For charitable organisations that rely on donations, it is essential to use their funds efficiently and target donors. Understanding donor behaviour plays a key role in this - from the amount and frequency of donations to preferred donation purposes and payment methods. In this article, we take a look at how to analyse these aspects using Self-Organising Maps (SOMs) to gain valuable insights for your fundraising strategy.
With the help of Viscovery SOMine, we analysed the donation behaviour of 5199 donors of a non-profit organisation (NPO). The self-organising map helps to display complex data in a two-dimensional map and to identify homogeneous clusters.
For our analysis, we took various factors into account, including the average amount of a donation, the number of donations per year, the total amount donated per year, the month of donation, the proportion of free donations, the payment method and the purpose of the donation. We also differentiated between private individuals and organisations. For data protection reasons, we did not have any further information on the donors, such as gender, age or profession. Nevertheless, the analysis provides revealing information about donation behaviour.
The resulting model shows 9 clusters:
Project donor organisations: Organisations (especially companies) that donate almost exclusively on a project basis
Free donor organisations: Organisations (especially companies) that give a significant proportion of their donations as free donations
Frequent donors: Almost exclusively private individuals who donate very frequently
Direct debit donors: Almost exclusively private individuals who make fixed donations via direct debit
Exclusive free donors: Private individuals who only make free donations
Mixed donors: Private individuals who make both free and project-related donations in significant proportions
Project-A medium donors: Almost exclusively private individuals who donate primarily to Project-A and give medium to high amounts (on average around €300 per donation, or €360 per year, with some very high donations of up to €50,000)
Project A small donors: Private individuals who primarily donate to Project A and tend to donate smaller amounts (on average around €30 per donation or year, and up to around €150 per donation or year)
Project B, C, F & G donors: Private individuals who donate almost exclusively to Project B, C, F or G
Abbildung 1: Clusters of the different types of donors
As can be surmised, the average donation amount varies from cluster to cluster. There is also a clear variance within the clusters. Organisations in particular often make higher donations. The following illustration shows the average donation amount in the individual areas of the donor map.
Up to 70 very similar donors are assigned to each hexagon in this map - an average of 5.3 donors per hexagon.
The average donation amount is represented by the colour of the hexagons, with dark blue corresponding to €0, light blue to €30, turquoise to €70, green to €250, yellow to €1000, orange to €3000 and red to a donation of €5000. A logarithmic scale (with offset = 10) was chosen in order to be able to resolve the different areas well):
Abbildung 2: Mittlere Spendenhöhe, logarithmische Skala (Offset = 10)
What is also interesting in this context, however, is the donation frequency, which is higher for private individuals, especially for frequent donors, but also for direct debit donors and mixed donors.
In the following illustration, the colour coding is given by the donations/year, where dark blue means 0.5, light blue 1, turquoise 2, green 3, yellow 6, orange 8 and red 12. Once again, a logarithmic scale (offset = 1) was chosen for the visualisation.
Abbildung 3: Mittlere Anzahl der Spenden pro Jahr, logarithmische Skala (Offset = 1)
This results in the following more balanced picture for the donation volume calculated over one year. This is again a logarithmic scale from 0 to 5000 with offset = 10.
Abbildung 4: Mittlerer Spendenbetrag pro Jahr, logarithmische Skala von 0 bis 5000 (Offset = 10)
In this context, we are particularly interested in frequent donors, as they generate an ongoing cash flow that adds up to considerable sums over the year. Many of these donors also pay by direct debit, which is a sign of a closer bond between the donors and the NPO.
The following chart shows which projects the frequent donors have donated to most frequently. The percentage of donors in this cluster for the respective donation purpose is shown:
Abbildung 5: Auflistung der Projekte nach Häufigkeit der Spende
Finally, we are interested in which fundraising campaigns the frequent donors responded to with an above-average number of donations. For this purpose, a profile value (Hedges g*) is calculated, which relates the difference between the cluster and the population to the average statistical fluctuation. Only those differences are shown that are statistically significant (significance level 95% taking into account a Benjamini-Hochberg multiple testing correction).
Abbildung 6: Auflistung der einzelnen Fundraising-Aktionen nach Erfolg
It can be seen that Aktion-PP-nn is very strongly overrepresented. In fact, 71.9% of frequent donors were (also) approached through this campaign. Across the entire data set, this applies to only 2.7% of donors. Just like the second-ranked campaign-MF-nn, the campaign-PP-nn is a permanent campaign that extends over the entire observation period. Unfortunately, we do not have any more detailed information on the individual campaigns. The time-limited campaigns N21, N13 and N15 are also overrepresented in this cluster. Action-HS-nn and Action-KK02 are the most underrepresented FR actions. It can therefore be seen that varying target groups (here, for example, the frequent donors) are also addressed very differently by different campaigns.
Our model can help to analyse the fundraising campaigns with regard to previous donors and to define hypotheses as to which types of campaigns can lead to success with which contact person. Quantitative testing of these hypotheses is not possible with the model, as negative results (people who were contacted by the campaign but did not donate) are missing from the data. Such data would have to be collected when implementing future FR campaigns. Personal or company-related data on the individual donors would also be of great benefit for optimising target group selection.