The Covid-19 pandemic put huge pressure on all not-for-profit organisations as donations dried up almost overnight and fund-raising activities were curtailed across the board.  Greenhouse Intelligence had just completed an AI strategy for our client, a well-known cancer charity, but the pandemic meant that our work from then on focussed on finding ways to use AI to increase donations into the charity. Specifically, this meant trying to predict donors propensity to donate so that they could bring in more donations whilst targeting fewer people.

Initially, to get results as quickly as possible, Greenhouse brought in a software vendor that had pre-built models for some standard use cases, such as predicting Individual Giving donations. This approach worked well, and the campaign during late 2020 demonstrated that using AI could materially increase the return-on-investment from their fund-raising activities.

Having seen the benefits of AI in action, the client wanted to exploit the technology further by developing their own in-house approaches and models, rather than relying on a vendor. This would give them greater flexibility, transparency and control over what can be predicted, and, it was hoped, even better fund-raising outcomes.

Campaigns have historically been run using segmentation techniques, with both warm and cold campaigns run through the year. Traditional segmentation approaches typically only consider a very limited set of factors, and individual donors end up being treated as a single group, rather than an individual with their unique background, desires and behaviours.

Artificial Intelligence (and specifically Machine Learning) takes into account many more different factors and is able to spot trends and signals that are undetectable using more traditional methods.

Most importantly, instead of splitting a database into basic segments, using an AI-based approach means that every donor gets their own propensity score, which recognises the reality that no two people are exactly the same. This can result in much better targeting, ensuring only donors interested in the messaging for that campaign are contacted.

AI is also much more dynamic: it can also be updated on a frequent basis across the year to take account of changing seasons or trends. This is particularly important when people’s attitudes are changing on a regular basis in reaction to the Covid-19 crisis – the models can be updated to reflect this is near-real time.

Within the client, data has, historically, not been seen as a strong enabler for the teams, although this is changing with the development of an in-house data team. However, as the client is unlikely to employ full-time data scientists in the near future, it made sense to focus on using a data science platform, which has less of a reliance on these specific resources.

The project sought to test the following scenarios:

  • Replicate and compare the previous vendor-based work
  • Create a new Individual Giving (IG) propensity model, and test it with the next campaign
  • Create a new propensity model for a different use case

Three different data science platforms were used to build all the models, and were compared with each other for accuracy, ease of use and functionality.

The comparison with the vendor’s IG model was very favourable, indicating that the model created by Greenhouse Intelligence would have out-performed the vendor model by up to 15%.

The IG model was then updated with the most recent data and used to predict the donation propensity for the current Christmas campaign. This is not an ideal test as the Christmas campaign is the largest of the year and tends to target most potential donors, therefore making any gains from the AI model harder to achieve. However, from the results that have come through we have already seen that the model has been very accurate in predicting those who have donated, and has been able to target additional donors to bring in further funding. The campaigns during the rest of the year will see much greater gains as they are targeting a smaller set of donors.

The third part of the project has focused on building a model that predicts the likelihood of a donor becoming a Regular Giver (RG), which are valuable long-term supporters of the charity. The model that has been developed is showing Recall scores of over ninety per cent in tests, and will be validated in the new year on a real campaign.

Using and testing the data science platforms was also a success: the comparison of the three different solutions produced a clear winner, and the charity has agreed to adopt that platform starting next year. Greenhouse Intelligence will continue to support the charity to develop new models and upskill the data team to make the most of the platform across the whole organisation.

If you’d like to understand how AI can benefit your business and customers, then please get in touch using the Contact form.