basic steps, which are described in more detail below.
The basic first step is the alignment of the raw donor information. For this purpose, all existing donor data is fed directly from various sources into the analysis tool. If the software does not support access to different sources, the data must first be merged in a data warehouse. Data that can be accessed here includes demographic data (age, income, occupation, marital status, etc.), campaign data (contact history, responses and donations, results of test campaigns, etc.), survey data (feedback from social media, results from emails or surveys that provide information about wishes and preferences, etc.) and all structured or unstructured data relating to donor activity.
In the second step, the information is analyzed using predictive models to understand what donors will do next (anticipate). The models identify the ideal donor segments, weight the data and predict the probabilities of future events (e.g. how likely is a donor X to respond to a particular marketing campaign Y?). Another important point in this step is decision optimization, which additionally suggests how the information from the predictive model can be used most effectively (donor X is most likely to respond to campaign Y when approached via channel Z).
The third step is to integrate the knowledge gained into the operational processes and systems and to address the sponsors according to their personal preference (Act). The insights gained can also be used for strategic decisions (Which measure receives how much budget? Which activities pay off in the long term?
Since predictive analyses are not linear, the prediction is optimized last. New findings from operational implementation refine the existing sources and new data sources can also be added. As a result, the model improves with each repetition and delivers better and better predictions.
Predictive analyses can therefore help to plan all fundraising measures efficiently, analyze the results of the measure and incorporate them into the planning of subsequent measures. In order to segment target groups, a basic understanding of the target group is required on the one hand and, on the other, data of the appropriate quality is needed to apply such techniques. However, if the relevant data is available, a forecast can be established that makes it possible to target people very specifically for a new investment opportunity [in terms of projects or campaigns].
It is important to differentiate between how an organization obtains its data. In the case of supporters who have been with the organization for years, i.e. where there is a certain level of trust and the relevant data and preferences are known, additional profile enhancement through predictive analyses can create a win-win situation.
However, when it comes to exploring new sponsor data or exploring the data of people who have only recently become sponsors, organizations are operating in a highly sensitive area in which ethical and data protection issues need to be addressed. In general, it is easy to extract information from the Internet using web mining or web content mining and make it usable.
How far an organization wants to go here, whether it collects data about its donors through years of support and donor journeys, or obtains data from sources external to the organization, must be weighed up and decided on an individual basis. In general, a reflective approach to data and the results of corresponding analyses is required, as distorted images can also be explored. Predictive analyses only work if patterns are actually contained in the data structures. In many cases, this is the case, but the final step is always common sense, because every prediction is and remains a statistical analysis.
The fields of application for data analysis for NPOs generally go far beyond target group segmentation and the selection of suitable campaigns or contact channels. Data analyses and predictive models can also be used to carry out needs or impact analyses. The American section of Amnesty International, for example, worked with DataKind to analyze 5,000 requests for help received via the Urgent Action Network over the last 25 years. Patterns were identified that indicated which of the cases had escalated into crises in the past. The patterns can be used to estimate urgency ratings for new urgent action calls, which help to prioritize cases. Further application examples can be found on the DataKind homepage(www.datakind.org).
If an organization decides to work in a data-driven way, it must not ignore the issues of data protection, data security, ethics and privacy, as awareness of these issues will continue to grow in the general population in the future. Transparency when collecting data is therefore of great importance, as the issue of trust plays a major role in fundraising anyway and many sponsors still lack this in relation to digital channels. So if methods such as predictive analytics are used, which can have enormous benefits and effects, then this should be done transparently for the individual user and with their consent.
From a sociological perspective, however, it is not necessary to go down to a 1:1 level and make the donors "transparent", as it would basically be sufficient to look at segmented group levels. In a survey conducted by the Frauenhofer Institute for Secure Information Technology, the majority of respondents stated that they consider anonymity to be guaranteed from a group size of 100 people per data set.
IBM's "Leap before you lag" study, in which managers and employees from 330 large organizations of varying financial strength from 34 countries took part, shows that it is worthwhile for organizations to work in a data-driven manner.
