Andrew Burgess, Founder & CEO
Gareth Ellis-Thomas, PCUK
Daniel Besson, PCUK
Amy Oberholzer, PCUK
Holly Matthews, PCUK
Neil Chong, PCUK
Country: United Kingdom
Organization: Greenhouse Intelligence Ltd
Greenhouse Intelligence is a strategic AI consultancy. We work across the AI lifecycle with a focus on building momentum and value during the early stages of the AI journey. We educate our clients about AI and develop AI strategies to ensure all opportunities are aligned with business objectives and are value-led. We help our clients build innovative and transformative solutions quickly, robustly, and ethically, or help them source and manage the very best AI vendors. We think big but act small – always starting with a low-risk proof-of-value. Sustainability, ethics, and the future-proofing of AI are embedded in everything we do.
Prostate Cancer UK, like many other charities, were looking at opportunities to maximize the value of their base and generate additional funds from their existing warm supporters. This money typically funds research, support, and services for men affected by prostate cancer, one of the most prevalent forms of cancer in the world. In the UK alone, one in eight men will be affected by prostate cancer in their lifetime. Prostate Cancer UK, therefore, needed to ensure that they could maximize the return from every campaign they undertook.
They felt that the traditional approach of manually segmenting their customer base could be enhanced using Artificial Intelligence, but they didn't want to simply hand over all their data to a vendor. They wanted to keep control of their data and have full transparency and explainability of their actions, but they did not have experienced data scientists as part of their team to code everything by hand.
Working with their strategic AI partner, Greenhouse Intelligence, they decided to use Dataiku DSS to build the necessary models that could predict their customers' propensity to donate. Whilst the charity was keen to maximize the return on their data and had an enthusiastic data analytics team, we needed to be able to convince the fundraising team of the value of this very different approach to campaign selection. We also had to deal with data that was not always as complete as we would have liked. In fact, using Dataiku highlighted specific issues with some records that the charity had not been aware of.
Greenhouse Intelligence brought their experience in AI strategy and implementation to advise Prostate Cancer UK on the best approach to achieve their goals. Without an experienced data science team in-house, and with little appetite to recruit one, it was decided that Prostate Cancer UK would use a low-code platform to do the 'heavy lifting' of the data science.
Greenhouse introduced three platform vendors to test out, building a proof-of-concept for the propensity models on each. Dataiku came out top across most of the criteria, including ease of use, breadth of functionality, and company engagement.
The initial models were built by bringing in all of the available CRM, gift, and appeal data into Dataiku. Working closely with the data analytics team and the fundraising team, the models were refined and checked to ensure they were giving accurate and meaningful outcomes. Dataiku DSS allowed the team to try many different algorithms with different combinations of hyperparameters quickly and efficiently.
In the workflow, new features were derived from the available data that would be more relevant to the model. There was also a large amount of data manipulation to be done to ensure that appropriate permissions were adhered to and that the final selection would only go to relevant people.
The three datasets that were used covered over 1.5 million customers, over 5.5 million gift transactions, and over 15 million appeal activities. In total there were 66 recipes used, which all led to a single output file that ranked each relevant customer’s propensity to donate. These scores were then fed back into the charity’s CRM system to enable campaign selection.
Whilst Greenhouse Intelligence carried out most of the early build work, they were shadowed closely by the client team who quickly learned how to manipulate data and build their own models.
Business Area: Other - Fundraising
Use Case Stage: Proof of Concept
The Christmas campaign is traditionally the charity’s largest and most important of the year. As this was the first time that the fundraising team would be testing AI-generated insights on specific segments, the score rankings were closely monitored for performance alongside the traditionally selected segments.
The campaign sent just over 100,000 direct mail appeals in the post to potential donors. Since each flyer has a material cost for printing and postage, being able to reduce the number of unnecessary appeal packs can have a large impact on the ROI of the campaign.
To assess the value of the Dataiku-scored donors, a ‘virtual’ campaign was created post-hoc where only the Dataiku scores were used to determine who would get sent the campaign. Selecting only the highest scoring customers (just over 50,000 customers) gave a 50% higher response rate, whilst those low scoring customers that did not get virtually selected would have actually had a negative ROI. This meant that the Dataiku-driven campaign would have been able to raise just as much income as the traditional approach, without the hard work of manually selecting segments and without sending out unnecessary appeal packs to people who would not respond.
This first proof-of-concept clearly demonstrated the value of using Dataiku-generated insights to raise funds for the charity. Whilst the charity will continue to use a combination of AI and human insights to build their campaigns, they will be able to refine the models over time to deliver even better results and become more confident in relying on the AI insights. This will allow them to develop campaigns quicker and more efficiently and be able to raise more of the much-needed funding.
Prostate Cancer UK would never have been able to develop the level of propensity models that they did without Dataiku. They do not, and have no wish to, have the necessary data science coding skills in-house to build the models by hand, but Dataiku gave the existing data analytics team the tools and skills to quickly create models.
Their previous experience using a vendor with pre-built models did not give them the control and flexibility that Dataiku ultimately did. In particular, the ability of Dataiku to explain the models’ outputs was key in building confidence amongst the fund-raising team and certainly aided the adoption of the AI-led approach.
Dataiku has delivered a range of benefits to Prostate Cancer UK that will only get bigger as they use the platform for more use cases. After demonstrating the success of the first model, they have since used Dataiku to predict those customers likely to convert to regular giving and those regular givers who are at risk of churning.
They are also currently building a model to predict those customers likely to leave a legacy donation, which makes up approximately 20% of the charity’s income. Dataiku managed, just in the first six months of the subscription, to cover its costs.
As more and more models are built the benefits to the charity, and therefore those who are living with prostate cancer, will only increase further.
Value Range: Hundreds of thousands of $