The future of student recruitment is here. So many times I have had university recruitment staff talk about their frustration with trying to find serious student prospects amongst hundreds of leads. The challenge is to identify those prospective students most likely to enrol, thrive at the university and graduate. This to improve their early engagement experiences with the university and streamline processes for them. Why expend limited recruitment resources on “tyre kickers”? Read on because Artificial Intelligence is here to help.
One part of Artificial Intelligence is “Machine Learning”. This is where we create advanced computer models which analyse large amounts of varied data and they literally learn from this and adapt their modelling accordingly. The models can make linkages across data sets which as humans we simply would not consider. At Marketing Decisions we have been learning ourselves in this space. Since 2014 to be precise, when we created our very first Machine Learning predictor for a Medical Practice. We have learned much since then eg:
- The biggest data sets do not necessarily result in better predicting models – in fact some data points may reduce performance
- Data needs to be prepared/transformed into states best suited for predicting
- The choice of model type is crucial to optimise performance
In essence, the “secret sauce” is in the (data) preparation.
The Role of Marketing Automation in Prediction
Most Marketing Automation vendors provide lead scoring as part of their solution. This is where we “guesstimate” how to apply scores. We use “gut feel” or maybe in more advanced situations, historic data analysis of trends to assign score uplift in order to identify the high scoring leads – those most likely to move from student enquirer to applicant or from offer to acceptance etc. These serve a useful purpose as a first generation Marketing Automation tool to help us to better understand our database of student prospects. However these models need to be maintained. They are fixed until we refine them manually and they tend not to be scientifically deduced. Sometimes the models simply prove flaws in our scoring criteria as we surface hot leads which fail to make the grade.
Machine Learning prediction on the other hand teaches itself based on timely new outcome data and adjusts its modelling accordingly. The advanced models also surface those attributes most influential in making predictions. This can help recruitment and marketing staff to pay particular attention to these, especially when they are unexpected, and we have experienced this.
Marketing Automation provides the perfect hosting foundation platform for AI prediction. Especially in the higher education space where invariably we bring a wide variety of data into the marketing automation platform for personalisation and triggering of campaigns: behavioural, demographic and related such as CRM data. The automation allows us to process, move and transform such data. Once prepared we can automatically transfer to our prediction model which returns a prediction score for each record.
Custom reporting in the Marketing Automation platform then allows us to surface prediction insights eg current database of high scorers and trends.
So Why Predict?
Prediction of future states allows universities to derive more operational efficiencies and optimise recruitment efforts. We can focus our resources more on those most likely to be impacted in a positive manner for both student and university. As we see in one of the examples below we can even predict those enrolled students most likely to thrive and graduate for this is also a key metric for universities as we strive to elevate academic performance. Our AI Predictor can analyse a complete database of prospects or applicants or offer holders at any point in time and assess likelihood to progress to the next stage of the conversion funnel. And do this with a degree of certainty which managers and planners need to make better informed decisions.
Higher Education Examples
International Student Recruitment
We have worked through many static first generation lead scoring models and seen some success with this. Our scoring model outcomes even steer personalised messaging in automated campaigns. University staff can list source countries most likely to convert directly from experience and generally they are correct. But there are always nuances and outliers. As these outliers unexpectedly convert, an AI Predictor will take account of this and adjust its modelling accordingly. One of the unexpected benefits of AI Prediction is that it mandates a level of data discipline within an organisation. If we need a particular data point to predict but staff are inconsistently collecting this eg at events, on various web forms etc then it reduces our ability to predict accurately. As the Predictor surfaces those attributes most impacting predictions then staff can focus on ensuring collection of these data points as a minimum and as part of our data preparation we may experiment with a range of additional data collection to test impact. An impactful data point will then be normalised into collection practices. As the Predictor finds international prospects highly likely to convert, recruitment strategies are prioritised to drive that conversion eg personalised invitation to chat with a student ambassador. This all happens within Marketing Automation.
Student Recruitment Funnel Stage
There are multiple funnel stages in the student recruitment cycle eg enquiry to application, application to enrolment etc. The more granular we become with stating our prediction goal eg predict an enquirer will apply vs predict an enquirer will enrol the better our prediction results will be. Different Universities that we work with have differing recruitment challenges. Some have no issue with acquiring leads at the top of funnel which may be converting well into the funnel, others lack volume at this point, others have volume but very poor quality. AI Prediction can be deployed to address the very specific needs of a recruitment funnel. Logically there are many influencing data points eg source of the lead (eg recruitment event, open day, website enquiry etc); location; prior education and school; course of interest etc. At different stages of the funnel, differing data points will have a range of impact eg higher in the funnel certain source countries may be more likely to convert to the next stage but lower in the funnel, different countries may be more likely to convert.
Much of our focus in higher education with marketing technology has been in the area of future student recruitment. However we have had great success in the few cases where we have focussed on student retention. We use AI Prediction to surface an at-risk prediction score for current students. Provided we are collecting the relevant data points in the Marketing Automation platform then we can predict. For current students this involves integrating a range of data points relating to their current studies and then raising a red flag when warning signs appear. Our goal in these cases is to provide the best student learning experience and to strive to see students graduating and even elevating academic performance.
And the Results
Our first prediction models back in 2014 were delivering 60% correct predictions. This was simply not of value, being so close to a 50:50 call – we could almost make a random guess and be close to this performance. So we experimented and we refined. We learned what made a positive improvement and what had little or no effect. Obscure changes such as redefining a postcode data point to distance as the crow flies to the campus instead had a positive impact. Today we consistently achieve three out of four predictions correct. In many cases we land at 80-82% correct. We see results delivering 30%+ improvement in conversion via outbound call when call lists are generated from AI predictions, ranked according to likelihood to convert.
And yet we continue to work through opportunities to improve. Our AI Predictor can predict almost any business outcome provided it is supported by sufficient available relevant data and can operate within most Marketing Automation platforms.