Advancing digitalization is opening up new horizons in fundraising - and at the forefront is artificial intelligence (AI). AI is fundamentally changing the way we communicate, acquire donations and spread our messages. In an area that has traditionally relied on personal relationships and emotional appeals, technology presents both challenges and opportunities.
Whether producing mailings, creating email newsletters or social media posts, fundraising is all about finding the right form of communication. This also includes creating texts, a task that AI systems could take over in the future. There are already software solutions that use NLP techniques to automatically generate texts from a given database, such as AX Semantics, text-on and Retresco for the German market.
An AI system can be used to create personalized communication based on the knowledge gained about the target groups through data analysis. By using AI methods, fundraisers no longer have to use A/B tests to decide between a few text variants; instead, many differently tailored variants of texts can be created that can arouse a much greater willingness to donate or get involved.
With regard to the quality of the texts, it can be stated that in the final instance the emotional corrective of the person will remain indispensable. This is because components necessary for emotional appeals, such as empathy and feeling, will not be able to be replaced by software and algorithms in the foreseeable future.
In addition to writing texts for mailings or emails, artificial intelligence techniques can also be used to design communication with the user on the organization's website and to improve user conversions in a targeted manner.
On the one hand, the texts on the homepage, as described in the previous point, can also be created automatically on the basis of data. On the other hand, the content on the homepage can be dynamically displayed according to user preferences. Based on known data from web analytics, CRM, geographical data, etc., AI-supported applications can dynamically display content for each user group and at the same time analyze the click behavior of the users and immediately reuse the knowledge gained as to which displayed variant delivers better results (in terms of previously defined conversions). However, the website does not automatically become a self-runner, as the non-personalized variant should already be convincing on the first visit to the website.
The dynamic display of content on the website requires technical prerequisites, so the website must be set up accordingly. In connection with personalized content on the website, however, identity-related questions also arise with regard to the needs and wishes of users. What happens when users are always shown exactly what they are supposedly interested in? Is a certain discovery experience lost when visiting a website? It is therefore advisable to make dynamization optional as a convenience feature and thus controllable for the individual user. Users who do not want to use this option are shown the homepage provided by the online editorial team.
Another promising application for artificial intelligence techniques can be found in the area of donor or supporter services. The majority of inquiries sent to an organization by existing or potential supporters can be described as so-called standard inquiries (changes to bank details, requests for information, etc.). One of the experts from the interviews puts the proportion of recurring communication at around 80% and also states that on average it takes around one employee per 10,000 supporters to process all inquiries.
This is precisely where chatbots could be used. Chatbots are online dialog systems with which users can communicate via text or voice input and through which requests can be answered automatically in the form of text or voice output. This does not necessarily require AI, but the use of intelligent systems has the following advantages:
The use of chatbots or digital assistants in support services brings two significant advantages for organizations. On the one hand, chatbots are permanently available and not tied to business hours, meaning that inquiries can be processed quickly at any time of the day or night and users always have the information they require at their fingertips.
Provided the bot delivers good results, it can be assumed that the user experience for supporters will be more positive than having to wait several days for a human response. RaiseNow fundraising expert Nico said in an interview:
If, on the other hand, the economic perspective is considered, considerable personnel costs can be put to better use elsewhere. According to the calculation above, an organization with 150,000 supporters needs around 15 people in the supporter service. Assuming a gross employer salary of €3,000 per month and assuming that 80%, i.e. 12 people, are involved in standard communication, €216,000 in personnel costs can be invested elsewhere per year, assuming that only half of the work is done by an AI-supported chatbot.
In practice, it is unlikely that 15 people will be deployed in the support service of an organization with 150,000 supporters anyway, which is why there are often waiting times in answering inquiries, which makes the first advantage described even more meaningful.
Before a bot is used, it is essential for organizations to clarify what the bot should do. A distinction can be made here between information bots, which only answer questions, and service bots, which perform tasks and answer questions. The use of service bots seems to make sense for organizations.
