Connect with us


AI-driven precision can revolutionize jewellery ordering, says Dr Asim Tewari of IIT Bombay 




Dr Asim Tewari of IIT Bombay says finely-tuned machine learning models can completely transform the ordering process of jewellery

Artificial Intelligence (AI) systems are now virtually all around us, the latest and most famous of them being the Large Language Model (LLM) ChatGPT. With AI being so widely discussed, it is easy to wonder what can these techniques do to help improve traditional businesses. One such application is that in the jewellery retail segment.

Of the several applications that may be possible in everything from inventory management to customer facial recognition and tagging which address problems of businesses in general, more interesting and directly impactful are the ones which affect the ordering process of jewellery.


Consider the peculiar case of a jeweller pondering over the quantity of jewellery to be procured for the next quarter. The first intuition one would have is to think what sold well last year will also sell well this year. This hypothesis does not take into account several things like ever-evolving design sense, customer segment related shifts, temporal, i.e., time-related, movements in the calendar, etc.


While add-ons on such as a model to account for changes in most things is a working solution that most jewellers seem to have adopted, such a model is clunky and not amenable to change, looks at just one year of sales data and essentially is dependent heavily on the person designing it.

Finely tuned machine learning models have the ability to completely transform this landscape. A machine, unlike a human, can keep track of millions and millions of parameters whereas a human can at best remember and use a dozen.

“At IIT Bombay, the problem of jewellery demand forecasting was converted into a craftily constructed classification problem”

Dr Asim Tewari
Professor, Center for Machine Intelligence and Data Science (C-MInDS), Department of Mechanical Engineering, IIT Bombay


However, this process is not so simple. The demand forecasting problem in jewellery retail segments, especially in India, is differentiated from that of other luxury goods by its dependence on a complex combination of requirements and factors, including wedding seasons, religious beliefs, festivals and macro-economic market forces. Jewellery sales can be looked at as a periodic function with the time period of a year. However, unlike other sectors, a one-year period cannot be divided into financial quarters which are consistent in their timing. In the jewellery sector, buying cycles differ from year to year due to religiously-motivated time periods for jewellery buying, auspicious occasions, festival timings, etc. Therefore, conventional and established time series machine learning models are not directly applicable to jewellery demand patterns. A solution to such a problem was developed at IIT Bombay, wherein the problem of jewellery demand forecasting was converted into a craftily constructed classification problem. The machine learning model was built with the understanding that outliers are actually extremely significant days in the data. Thus, the problem becomes one to identify and predict such days rather than ignore them, thus turning it into a classification problem, wherein the task is identifying how a particular day is, rather than accurate prediction of sales.


The case cited here is just one avenue of application of AI in the jewellery retail business. Several applications exist, originating from using objective metrics to identify how dissimilar images are from one another, i.e., distinguishing designs and correlating them to sales, to come up with a grand model which identifies currently running jewellery designs, suggests new ones as well as informs which ones are going to be the best-selling ones in future.

The Retail Jeweller India magazine

Continue Reading

Latest News