In a recent InternetRetailing webinar, How to acquire unexpected new customers through AI-powered, data-driven prospecting, we heard from Dr Neal Richter, CTO at Rakuten Marketing. Here’s a bulletpoint overview of his presentation.
- Neal opened the webinar with an introduction to Rakuten Marketing.
AI algorithms
- In product selection: how AI is used to show the most relevant and profitable items
- In advertising: through a predictive model that says what groups of people are likely to respond to what items.
- Putting together relevance and profitability rankings
- The difference between targeting shoppers at different points of their journey.
The machine-learning process: the different steps
- How machine-learning works to analyse behaviour
- How that is applied to advertising
- Creating a predictive model
- How that applies to advertising. “Almost everyone at this point is using some sort of retargeting”: catches people who are already interested in a product.
- Prospecting is about finding new shoppers earlier – and how that works
- Data collection/storage – feature engineering – seed creation – internal lookalike – activation through DSP – report and feedback.
The internet audience
- Made up of bots (don’t have family to buy for); people who are browsing not buying; and buyers. “If you can score the identifiers and score the behaviour you can cut out the amount of advertising spend.”
Machine learning for prospecting
- Use scoring and filtering algorithms to separate into users that didn’t respond, users that responded to an ad, and users that clicked, browsed and bought. “Once the campaign is live we’re not only refining the ads that are shown but in some cases going back and refreshing the seed.”
Audience strategies
- Prepared segments – canned or seed audiences.
- Direct activation: bring your own seed.
- Smart seed: build a seed for me.
Showing ads to that audience
- Train ad engine on past-purchase behaviour and pre-purchase shopping behaviour.
- Show an ad: different sets of consumer-learning algorithms.
- Ways to drive sales of particular items once you’ve found an audience that has clicked, can then optimise what happens to the basket.
Case study
- How many new customers did I get: if the value is high, then showed ads to new customers. Case study achieved a return 51% higher than the benchmark.
- AI-powered prospecting: case study 20% higher visit rate; 229% higher conversion rate; 51% higher new customer rate.
Data-driven retailer strategies
- Contrasts intuitive ideas about what groups of shoppers want to buy with their actual behaviour.
- Case studies: beachwear and backpack companies.
- Found: beachwear buyers more likely to use a Windows laptop. Converted at a higher rate depending when they had previously bought beauty products.
- Backpacks: a repeat purchase: whether they’d bought backpacks previously in the past indicated likelihood of buying a new one. Highest propensity to buy among mobile users.
- “It’s hard to do in your head, it’s hard to do on a spreadsheet and you can’t do it on a piece of paper. What machine learning algorithm does is create a dense model with a probability and ROI to each, then apply that at scale.”
- Based on personas rather than specific user.
Rakuten Institute of Technology
- The Institute underpins a lot of the AI work that Rakuten Marketing does.
- Internal team in Japan and the US.
- Enables and supports retailers to take algorithms and to experiment with them.
Final points
- Shelf space: Brands compete for physical space and premium shelf space, but can create AI models for shelf space.
- Unified approach to marketing: attribution, search, affiliate, display: helps to drive marketing spend.
The webinar was followed by a Q&A session. To see the webinar, slides and the Q&A session in full, visit the Rakuten Marketing webinar page here.
