How to double your eCommerce overnight: Use AI to increase sales and conversions

Introducing Matrix Commerce Coefficient™, a custom-designed machine learning (AI) solution for maximising the profitable choices in data-driven product personalisation for online shop visitors.

Read for details and a case study showing 2.17x growth in sales, and 3.2x increase in profitability achieved ‘overnight’, without a need for any other changes to the shopping platform, channels, or product inventory.

In early 2022 e-commerce space is as hot as it gets. The last two years saw the biggest growth in history (due to lockdowns and societal changes caused by the COVID pandemic), following in the footsteps of a fantastic prior decade of a steady march towards dominance and destruction brought to traditional retail. The opportunities abound, but competition is also heating up, with the space becoming increasingly crowded. The barriers to entry are relatively low, with cheap and easy platforms such as Shopify, BigCommerce and WooCommerce (a WordPress ecommerce plugin), and digital advertising platforms available to all. 

Established and (increasingly abundant) new players are competing for the wallets of the same (albeit growing) group of customers. Everyone is using the same best practices to attract, convert and retain a limited footfall of visitors, who are choosing online over traditional retail. Due to growing competitive pressures, and relentless technical progress and innovation, winning in this game, or even retaining previously covered ground, is not a given.

The need for better methods

While everyone is sharing the same best practices and growth hacking ‘secrets’ (not really a secret if it’s known and used by everyone else) there remain a number of new approaches that are relatively inexpensive and easy to implement, while sitting at the very core of the maths determining the profitability of online commerce. Unlike the overused and popular methods, whose effectiveness is at best neutral now that everyone shares them and must adhere to just keep up, those new techniques are less obvious, because they require more specialist knowledge and work to adopt. But unlike the low hanging fruit that has already been picked by everyone, they offer true and significant competitive advantage. To make it even better, they do not require any significant changes or disruption to the processes and structure of ecommerce operations, making them relatively easy to use. 

If this is true, how come these are not more widely used already? The answer is that they are, but only by the strongest and most forward-looking players, the likes of Amazons and eBays of this world. What is stopping the others? It is hard to know for sure, but our bet is the lack of skills, resources, knowledge and patience. Moreover, many ecommerce operators believe that they are doing this already, as the key premises are not new, and not secret. The secret is the implementation, and the depth and effectiveness of the methods that cannot be achieved with simple and common off-the-shelf solutions used by most. To achieve true success, a more involved approach is needed, utilising the latest and most effective algorithms based on Machine Learning (ML, a sub-area of Artificial Intelligence), trained on your specific and fresh data coming from multiple sources, but the results are significantly better. Before we talk about the technology, let’s look at the math and the process making them so profitable, and illustrate this with some real-life examples. 

To achieve true success, a more involved approach is needed, utilising the latest and most effective algorithms based on Machine Learning (ML, a sub-area of Artificial Intelligence), trained on your specific and fresh data coming from multiple sources.

The math behind profitable sales

The first step to successful commerce (online or traditional alike) is bringing a visitor in. For ecommerce shops several digital methods are available, ranging from search engine optimization (SEO) and advertising (SEM, PPC), through social media, influencer marketing, affiliate programmes, product aggregators, banners, marketplace channels, email and text marketing, apps, and so on. One thing that these have in common is that they are expensive and highly competitive and make it important to maximise the chance that your ad or banner is clicked with a genuine buying intent, and that you achieve the maximum value from those hard-won visitors once they land on your page.

As you’ve paid for the visit, unless you can make a sufficiently profitable (on average) sale, the cost may outweigh the profit (even if you take LTV into account). This is where you can maximise the profitability of your overall operations by either improving your conversion rate, or by increasing the average order value. Some methods of doing this are relatively well known, and include great quality product information, UX of the product page, ease of navigation and search, but also recommendations of other products that are worthwhile considering, because they’ve been purchased by customers sharing similar characteristics, go well together, or are potential alternatives. 