It is also important to decide how the bot should be set up. At this point, organizations must decide between using ready-made bot frameworks from well-known chatbot development platforms (e.g. Botpress, Facebook's Wit.ai or Google's API.ai) and programming their own. Although programming on your own infrastructure is significantly more labor-intensive and costly, it has two advantages. On the one hand, the organization retains control over sometimes sensitive data of the sponsors (e.g. if address changes etc. are to be mapped via the bot) and on the other hand, the codes can also be used elsewhere in this way.
Once the essential decisions have been made, the use cases for the bot must be defined (e.g. "change of donor data" or "provision of information") and the conversation strategy developed. In order to understand the user's intention , the chatbot must correctly identify the utterances, intents and entities before it can react appropriately. In order to recognize sentences and intents, the bot needs AI rules that define what should happen with which intent. The bot should be given several phrases per rule so that the meaning of similar phrases can be recognized later. For example, the statements "I want to change my bank details!", "Change bank details!" and "Can you help me change my bank details?" should recognize the same intention.
Before the bot goes live, it should be tested extensively. There are automatic test frameworks for this (e.g. Botium), but alpha and beta tests are also necessary to test real human behavior patterns. Once the bot is available to the public, it should be continuously improved. Negative feedback should be incorporated and misunderstood utterances analyzed in order to achieve an ongoing learning process and thus a continuous improvement of the bot.
The scope of application of chatbots goes far beyond processing standard inquiries via their own homepage. Chatbots can also be used in messenger services (such as Facebook chat or WhatsApp) to communicate with users, send followers push messages for mobilization, initiate personal contact with employees of the organization or even take over the donation process if a payment option is stored (here, however, the inhibition threshold in Germany is likely to be quite high).
The bot could also act as an early warning system in order to recognize positive or negative trends. For example, if it is asked by users about the topic of "termination", the bot could send a signal to the fundraiser, who then initiates countermeasures to retain the supporter. On the other hand, the bot could also give a signal if there is a chance to raise the supporter to the next level of the donor pyramid (for example, the supporter asks the bot about the topic of "major donors" or "wills").
There are a few things to consider when using bots. On the one hand, users who interact with the system should be made aware that they are communicating with a bot. This avoids creating the expectation of communicating with a human from the outset and then disappointing them when this is not the case. In the same step, it should also be made clear to users what the bot is and is not capable of doing in order to prevent any disappointment with regard to how it works. If the bot reaches its limits in dialog, users should always be given the option to switch to human contact and, in order to preserve the maturity of the users, they should be given the choice from the outset whether they want to communicate with a bot or with a human.
There are certainly signs that some organizations are already addressing the issue. The Swiss Red Cross (SRC), for example, set a challenge to create a chatbot as part of the "BärnHäckt" hackathon. The requirements for the bot included the ability to integrate with messenger services, the ability to connect to the planned CMS and CRM system and the availability of Swiss payment methods, which indicates that the aim is to generate donations via the bot.
An overview of active chatbots is provided by the online database chatbottle.co, which lists over 10,000 bots for Facebook Messenger. The search term "non-governmental organization" lists 565 organizations that already have a chatbot in Messenger, including Oxfam Australia, World Vision USA and various Greenpeace sections.
In summary, it can be said that there are various ways in which artificial intelligence can be used in fundraising to relieve the burden on employees and drive forward the further professionalization of fundraising, which can counteract the constantly growing pressure on the donor market through efficiency and effectiveness.
It is undisputed that AI will play an increasingly important role in the future and the use of AI for organizations goes far beyond the functional area of fundraising, so that, for example, the integration of AI in the areas of HR or accounting would be conceivable. The automation of standard processes through self-learning systems in particular appears to have great potential for nonprofits in the future in order to deploy scarce human resources more effectively. However, the content-related work of organizations can also be supported by the use of data analysis.
We would like to take this opportunity to thank Katja Prescher, Maik Meid, Dr. Kai Fischer, Nicolas Reis, Prof. Dr. Christoph Benzmüller and Tobias Hübers for their time and participation in the interviews.