Machine learning misconceptions. Forget what you think you know!
Tim Paris – 4 March 2019
There is a lot of information out there about what machine learning can and cannot do for businesses and charities, but there is also a lot of misinformation! In this article we would like to address a few of the common misconceptions we’ve encountered in the NFP space.
Misconception Number One: “The model won’t work. Our donors are completely different!”
It’s true that every organisation and their donor group is unique. For example, donors supporting an environmental organisation may have very different motivations compared to donors supporting a health-based NFP. Donors in each group might churn or upgrade for very different reasons.
Understanding the differences between organisations is essential to predictive modelling, and is the reason why Dataro does not use some ‘master’ model that we try to apply everywhere. To measure true effectiveness and for scientific validity, when we approach your data we bring no preconceptions and make no assumptions about what might or might not drive donor behaviour. Rather, we train a model for each organisation based on your own historic data and performance, which means that each model is necessarily going to be different.
Having provided propensity models for some time, we also know that these models change over time even within the same organisation. For instance, churn may be driven largely by donor age in March, but by socioeconomic factors in September. For this reason, our system is constantly retraining models to give you the most accurate predictions at any point in time.
Misconception Number Two: “This all sounds too technical/I need to hire a data scientist!”
Machine learning certainly IS complex, which is why it has traditionally been utilised primarily by the major corporates. This is one of the key things Dataro set out to solve by developing a complete system that ingests your data, safely handles personal information, trains models and outputs your predictions. You don’t need advanced data skills on your end to handle the initial data input – we provide detailed and clear instructions so information can be simply uploaded and we have a technical team on hand if you require additional assistance. We’re also adding more integrations to our systems all the time, making seamless data transfers even easier.
Misconception Number Three: “How do you handle personal information? It’s too risky from a privacy perspective!”
Fair question. Privacy and security is an important consideration whenever an organisation is talking about data and personal information. Whilst Dataro has developed an extremely secure system, the best security when it comes to personal information is to not take any more data than is absolutely necessary. For that reason, our models need just a very small amount of personally identifiable information about your donors. All information is stored securely in AWS private clouds based in Australia and all personally identifiable fields are fully encrypted. We also encourage the use of anonymised, tokenised data where possible.
So there you have it! Adopting any new technology can be scary, but machine learning already touches our lives in so many ways. In our view, the potential upsides presented by clever propensity modelling means the technology will soon become the new best practice for fundraising teams.