The important goal is to avoid the customer bouncing back to the search engine, product comparator or other channel that brought her to you. This is achieved either by serving a well selected product that maximises the chance of transaction at the stage of the channel banner / ad, or offering her a personalised, well-tailored selection of satisfactory products once she enters your shop. By using all the information available about the customer, your products, channel and context of the visit to select the products with the highest likelihood of conversion, you can increase the statistical rate of success, and the expected / average size of each order.

Advantages of data-driven AI engines

Most shops are designed with this in mind and do this to an extent allowed by simple algorithms driving off-the-shelf plugins, or simple database queries. Although those methods are quite simplistic and far from optimal, many of the shop managers believe that they are doing a decent job at this, and their results are as good as possible based on industry averages. Little do they know what is possible with advanced and truly intelligent personalisation and product recommendation engines, driven by custom-designed and trained machine learning (AI) models. 

The secret is in the richness, depth, accuracy, completeness and timeliness of the data used to maximise the relevance of the recommendations shown to a particular customer, which cannot be optimised using the usual simplistic methods. A machine-learning approach is able to profitably use multiple sources of data and is able to make instant decisions based on dynamically changing circumstances, optimising and maximising outputs of every single visit and interaction. This leads to fractional improvements at each step, which add up to a very sizable and consistently repeatable increase in the overall results.

The secret is in the richness, depth, accuracy, completeness and timeliness of the data used to maximise the relevance of the recommendations shown to a particular customer, which cannot be optimised using the usual simplistic methods.

Because the advanced AI-driven methods differ only in the extent (or depth), rather than the general functional principle, most online shops can deploy them with minimal disruption, risk and difficulty. The user interface / front-end is usually ready and requires minimal (if any) changes. Equally, no extensive modifications are needed in the back end, as the machine learning models are run independently in their own loosely coupled microservices architecture containers. Once correctly designed, trained and implemented, the results driven by those custom AI models often exceed expectations, and are a huge positive surprise, making a vital change to the overall profitability of ecommerce operations.

Enter the dragon: the Matrix Commerce Coefficient™

To give a specific example, we have recently completed deployment of our custom Machine Learning / Artificial Intelligence product recommendation engine, called Matrix Commerce Coefficient™, for a major fashion marketplace (product aggregator) platform Avanti24. The client was adamant that they had industry leading results and they were (this may even be an understatement) sceptical that these could be improved any further. They agreed to the project, but with low expectations. 

By training our dedicated AI architecture with combined internal and external customer interaction data taken from the client’s platform, affiliated shops and other channels, to generate real-time automated and highly optimal decisions (inference) driving selection of products displayed for each customer, we were able to significantly improve the overall profitability of the client’s operations. 

The engine was able to efficiently predict which products had the highest probability of getting purchased by each individual visitor. This was an advanced and exceptionally effective exercise in personalisation. It isn’t generally easy to achieve, and this is where our custom Matrix Commerce Coefficient™ engine, which we perfected over many years of R&D and projects developed in partnership with the leading Machine Learning / AI experts and largest e-commerce operators in the UK and Poland, shows its real strength.

Product data retention and attribution

One of the difficulties was, as is usual with large product catalogues, that products rotate, and that the previous interaction data (views, clicks, purchases) was only available for a relatively small subset of the catalogue. This meant that for most of the product offering it had to be extrapolated from similar products, for which the data was available. This meant the need to understand and create a graph of product relationships, similarities and classifications, and use this information for new products. 

The upside was twofold: even new or previously not tested products – which did not get the chance to be shown to visitors yet, and so for which it was otherwise impossible to estimate attractiveness to different categories of customers – could be efficiently shown to the visitors with the maximum purchase intent, and that the data gathered for old products, even when those were removed from the catalogue, was profitably used rather than lost forever. This meant that the knowledge base of data available for product and customer interactions continued to grow and the efficacy of the algorithm kept improving over time.

How it works

As you will appreciate, we cannot disclose full details of our proprietary technology. Broadly speaking, Matrix Commerce Coefficient™ engine is based on a sophisticated mix of several machine learning and AI algorithms and architectures, including Reinforcement Learning, Contextual Multi-Armed Bandit, Random Forests, Gradient Boosted Decision Trees, Epsilon-Greedy, and Bayesian algorithms, implemented in Python. 

