Implementing advanced analytics projects, whether predictive or prescriptive, is hard. Numbers show that it is very easy to fall into the “proof of concept purgatory” for different reasons: inconsistent or fragmented data, lack of knowledge, lack of an end-to-end data management structure, strategic direction, etc.
Let us share with you some key learnings from projects in the advanced analytics domain.
1 – If management believes in it, it will happen. If management believes in the project, they will put the right team in place with the required dedication, they will involve the key stakeholders and stress the importance of the project.
2 – It is about people. Understand their problem. Prescriptive analytics tools do not replace people; they offer them a more powerful tool to do their job. For example, the person in charge of production planning will have a tool that will quickly propose plans so that the she will no longer have to “solve the sudoku” – the machine will do it. As a result she will have more time to test scenarios, prepare contingency plans, etc. When the user perceives this, a virtuous cycle begins: the user will become involved in the tool development, which will become more valuable, and the user will want to use it more, she will have new ideas, think about new features which will lead to new developments, etc.
3 – Start small, think big. Understanding the whole arc of a prescriptive analytical solution can go a long way in directing the effort. In the short term, it is appropriate to identify what may provide the quickest return, but without adding much to the cost you can develop the tool so that it can be extended more easily at a later stage.
4 – Test, test, test … and, after testing, test again. Failure during the testing phase is extremely cheap compared to failure in production. Each error of a tool in production affects the solution credibility. In many cases, it takes a long time of repeatedly reliable results for the solution to gain the trust of a user. It is easy to lose faith in it when it fails or when it offers inconsistent results.
The most serious mistake is that the solution makes unviable proposals once in production. If the user encounters this, she can hardly trust the following proposals. Failure is allowed during the development phase, it inevitably happens and when that is the case, it is time to humbly learn from the user’s knowledge, usually an excellent and experienced connoisseur of the system.
5 – The truth begins with the delivery. Accompany the client when she receives the application, make sure that it serves its purpose, that it responds to what she needs and identifies which particular cases can be better addressed: they are future developments that will make sense once the user trusts it.
If the user does not feel comfortable with the tool, the problem is with the tool. The user has been making decisions for a long time before the tool existed. If you don’t make her comfortable, the tool will have a very short life.
6 – Do not fall in love with the technology. Technology is at the service of the business, not the other way around. If the algorithm is sophisticated it is because it is more profitable. You do not want to publish a research paper, you want to offer the solution with the highest return in the shortest possible time. Strive to understand how the users can make better, quicker decisions so that she perceives the usefulness of the application.