Xomnia’s Analytics Translator Jelle Stienen participated as a panelist at the Shopware United AI in Commerce Event, answering some of the most pressing questions within the eCommerce community regarding the role of AI within this sector.
Watch the entire discussion on the video below, or scroll down to read a summary of Jelle’s answers:
1) When does intelligence become artificial intelligence?
Artificial intelligence involves automating a cognitively demanding task, whereby a human needs to evaluate a decision or a process. This encompasses quite a broad range of subfields of AI, like machine learning, deep learning and reinforcement learning. Those subfields collectively can help a computer more dynamically respond to users or events.
AI can help its users arrive at predictions based on and learned from historical data, which can be in the form of pictures, custom text (such as feedback forms) or structured data (such as spreadsheets). Using this historical data, the computer can learn patterns, and once an AI-based model is deployed in it, it can also make decisions by itself or, in most cases, give suggestions. Depending on the implementation, an AI solution can learn new patterns over time and improve decision making, which leads to artificial intelligence.
2) How can AI help in eCommerce?
For our eCommerce clients, the first step to making predictions and performing analytical tasks with AI is centralizing their data, which often comes from different platforms like Google Analytics or Shopify. Next, they need to add to this the data coming from more legacy data and systems, like financial or ERP-like systems. These two steps naturally also represent a challenge for many eCommerce companies that want to start using their data to become AI-driven.
eCommerce is a place where we find a lot of data, and what we often see is potential in personalizing the customer experience, which includes product rankings, product recommendations and others. Another way that AI can help in eCommerce is assisting retailers in understanding customers in a better way by, for instance, performing client segmentation for marketing or next-best action marketing. Other ways are more focused on the ‘backend’ processes, encompassing stock availability predictions, product demand forecasting, among others.
3) What are the factors to consider when identifying opportunities for using AI in eCommerce?
Applying AI should be a means to an end. As a rule of thumb, good AI meets the following 4 categories:
- It solves the problem at hand
- It can be put into production
- It is compliant with legal requirements and the GDPR
- It is reliable and trustworthy for the end user
Having said that, it is important to point out that AI shouldn’t be viewed as a solution to every problem. Most of the time, you can find easier solutions that are based on business-like rules. Some of our clients say that they want to implement a fancy deep learning model which can be interesting to develop but actually costs a lot of time and doesn’t really solve the problem. To know more about Xomnia's value proposition method for AI products, download our way of working whitepaper.
4) How can eCommerce merchants make sure that they are compliant with GDPR and privacy regulations when using AI?
We often encounter questions about GDPR, especially in the public domain where personal data is used. My advice is to involve your legal and privacy officers from the first stage of creating an AI-driven model or solution, and not at the end. Also, my advice is to be completely transparent with your stakeholders regarding what the model does, what are its outcomes, who are its end-users, and what are its regulatory instances. It’s important to add that sometimes, human involvement is necessary to evaluate certain decisions and to keep the ethical aspect in balance. Therefore, it might be a good call to never strive for full automation, but to always try to optimize the process nevertheless.
5) Does my data quality affect how my AI is trained or performs?
Yes! Garbage in, garbage out. When the input data is not of a decent quality, is biased or affected by seasonality, it affects the outcomes of the model. It also increases the development time greatly, causing data scientists to spend more time cleaning the data than actually modeling the data.
6) How does the ‘blackbox nature’ of AI affect retailers’ understanding of their customers and segments etc.?
We see a lot of willingness among retailers to optimize their processes, but regardless of the domain, it's hard to work with a solution that is not transparent and where the problem owner wasn’t involved in its making from the start.
Therefore, we think it's crucial to work with the end users from the start of the project, taking them along in the process of developing and implementing a model. This makes the process of AI transparent to them and helps in adopting the solution.
7) What do you expect to be the next big thing in AI five years down the road?
Ten years ago, the process of developing a website was very code heavy. Fast forward to today, I created a website in a few hours without having to do any coding. I see we are moving in the direction of auto machine learning, which we see all big cloud providers pushing nowadays. Therefore, mostly for somewhat simpler models, we envision that merchants will be able to ‘click an AI model’ within the coming five years.