When coming to work in the mornings, our fundamental goal – similarly to other Data Scientists, has always been to make a positive impact on the business.
We take the Machine Learning magic and deploy it within the organisations to make conversion rates go up, revenue to climb and costs to go down.
However, there are major issues with the Predictive Analytics delivery process which prevent organisations from truly benefiting from it. And it’s not due to the shortage of Data Scientists.
Those issues were the primary reasons why a while ago we took a break from back-to-back project delivery and looked for a solution that would make Predictive Analytics efficient and applicable at scale by organisations of any size.
In the last 2 years we also talked to over 50 Data Science teams around the world and with a few exceptions – all of them grimly agreed when we asked about their view of the issues.
“We need 25 Models ASAP”
There are different products to promote, different customer segments with varying interests, there are certainly many reasons and ways customers abandon their service providers – and those differences often cause the need for separate models.
To give an example – in one of the telecoms we identified a room for around 60 different Churn Models. Wenarrowed it down to 25 that made practical sense to build. Early churn, BillPay churn, Inactivity vs Port-out churn, models to predict 2 weeks ahead and different models to predict 8 weeks ahead, etc, etc.
That’s a LOT of models.
Add ca 10 Cross-sell models, 5 Up-sell models and 2 Risk assessment models for new customers.
And all that just for personal customers – what about the 1/3 of the company which sells phones to businesses?
We had a roadmap for 3 years into the future, knowing that at least half of the plan won’t happen due to other workloads.
“Traditional Data Science does not scale”
Model development is slow. But it’s not the Machine Learning that takes time.
The actual Machine Learning has been automated for decades – after all, it is the Machine which is Learning while we go for coffee or work on something else in that time.
Continuing the Churn case study – even with a team of highly skilled and experienced Data Scientists wemanaged to deliver 12 models within 2 years – most of them by the end of the 2-year period.
What the business needed was 25 models 2 years ago – and we could not possibly deliver that.
Because Data Preparation for Modelling was still a manual process.
We were coding and running SQL transformations day and night. The process known as Feature Engineering was consuming most of our project time and it was sucking the life out of us and our teams. It’s slow, error-prone and completely not-scalable. Every new table on the input side requires new coding, as does each additional Feature the we think of building.
It’s been like that on every project since the beginning of time and in every company we worked for in the past. Banks, Telecoms, Gaming, Insurance – the experience had been astonishingly similar.
A 3-year long roadmap for Predictive Analytics is a joke. Businesses need the models ASAP, not 2-3 years from now.
If we – Data Scientists can’t deliver those models quickly – we are not making the very much needed and expected impact.
With the increasing speed of new data sources appearing everywhere – data preparation has been consistently becoming even more time-consuming.
In the era of Big Data the traditional Predictive Analytics delivery process is clearly broken.
Xpanse AI does not solve all problems in the Predictive Analytics delivery process.
But it solves the most time-consuming ones by automating the Feature Engineering from raw non-aggregated data and integrating a couple more useful functionalities.
Likely there will be a need for Data Scientists to tinker directly on source data from time to time before taking a full advantage of the Platform but Xpanse AI covers the most common ways of storing information – and wewill keep widening the range of digested data sources.
Additionally, Xpanse AI contains fully integrated Machine Learning (because why not?) and generates much needed visual Insight, which helps to understand the underlying correlations.
Lastly – based on some reactions we’ve seen – Xpanse AI does not aim to automate anyone out of the job. It is meant first and foremost to empower.
With a tool like that you can address many business problems much faster comparing to the traditional approach. You can move between different predictive objectives swiftly, test tens of thousands of Features in the space of hours and produce Predictive Models effortlessly.
We hope you will find Xpanse AI helpful in making an impact for your business.