Kamelia Aryafar explains why keeping data science, ML, and AI front and center is essential to the long-term success of e-commerce companies in every vertical.

Avenue Code: Tell us about your personal career path. What inspired you to pursue a career in computer science and machine learning, and how did you get to where you are today? Was there anything surprising that you learned about yourself and your abilities along the way?

Kamelia Aryafar: I think I have been incredibly lucky to have had support along the way in different steps of my journey in addition to the passion for computer science. I have been drawn to computers ever since I programmed Hangman on my computer at the age of six. I never considered pursuing anything else - I was obsessed with exploring the capabilities of computers, which naturally led to AI. AI is one of the most impressive inventions in human history, and the possibilities are endless.

What I’ve learned about myself is also what I tell other people who are pursuing careers in ML/AI - patience is required. It can be so frustrating to spend hours hunting for a tiny bug that creates a massive problem, but that’s what it takes to optimize constantly. 

AC: As a Board Member of Persian Women in Tech and the author of several compelling articles, you’re a well-respected advocate for encouraging women to pursue STEM careers, especially at the executive level. What do organizations gain when they embrace a balanced leadership model?

KA: Women add a much needed perspective because they view products from a different angle. When women aren’t involved, product design misses crucial considerations. For example, there was a time when the automotive industry was only testing product safety features for the male body.  

In STEM fields, women excel. In fact, the first generation of computer scientists were women, but now there are more males than females in STEM. The ratio should be more balanced at all levels, including at the executive level and in board rooms. But we are making progress. The more we talk about and normalize women and under-represented groups in STEM, the better it will get. 

I should note that having a balanced leadership model is valuable outside STEM as well.

AC: Why is it vital for enterprise organizations to add a CAO role to their executive team?

KA: Every company is trying to extract value from their data and use it to personalize product recommendations to drive sales. Because it’s relatively new, there’s an executive-level leadership position for every essential business area except for AI and ML. Several companies are beginning to add this role, and that’s what closes the gap between excitement over the potential of these technologies and implementations that lead to actionable insights.

AC: What are the biggest challenges of walking into a company where this role isn’t already established?

KA: Expectation management. AI and ML are tools for a long-term game and don’t often generate immediate results. It takes time and patience to go through all iterations. Change management is also crucial, as implementing this role necessitates a departure from historic business strategies. 

AC: Organizationally, can ML and data analytics teams operate relatively independently from other departments, or does there need to be a wider, cross-organizational cultural shift in order to utilize these teams to their fullest potential?

KA: Eventually, the paradigm needs to be highly cross-functional and collaborative. When I talk to companies at the infancy stage of establishing an AI or data team, it makes sense to centralize it because you want to establish the pipeline, infrastructure, and organizational alignment on prioritization. But down the line, AI needs to be heavily matrixed into the organization so that its value can be added into every product team.

AC: How does the scale of data change an organization’s ability to utilize it effectively? 

KA: Sparse data equals less accurate models. Having diverse, large scale data is better because it lets you create more generalized models, but there are other challenges related to storage, cost, data utilization, and infrastructure. Algorithms that were built for smaller-scale data are not applicable anymore, so it’s really important to take advantage of data by thinking through architecture and system design from the outset. 

A lot of companies are struggling right now because they have siloed, fragmented data in legacy systems. They have to create a data lake before they can create a meaningful machine learning model. It’s relevant to note that we are generating data exponentially, and most of the data in the world has been created in the last few years. 

AC: What are the biggest opportunities in harnessing the power of data for recommender systems and personalization? 

KA: There’s a big emphasis, especially in e-commerce, on creating personalized recommender systems that present products aligned with each consumer’s style and preferences. These machine learning algorithms are based on demographics, purchase history, searches, etc. and

are designed to make the world smaller for us so that we can quickly find relevant products. In brick and mortar stores, we find what we need and walk out. But in e-commerce, no one has the time or patience to sort through millions of products to find what they’re looking for. The potential for this technology is massive and extremely helpful for buyers and sellers alike.

That being said, there are two things we need to pay attention to as an ML community. The first is privacy. We need to consider the ethics of AI and what kind of data we’re extracting for our personalization systems. This leads to explainable AI. A lot of companies are focused on using AI models for personalization that explain to the consumer what data was used to generate the ad. For instance, many ads now include tags that say, “Why am I seeing this ad?” This is something we need to pay more attention to.

The second consideration is bias and lack of diversity. For recommender systems, it’s easy to create models without enough diversity. There’s a danger of confining consumers in an information filter where they see more and more of the same product and nothing else. It’s challenging to maintain personalization while creating enough exposure to new products and growing along with the consumer. 

There are different ways to grow with users. For example, if you’re shopping for mid-century modern furniture, we don’t want to box you into this product profile exclusively, so we inject new styles into recommendations while maintaining relevancy to your interests. 

AC: What is the future of ML and AI in e-commerce?

KA: E-commerce is one of the biggest consumers of AI. As we shift increasingly toward online shopping, AI will also increase in importance. Personalization is essential for helping consumers find what they want and convert quickly. AI and ML are also key for marketing spend optimization, as well as helping e-commerce platforms identify and customize offerings for a clear target audience.  

Historically, a lot of companies have used CRM to define static clusters of consumers and segments to pursue. But with ML and AI, you can personalize to particular consumers and create a one-on-one relationship as opposed to a cluster-based relationship. 

AC: What do you think is the key to success in terms of strategic partnerships?

KA: The successful partnerships I’ve been a part of have all had an aligned vision, goal, and roadmap very early in the process. The unsuccessful partnerships I’ve seen are those that are trying to explore a vision together. So the first question I ask when I go to review meetings with new technology vendors is “what is the goal?”

AC: What do you do to stay abreast of innovations in tools and technologies?

KA: I read as many research papers as I can and volunteer to serve on several review boards for conferences. It’s a fantastic way to stay current on ever-evolving technology, because to review someone’s work, you have to understand it deeply. I also give back to the community by publishing my own work. When you’re writing a research paper, whitepaper, or blog post, you generally start with a survey to understand the current state of the world before adding your own contribution.

I also connect with other researchers so that I can hear about their work and ideas, both by attending conferences and joining social media platforms, which are great at building these communities for AI researchers! 

AC: Thanks, Kamelia, for sharing your expertise on the potential of AI and ML in e-commerce, as well as your inspiring vision for women in IT leadership roles!