DataIQ Talent Awards 217 - Data Science/Big Data Leader: Ian Smith, RAPP

DataIQ

DATA SCIENCE/BIG DATA LEADER

Ian Smith, principal analyst, RAPP

John Markham, RAPP, with James Morgan, Sainsbury'sWho is he?

Ian is a senior member of the insight team at RAPP where he supports the development of analytics for clients, as well as building the statistical and technical capabilities of the team. He has been the lead analyst for clients including Christie’s, Citroën, Guide Dogs for the Blind and Direct Holidays, as well as supporting the creation of analytics roadmaps for the agency’s other clients.

With a BSc in Maths, his career has included spells as a marketing analyst both client and agency side, before assuming his current role in late 2011. His responsibilities cover data discovery, targeting and segmentation, campaign analysis and dashboard design. His skills as a data scientist regularly see him using maths and stats techniques in new ways, as well as combining structured data from CRM systems with unstructured data from online sources and third-parties.

(John Markham of RAPP accepting on behalf of Ian Smith from James Morgan, Sainsbury's)

What does he do?

A breakthrough solution for a recommendation engine for fine art auctioneers Christie’s is evidence for Ian’s approach as a data scientist and big data leader. The cold-start problem confronts any system which needs to provide suggestions where there is no historical behavioural data on which to base them. For Christie’s, that ruled out the conventional approach based on other items viewed by previous visitors - the items coming up for auction are often new to market and there is very little data available.

Ian’s approach was to look instead at the detailed descriptions of the items being auctioned which are written by Christie’s teams of domain experts. He applied a bespoke string-searching algorithm to examine the text created and build metadata for every piece. Where Christie’s customers have indicated their specific interests, these were used to build weightings for each department. Machine learning was applied to ensure that these weightings adapt over time as those interests and customers’ tastes change. 

Based on 15 data sources, the technique ensures that the best matches to interests are presented to customers within Christie’s auction communications. New items which have not been seen before can be included because of the way the technique looks at the fit between the customer and the item.  

The recommendation engine is now on its third iteration as Ian looks to continue to tune it and apply more data to ensure the best possible fit. As an indication of its impact, the realised prices for items at auction has risen by 20 per cent and now regularly exceeds the specialists’ published estimates.

What did the judges say?

The judges were impressed by a novel approach to a well-known problem. They recognised the complexity of the problem, especially the absence of historical data, which made it tricky to resolve, but applauded the impressive results.