In the case of this implementation, it used the following sources of data to infer the optimal product selection, maximising the click-through rates from product ads displayed in a partner site:

  • Product characteristics and history, with relation to other products based on the product knowledge and relationships graph
  • Customer interaction history with this, similar and other products on the shop and on other channels and partner platforms (views, shares, adds to basket, saves, purchases)
  • Context of the visit – the path that led the customer to the product grid page
  • Customer characteristics from cookies and external sources, such as Google Analytics data

In this case the back-end and a store of product information was a PIM solution Pimcore, but Matrix Commerce Coefficient™ can be as easily implemented for Magento, Shopify, WooCommerce, BigCommerce, Prestashop, or any other eCommerce platform.

Case study: Fashion marketplace

Results achieved by Avanti24 – the largest fashion marketplace in Poland, operating within one of Europe’s largest media groups – were nothing short of stunning. Some of the ratios previously believed to be industry-leading improved by 2.17x on average, or even 3x for selected product categories. The click through rate (CTR) has increased from 0.6% to 1.3% (on average) or even 1.8% (for previously underserved product categories).

Some of the ratios previously believed to be industry-leading improved by 2.17x on average, or even 3x for selected product categories. The click through rate (CTR) has increased from 0.6% to 1.3% (on average) or even 1.8% (for previously underserved product categories).

This translated into a dramatic and roughly corresponding improvement of the overall sales and an even greater profitability (as the cost of ads remained the same, with ROI increasing) of the platform ecommerce operations.  

The results were so good that based on the recommendation received from the client we were approached by one of the largest DIY and building materials retailers from the Kingfisher Group, Castorama, to apply Matrix Commerce Coefficient™ in their online shop. In this interesting case study of a multichannel merchant, who operates a chain of building materials and DIY hypermarkets, an additional data stream is provided from all daily in-store transactions, enriching the data available for product histories, associations and interactions.

We are also in the process of preparing a similar project for the UK’s first building materials marketplace, Go Banana.

Summary: the case for starting now

Improving profitability of eCommerce operations with methods that are less obvious and less widely used at present is a great way of achieving a strong competitive advantage. And, of course, it is simply great to be more profitable. 

But the story also has another side. As is the case with every new strategy, the benefits for early adopters are outsized, and then diminish over time as the same methods become widely adopted by everyone else. But in a competitive situation, the initial advantage can make a massive and lasting difference, as winners keep winning and losers cannot ever make for the initial loss of ground. 

New and more effective methods distort the balance in the entire profitability calculation, which has a large impact on the competitors. More profit from every visitor (higher ROI) means a higher marketing and advertising budget, and ability to bid more for the same clicks, which are now more likely to result in a larger than before average sale. 

It will not be very long before this strategy ceases to be a way to get ahead of the game, and the time will come when – as with every new profitable strategy – it will become a necessity, required to simply keep up and stay in the game. The time to make the best use of Matrix Commerce Coefficient™ is now, and it will handsomely reward the early adopters.

The good news is that these are still early days, and it is not too late to be among the early winners who will reap most of the rewards.

If this sounds like something that you would like to explore, do get in touch with us, and we will be happy to help you start to maximise your results with our proven leading-edge AI technology.

About the author:

Dr Rafal Bergman is the founder and CEO of IQDEV, and a CTO in a number of eCommerce technology start ups, including the UK’s first building materials marketplace Go Banana. With over 22 years of hands-on experience starting as a software engineer building some of Europe’s largest eCommerce sites in the golden dotcom era (before the crash of 2001), with his later original patented software and big data / AI products used in over 30 countries, with multiple National-level deployments. Recipient of the King of Bahrain ‘National Project of the Year’ Award. Rafal is passionate about helping businesses develop a strong and lasting competitive advantage with the use of software technology.


Subscribe To Our Newsletter

Send a Quick Enquiry