I talk to lots of business leaders about Artificial Intelligence (AI) and Machine Learning (ML) and what their plans are for the future. The overriding perception is that AI & ML are for innovation or "new" things and that to even start thinking about a project they must hire a team of PhD Data Scientists, then spend the next 6 months thinking of use cases and labelling data, before training models across newly acquired data sets - and all of this before any value is realised. Happily, this doesn't need to be the case to get value from AI & ML.
I've seen white papers on the optimum team structures for a data science program that include the following roles:
I appreciate that in an average organisation, many of these roles will overlap; it is nonetheless daunting for anyone starting out and a small wonder that formal initiatives are planned "down the line". The hiring costs alone require a demonstrable ROI on any future project, let alone any software and infrastructure costs to make it happen.
Whilst this may be the case for larger Enterprises and there is no doubting the value that can be derived from these untapped mines of data, there is also a huge amount of value in pointing machine learning at things you are doing today. These can be things that you already know the value of and are aware how useful an increase in accuracy could be.
Typically this could be Sales Forecasting within most businesses. We can all appreciate the value in knowing where to focus our sales efforts or marketing budgets for best return, or which product performs best in which sector.
Most sales forecasts are built up with rules - select sales where X happened in Y geography. Tell me how long the average sales cycle is for this product. Who is the highest ranking sales person? Every Sales Director will then tell you that the data on its own won't help you and that it needs narrative. X performed best, but only because of Y.
This is where Machine Learning comes in. Feed in the sales data you are already reporting on (doesn't need Data Analyst or Data Engineer), give it some data with your results fields (labels) to train on (doesn't need, Statistician, Ethicist, Social Scientist, Researcher because you already know the definition of the "answers"), then run the model under your existing analytics team.
With this approach you will find out the relationship between all available inputs. For example, you may discover that Tenders responses where the company has never visited your website have a 30% reduced chance of winning; or, for Salesperson A, any opportunity that remains static for over 30 days has a 90% likelihood of being lost, where the value is less than £10k (this may prompt you to move smaller deals to another sales person). Without Machine Learning your answers would only be the net result of a specific question, as opposed to the net result of analysing all available data.
You have lots of options around models and processes that will give varying estimates in confidence - did you know that you can run certain ML capabilities as SQL statements on Google's Cloud Platform, for instance?
My point is that if you currently gain business benefit from analytics (supply chain maybe) then consider how much it would be worth to improve the accuracy or extend the scope. If you can answer that then you should consider an ML proof of concept to do 'better'.
If you are interested in a funded PoC around Sales or Supply Chain data or would like to discuss your path to becoming a data-driven business, please contact us and we'll come back to you with next steps.