Predictive Analytics – when to get started?

When should we perform maintenance on our equipment? How much stock should we purchase of this item? Will this project stay within budget? These are a few of the many questions which can be addressed with predictive analytics. However, the question of when to get started with predictive analytics might be difficult to answer, as it has many possible applications and it could be unknown territory for your organisation. The intention of this blog post is to give some support in answering this question.

Preconditions

In order to get started with predictive analytics, a couple of preconditions have to be met. Roughly, the following two preconditions apply:

1. The required data are available

The ‘required data’ means that sufficient observations are available and that the relevant variables are available as well. For example, if an ice cream truck business wishes to predict the ice cream sales for the next day, the ice cream sales of many days are required (in this case an observation is 1 day) and the temperature should also be recorded (the temperature is a relevant variable).

2. The organisation is ready

There should be willingness in the organisation to use the predictions as made by the predictive model. In case decision making is currently mostly done based on experience and gut feeling, this would be quite a change. Not everyone in the organisation has to fully embrace predictive analytics, but there should be some people who recognise its potential and are willing to approach a certain business problem with predictive analytics. Decisions do not have to be fully based on a prediction, but the prediction can be seen as an advice which is based on the available data. This way, the relations as extracted from the data are combined with the human experience of the decision maker. By combining the available resources (data + experience), decisions can optimally be substantiated.

Business case

If all preconditions are met, a business case can be created to determine whether predictive analytics would be a good investment. Based on some rough assumptions, the costs and benefits are estimated. Benefits mostly come from a higher efficiency and preventing wrong decisions. Costs mainly consist of the implementation project and software licenses.

Consider for example that 3% of the stock is thrown out each year because the items are expired or obsolete. If adopting predictive analytics results in a 50% more accurate estimate of demand, only 1,5% of the stock would have to be disregarded. In case the stock value is €10 million on average, this would mean savings of €150k for each year.

Pilot

Om te beginnen met predictive analytics hoeft niet meteen een langdurend project opgestart te worden. Het is aan te raden om te beginnen met een pilot. Dit heeft een aantal voordelen:

  • De kosten en baten van een implementatie kunnen nauwkeuriger ingeschat worden.
  • De randvoorwaarden kunnen beter getoetst worden.
  • Voor relatief lage kosten kan de organisatie kennismaken met predictive analytics.

Gedurende de pilot worden de benodigde data verzameld en wordt hier ten eerste wat data exploratie op gedaan. Hiermee wordt o.a. onderzocht of de verwachtte business logica terug te zien is in de data. Bijvoorbeeld dat je als ijstruck meer ijsjes verkoopt bij 25 graden dan bij 15 graden. Indien dit niet het geval is, is het van belang om eerst deze oorzaak te bepalen en dit op te lossen, voordat er daadwerkelijk met predictive analytics aan de slag wordt gegaan. Mogelijk zijn er niet genoeg observaties of zijn er ontbrekende variabelen.

Conclusie

Binnen vrijwel elke organisatie zijn toepassingen te bedenken van predictive analytics. Voordat hiermee aan de slag gegaan wordt, moet gecheckt worden of aan de randvoorwaarden wordt voldaan. Vervolgens wordt een business case gemaakt om te bepalen of er sprake is van een goede investering. Hierna kan van start worden gegaan met een pilot, waaruit mogelijk een implementatie volgt. Ben je benieuwd wat predictive analytics voor jouw organisatie kan betekenen? Neem dan vooral contact met ons op, zodat we hier (vrijblijvend) een gesprek over kunnen hebben.

Auteur Jelle Huisman, Founder & Data Scientist Cadran AnalyticsJelle Huisman

Jelle Huisman

Advice about our solutions?

Jelle Huisman, CEO at Cadran Analytics at Cadran, would be happy to talk about the possibilities for your organization.