They say data is power, but they are wrong. Analytics is the power, data is the source.
Most organisations are collecting and analysing data on product performance, customer trends, supply chain effectiveness and much more. To a lesser, but rapidly increasing extent predictive techniques such as machine learning applied to customer trends are prevalent. Using optimisation techniques for decision making is the current differentiator that can put your organisation ahead of the competition. These techniques give successful organisations the confidence that they are making better decisions and getting the most out of what they put in. Having data without taking full advantage of it in this way is like having your cake and not eating it!
Broadly speaking, there are three different stages of data analytics:
- Descriptive: looking at “What happened and why did it happen?”
- Predictive: exploring “What is likely to happen in the future?”
- Prescriptive: making suggestions on “What should we do about it?”
This list is essentially a hierarchy: you can’t predict what you can’t describe, and when decisions rely on future uncertainties, sensible suggestions require sensible predictions!
Descriptive analytics are everywhere and give you a sense of where you are and where you have been.
Predictive analytics are getting a lot of attention currently and allow you to see where you are likely to go.
Prescriptive analytics are currently under-represented and use your data to actively help you make better decisions.
Therefore, prescriptive analytics provide the greatest opportunity to add real value and competitive edge. This can be achieved by adopting advanced optimisation techniques: aligning resources in a way to get the most efficient outcome.
Prescriptive analytics is not new.
The fundamental algorithms have been around since 1947. In the 1970s and 1980s there was hype about optimisation and the thought was it could solve all kinds of problems. However, the technology was not ready, and the uptake was therefore limited. Optimisation solvers have come on in leaps and bounds over the last decade. Combining this with the increase in computing performance we are now able to solve problems at a rate that could not have been dreamt of 30 years ago. The unfulfilled promises in the start 1990s can now be realised because the technology is ready.
A delicious example
Imagine your company is a large-scale producer of cakes. Being able to describe your past sales and productivity effectively through descriptive analytics is vital but relatively commonplace nowadays and, in many cases, not thoroughly enlightening. Stick everything into Excel or Tableau, look at trends to determine peak cake demand, compare with what happened last year, cash the profit, maybe make high level predictions about the coming year and then… move on, right? Instead, let’s decide to do more with this information!
How about being able to forecast demand for your cakes using predictive analytics. There may be more subtle peaks and troughs in when people want cakes that are affected by less predictable features than what was observed in Excel. A spike in demand in September reminding us that there is a concentration of birthdays in this month or, less predictably, it could be that bad weather and bad news increase the demand for cakes. If so, we may be in the golden age of the cake industry. Predictive methods are widely available and easily accessible as a result of factors such as rapid advances made in machine learning techniques. Access alone, is not enough though. Machine learning can be applied to anything but in many cases may not be appropriate or worthwhile, just because you can it doesn’t mean you should. Only by having a good understanding of your business problem and knowledge of a variety of predictive techniques, can maximise your predictive power by matching this information to the most relevant model and algorithm?
Once you have accurate forecasts of demand for your cakes, the right resources can be aligned to meet that demand with the least effort, but this requires business planning which is supported by prescriptive analytics. You want to match this demand with supply as best as possible, optimally in fact – so that your customer is having the freshest possible cakes and you are generating the greatest possible profit. This will involve many factors with varying costs including but not limited to:
- ordering ingredients;
- financial benefits for greater mass production;
- logistics; and
- meeting sales programmes for retail partners.
Forecasting demand is about understanding external factors that impact your product. Planning requires an overview of the internal factors of the organisation that are involved in creating the product or serving the demand and is often a more complex task than understanding what the future demand is.
Consider the below demand and sales programme, as an example.
There may be serious repercussions, financial, reputational or both, for not meeting the sales programme, but overcompensating and producing too many cakes (as if…) could lead to waste. Time is also a factor; cakes don’t just appear at the supermarkets. The obvious choice is to make as much as the predicted demand. This is only a toy example, but you can easily imagine reasons not to do that. For one, November has a conflict, is it more sensible to overstock and satisfy the sales programme to keep the supermarkets happy or to incur a loss due to waste? Are there fluctuations in the price of ingredients that make it worthwhile to stock up during low seasons? Do you have the storage capabilities to consider such an option? What is the shelf life of your cakes and at what temperature does quality start degrading? How can you factor all this information (and more!) into the decision-making process?
Optimisation is a key part of prescriptive analytics. By codifying these questions and quantities into a well-designed cost function and set of constraints you can answer all these questions simultaneously by using optimisation techniques. By using methods such as mixed integer linear programming, optimised schedules can be produced to maximise profit not just revenue, balance supply and demand while reducing cost or to address other KPIs that may be important to your business.
Now that you have optimised the production and distribution of your cakes you can also apply similar techniques to other aspects of your operations. Instead of a one-size-fits-all approach to weekly staff schedules you can give your employees the option to include preference and build an optimisation model to determine staff schedules to match employees’ wishes and the business requirements of cake production and distribution. Similar techniques are used by Network Rail to improve rostering processes for their signalling and maintenance workers. Or, consider organising the fleet of lorries designated to deliver your cakes to the suppliers daily, on orders that are coming in as a live stream. The GPS in the lorry is already working its optimisation magic but the next level is optimising the fleet so that the fewest lorries are going to the most delivery locations in the least space of time. This kind of solution is already famously being used by Amazon, UPS and Tesco just to name a few.
Data analytics can take a company that extra mile ahead of everyone else, but only by thinking bigger than off-the-shelf. Optimisation techniques are the final frontier of data analytics and, as such, requires the commitment of time and resources to take full advantage of them. It requires active participation. Embrace it by developing the capability, by driving the necessary change management and by integrating it into real systems and deployment.
Those who master this can dominate their market. Being first with an innovative solution for implementing optimisation techniques in your business will set you ahead of the competition.