Often a donor will score well on several Dataro propensity models. For example, a donor may be likely to give to an appeal (DM Appeal model) AND likely to be a mid-level donor (Mid-Level model, AND be likely to convert to recurring giving (Convert to RG model).
This happens because the propensity scores are derived from independent models and an individual might be 'likely' to do more than one thing.
So how should you choose between them? There are several ways:
- Do they score far better on one model than the others? This is a strong indication of the right course of action for a particular donor.
- What are you organisation's priorities? In the above example, if your organisation is prioritising growing a mid-level program, the you would include them in the mid-level conversion ask rather than aiming to convert them to recurring giving.
- As a general rule, prioritising models based on the value of associated gifts is also a sound approach. So if a donor had high scores on multiple models, you might prioritise as follows:
- Major Giving
- Mid-Level Giving
- Giving >$500 to an Appeal
- Convert to RG
- Appeal Giving
- In relation to Planned Giving / Gift-in-will, it is highly likely that these donors will score well on at least 1 other model. Planned giving activities typically run alongside other fundraising activities and often there is no need to choose between them. We would typically suggest including all likely gift-in-will candidates in planned giving activity, even where they score highly on other models.
FAQ: unusual scores - why might a donor have a higher DM Appeal >$500 score than DM Appeal score?
Occasionally an individual may receive what seems like conflicting model scores. For example, the model scores may suggest they are more likely to give >$500 (DM Appeal >$500 score) than they are to give >$10 (DM Appeal score). Why does this happen?
- The propensity scores are derived from independent models (i.e. a donor's score from on model is not a factor in the other models). This means it is possible, although uncommon, that different models may give inconsistent information about a donor.
- The ordering of donors within a cohort (i.e. the Dataro ranks) is usually the best data to use for campaign audience building, as it gives you the most likely donors in order, regardless of their probability. It is, of course, very normal that a donor might be ranked higher for giving >$500 and ranked lower for giving a smaller amount, because there might be a lot of people more likely than them to give smaller amounts.
- The issue therefore only arises when looking at model scores (i.e. probabilities), not ranks. The "calibration" of model scores (i.e. having them reflect true probabilities) is a difficult process and is less accurate for rarer events like giving larger gifts, as there are fewer positive cases of these behaviours in the training data. This means that models for rarer event (like giving >$500) are inherently less well calibrated that models for very common events (like giving any amount to an appeal). We therefore do not recommend that model scores are used to compare and contrast individual donors between models - the 'accuracy' of scores is measured after-the-fact and in aggregate.
- From a practical perspective, the issue of a donor scoring more highly in the DM Appeal >$500 model than on DM Appeal model makes very little different to your campaign. If the donor has scored well on either model then they should be included in the campaign, and if the >$500 score is better then they should typically be included in a higher value segment.