Data and analytics

Most analytical work starts with a less glamorous question: which numbers are right? Figures from a till system, an online shop and an accounting package will rarely agree, and usually for mundane reasons. A sale is recorded on a different date in each. Refunds land in a different month. A product code changed two years ago and one report was never updated.

We find those differences, fix them or write them down, and build a single set of tables with clear definitions that refreshes as new data arrives. Everything that follows depends on it.

A forecast or a model is only as good as the assumptions underneath it. We state those assumptions plainly, test the ones the answer is most sensitive to, and say how much confidence a result deserves. Where the data can’t support a conclusion, we say so.

Forecasting, regression, classification, segmentation and the analysis of experiments all have their place. The technique is the last thing we choose. We start from the decision the work is meant to inform and use the simplest method that answers it reliably, because a model people understand is a model people use.

A one-off analysis should be written up so that someone else can follow the reasoning and repeat it. A question that comes up every week deserves a report that updates itself, built in the tools you already use where possible, with every figure traceable to the query that produced it.

We only access the data a piece of work needs and keep it in your own systems where we can. Personal data is handled in line with Isle of Man data protection law, and UK and EU law where they apply.

To discuss a project, email enquiries@quantuity.com.