How we used machine learning to sell more flowers

What should you do when you have a warehouse full of flower bulbs that you want to sell, but aren’t sure exactly where to spend your marketing budget to get the biggest and quickest return? That was the challenge our customer Sarah Raven was facing. This is how we used machine learning to solve it and clear the stock.

What challenges did Sarah Raven have?

Sarah Raven sells a range of seeds, bulbs, plants, gardening and kitchen kit online. The website wasn’t generating as much revenue as it should have been, which was largely due to a lack of traffic. The site has a lot of pages, but with restricted budget, the challenge was knowing exactly where to spend the money in order to get the greatest return.

The Sarah Raven website needed SEO help

What did we recommend?

We decided to identify the pages that offered the greatest potential for conversions. Importantly, we weren’t trying to find pages with the highest conversion rate, rather the pages more likely to convert.

This task required a large volume of complex data to be analysed, which was likely to take a great deal of time that we (and our customer) could ill afford. We used machine learning analysis to overcome this challenge.

Machine learning helped identify the best pages

We used machine learning to calculate the likelihood of a conversion from any page, having taken into consideration its performance. The machine learning model had to predict with a 70% or greater degree of accuracy, the likelihood that a page would convert based on the correlation of large numbers of conversion indicative metrics, including:

  • Sessions
  • Average session duration
  • Page views
  • Bounce rate

We used four different algorithms, with the intention of achieving the highest level of accuracy. They were:

  1. Naive Bayes
  2. Random Forest
  3. Logistic Regression (with and without balanced classification)
  4. Logistic Regressions with K-Fold Cross Validation

All four models identified the dahlias category page as having a close to 77% accuracy level, much better than the minimum accuracy level that we set as a benchmark. This was a very positive result. However, we wanted to test how the page might perform by improving behavioural metrics. We manipulated the data and increased the click-through rate by 5%, which resulted in an accuracy level of 85%.

Time for dahlias to bloom

Now we knew where to focus our efforts, we reviewed the existing metadata and optimised the information by including key differentiators that would encourage a higher click-through rate, especially for terms for which the page had a good ranking position.

The new metadata helped us to differentiate the page from competitors and deliver a new message to the searchers. The aim was to show that this was a transactional page and make it stand out in the search results.

What happened?

Results of the machine learning and SEO for the Sarah Raven website

What the customer said

“We appointed Vertical Leap in January 2017 and they immediately had a big impact on our business. They’re very data focused which meant the decisions they made for us generated a good return right from the offset. We very quickly saw significant increases in traffic levels, organic transactions and organic revenue, and this trend looks set to continue. I would have absolutely no hesitation in recommending them to others.” Ashley Pallet, Ecommerce Executive, Sarah Raven

Read more about the clever stuff we’re doing

Using predictive analytics to increase sales – without increasing your PPC budget

Expert Interview: Henry Carless on automating PPC analysis

Expert Interview: Lee Wilson on the power of automation in SEO

Save

Michelle Hill profile picture
Michelle Hill

Michelle joined Vertical Leap in 2011 as Marketing Manager, having spent the previous 15 years of her marketing career in the recruitment, leisure and printing industries. Her passions include dogs, yoga, walking, cycling, the beach, mountains and tapas.

More articles by Michelle
Related articles
Abstract image showing the different journeys e-commerce buyers take when navigating the internet

How to build a buyer intent model with keyword classification

By George Karapalidis
Trolley

eCommerce marketing trends in 2019

By Michelle Hill
rockets flying up into the sky

Data science in retail: 4 use cases to help you increase sales

By Dave Colgate