The stages are shown below.
In summary, the earliest stage of analytics competency is characterised by disparate data sources, some descriptive analytics and no dedicated team resource.
Next come dedicated staff and a change in culture to prioritise actionable insight. Lastly, predictive analytics capability is built on a single data lake, supporting data-driven decisions throughout the business.
A model is all well and good, but how are marketers moving between these stages of maturity. What challenges do marketers face in establishing predictive analytics capability?
In a new report (Achieving Predictive Maturity) Econsultancy and RedEye round up some feedback from brand marketers and analytics specialists.
As more companies begin to understand the possibilities of automation and machine learning, there’s a need to rationalise the time and resource needed to prioritise predictive modelling, and weigh this against potential benefits.
The report identifies four crucial aspects of effective predictive modelling. They are:
1. Appropriate sources of data
One of the most fundamental points to consider is whether data is indeed capable of providing an answer to every question that the organisation has. There may be data gaps. Matthew Curry, head of ecommerce at Lovehoney, picks up the baton:
When you’re building a new product, whether it’s some software or a physical item, are you going to be able to find all the data you need? I’m not entirely certain you are. We have sales data and buying patterns. We know our customers like rechargeables and different colours. So we build on that. That’s as complex as it gets.
There are many factors that determine the appropriateness of a data source, from regulation and privacy to volume and velocity.
Interviewees urged caution when branching out beyond first-party data. Nathan Ansell, director of customer loyalty at Marks & Spencer implies that third-party data is inherently less valuable than first-party data because it is not exclusive to your platform.
He says “A lot of the data we have exists within our own big data platform – transactional or weather data for example. Other sources of data around customer intent such as social media activity also exist in other organisations.”
2. Data cleanliness and usefulness
After deciding on the appropriate data sources to use, the cleanliness of that data should be assessed and, ultimately, its usefulness.
Lara Izlan, director of commercial platforms and operations at Auto Trader, picked up the theme of consolidation, warning that incorporating too many data sources (however clean) may ultimately be self-defeating:
Any new data sources that feed into the model can add value but you probably hit the law of diminishing returns after a while. Auto Trader already serves quite a large proportion of the UK car buying audience so we have already moved quite far along that data value curve.
Prioritising the data sources along the car buying journey helps optimise predictive models towards conversion. If you were starting with a wider consumer base and a platform where the media was aimed more at prospecting new users you might need more data inputs to get a purchase.
Of course, there is value to be found in third-party data sources, particularly using anonymous third-party data at the top of the funnel to target paid media. But marketers are focused on getting their own data in order first and can be surer of its provenance.
3. Automation and machine learning
Machine learning may be something that more and more martech vendors are trumpeting as a part of their software, but making use of these algorithms more broadly in your company’s analytics teams is not a plug-and-play scenario.
Auto Trader’s Lara Izlan describes the scale of the task:
Most people are looking at machine learning and trying to figure out if they can get the resource for it. Data scientists, technologies, not to mention the amount of cloud storage that you need. I’ve learnt over the last year that it’s not just about chucking data into a warehouse and accessing it.
Machine learning is a massive undertaking to structure, having all the data sources talking to each other and then making it all actionable in real time.
The new General Data Protection Regulation (GDPR), which comes into force in May 2018, is also affecting companies’ plans to skill up in this area. Consumers will be entitled to know how their data is being processed and can contest ‘black box’ style analysis that is entirely automated.
“We had great plans but now GDPR is coming in and it makes us question whether there is still a business case for machine learning,” says Katrina King, head of customer marketing at Direct Line Group.
King continues, identifying the potential difficulty when working with partner data: “Not knowing what is going to be okay under GDPR makes it difficult to move forward. The key thing will be explicit consent and we don’t have a sense today of how many partner companies will have that consent. Eighteen months from now when there is much more clarity, the companies and agencies that can prove they have their consent already in order will be at a distinct advantage.”
4. Meeting business objectives
The final step of putting effective predictive analytics in place is determining whether your activity is meeting business objectives.
Predicting whether a customer will click or buy in the short term is all very fantastic, but marketers must still consider the broader picture. Colin Lewis, CMO at OpenJaw Technologies, says “If you’re using predictive modelling for the right reasons, you’re doing it to build the best customer experience. The problem is when personalisation is designed around the idea that you only want to sell more stuff. Then you’d be just as well giving the customer an offer.”
What Lewis is getting at is that predictive modelling requires a lot of management for something that is only going to be used to move the needle slightly. Instead, it should really be used to enhance a service, for example, to personalise curated content in the case of Channel 4’s All 4 platform.
Richard Clark, marketing director of N Brown Group gives us the final word on the agility of leadership required to make best use of predictive analytics:
As leaders you have to know and believe in your strategy. If the model shows up something of interest, you have to make quick decisions about whether or not to pursue it. I’m all for change when it’s appropriate but not about just jumping on things because they look interesting.
Subscribers can download the Achieving Predictive Maturity report now.
Also, check out Econsultancy’s Advanced Data & Analytics Training.