What actions can you take?
The first step is to assess your data effectiveness index. Many of the data collected are not used as they are incomplete. For eg, we see some of the following incompleteness in data in our experience,
1. Granular master data with attributes that matter (30-40% completeness)
2. Granular availability of stock information (while sales are largely available stock availability by day is not very common)
3. Attribute clarity in a machine consumable form (for eg alphanumeric numbers describing functional attribute like shape, last in shoes etc)
4. Granular information of consumer and store attributes
Create a data clinic to ensure all new product creations and transactions are clearly accounted and preserved from now on. Create a taxonomy that is relevant and important for your business. Involve predictive minds in this exercise as what you need for models to understand is different from what you need for the current way of working.
If you would like to move from descriptive to predictive analytics, time-series information of past data is critical. Cleaning of the past data is an absolute must. The returns would be exponential.
With data being 80% contributor to the outcome of any predictive modelling, always choose partners who can contribute to data than those who just come with models.
In fashion, data takes one more dimension unlike CPG companies i.e IMAGES. As we have always maintained fashion can’t be narrated and hence textual attributes alone are not good enough to plough the data. Every pixel in an image is valuable.
|