We recently celebrated machinable’s first anniversary and are delighted to report that our quest to incorporate advanced analytics and machine learning as a mainstream part of the business optimisation toolkit has come on leaps and bounds.
We’ve helped clients across diverse sectors (financial services, logistics, the social sector, healthcare and the hedge fund industry) address a range of problems (from the strategic – “what will be the impact of AI on my business?” – to the specific – e.g., “how do I identify high quality, high propensity targets to increase acquisition performance?”).
Nonetheless, one thing unifies them. All, without exception, have been business leaders faced with material business opportunities and problems. Most have been leaders with commercial responsibilities.
It has been tremendously rewarding to help them and, intellectually, a super-stimulating ride for us.
We thought we’d share with you four things we’ve learnt along the way.
- Machine learning can indeed reach the parts other tools cannot reach
Machine learning techniques can lend much greater precision and insight to business decisions in a way that goes beyond traditional analytics.
For example, in developing more targeted acquisition and retention offerings: by readily incorporating unstructured information (e.g., survey free-text responses) into predictive models, machine learning bridges the gap between purely quantitative and purely qualitative approaches to reveal much more about customer or employee motivations and value-sets – and in a rigorous manner that exploits a machine’s innate ability to pattern-recognise.
- … but needs to be wielded in a very human-directed and human-centric manner
The value of an analytic model alone however can be minimal without strong human involvement in design and deployment.
Spending lots of time upfront with practitioners to identify the factors likely to be predictive is critical and enhances the potential for undirected discovery offered by machine learning. We have seen this on multiple occasions and in very different contexts. At a hedge fund thousands of man-hours are devoted to developing and rigorously testing predictive variables conceived of based on prior financial analyst experience – this being the “secret sauce”, rather than the particular machine learning algorithm used. In identifying high quality recruits in social care, careful dialogue with care staff themselves provided insight on the indicators of commitment, competence and suitability that feed a predictive model to improves conversion by 200+%.
Moreover, the model is often only part of the solution. In our work with a parcel delivery company, addressing poor delivery performance needed – in addition to a predictive model – new staff incentives, new operational controls and revised commercial terms with retailers to effect the desired changes in human behaviour.
- Be expansive in thinking and application
We get hired to solve – in a data-driven manner – big, complex questions that lie at the heart of company performance. When taking the trouble to do so, it would be negligent not to capitalise on the broader opportunities oft-times presented.
Building a model to optimise the allocation of a mobile workforce to geographically dispersed clients generates value (releasing, for our client, 35% of staff capacity to serve new revenue-generating customers), but using the insight to target recruitment to those locations which most increase travel efficiency generates more, as does more accurately pricing new commercial contracts based on predicted workforce loading rather than intuitive experience. Similarly, full value from lead scoring customers in a consumer finance intermediary comes from going beyond reprioritising outbound call activities to constantly adjusting the channel / delivery model to the quality of the customer pipeline, and still further by evaluating employee performance based on returns adjusted for difficulty to convert.
- Don’t let doubts over data quality get in the way
To a greater or lesser extent, our clients have each had doubts over the quality of their data. A cursory look at the data may seem to confirm this. There are, however, reasons for optimism.
Our clients’ data usually contains much more insight than meets the eye. For example, in separate organisations we discovered tell-tale signs of customer intent to buy – from the level of personal effort represented by their choice of contact channel, to the level of intrusiveness of the time of day and day of week they were prepared to interact on – in data that at first sight seemed distinctly unpromising.
Above and beyond this, our clients derive huge value from the precision with which machine learning has enabled them to specify requirements for large-scale data improvement. Knowing with confidence exactly which data items most increase business decision accuracy (and by how much) represents a paradigm shift in focusing big-ticket data transformation programs in advance of investment.
We have seen a wide variety of client approaches to machine learning adoption – from centralised capability-builds and project portfolio management, to more conscientiously experimental and organisationally distributed approaches. Most clients are focused on finding discrete (often third party) “killer” machine learning / AI solutions for automation, prediction, etc. There is value to be had here, but in our view there is still more through getting great at incorporating machine learning into the business-as-usual performance improvement toolkit.
If this is something you’d like to hear more about, we look forward to getting to know you in 2019!