Applied AI For Products

Alice Tang
May 21 · 5 min read

Credits: Thanks to our sponsor Sift, the Advancing Women in Product Team, Alice Tang, Kavitha Gowda, Yina Shi, AWIP volunteers and our panelists.

To maximize the opportunity AI and machine learning represents, product leaders must ask the right questions, understand the process of developing models, and encourage their organization to foster continuous improvement.

On May 7th 2019, Advancing Women in Product and Sift Science in Seattle co-hosted “Applied AI in Products” to a full house, where our panelists shared their best practices for integrating AI into a product.

Panelists

Our panel featured Shahar Ronen (Product Manager at Sift), Palak Kadakia (VP of Product at UiPath), and Quaseer Mujawar (Senior Product Manager of Alexa ML Platform at Amazon), all product leaders who had to incorporate AI in some way into their products they were building. The event was moderated by AWIP Programming Lead for Seattle and Senior Product Manager at Microsoft Azure, Kavitha Gowda. Our panelists’ varying experience, industry, and product portfolio made for a well rounded perspective on applying AI to products.

Shahar leads Sift’s Account Defense product where he focuses on stopping account takeover and preventing fake signups, which enables their customers to grow their customer revenue without risk. He uses AI to run risk assessment models that detect and stop security breaches.

Our next panelist, Palak, drives product vision and strategy for robotic process automation (RPA), a form of software automation that builds business processes and performs user action on behalf of them. In this context, AI is crucial for recognizing the kind of business processes they can automate with RPAs and the right way to build the desired automation.

Our final panelist, Quaseer, is part of Amazon Alexa’s Automotive team, whose goal is to make every car a ‘smart’ car. AI is an important part of this: an example use case can be to have AI suggest a gas station for the driver to fill up based on mileage, gas prices, and commute time based on distance.

Success metrics for AI features

Before embarking on any new product feature, defining success for the feature you are looking to implement is critical.

For Shahar, it’s important to reduce fraud and risk in real-time proactively versus reactively (waiting until after a fraud had occurred). Any delays could expose the customer is a big amount of risk. Compared to a business automation product such as UIPath, Palak defines these metrics with the goal of increasing productivity for the company. For example, how often did the business process fail? This helps flag areas for improvement so a feature can be added to help increase productivity. And lastly, success metrics can also be viewed from the lens of increasing customer retention. By using AI, he can predict customers who are contemplating cancelling their service and proactively address their concerns before the customer drops the service.

Data is King

Data is king, but even in this digital era, collecting accurate data is hard.

For Sift, the accuracy of the model depends heavily on the data that is fed into these AI models.

Shahar says “When we had an accuracy issue with our models, it was almost always a data problem.” In response,his team works extensively with customers to ensure the accuracy of the data in order to reduce AI prediction issues.

Palak explains how collecting data for an enterprise B2B platform is hard: some enterprises can be very protective of their data whereas others are ready to share data for productivity gain. She works with these customers to get data in a controlled environment.

In some cases, you don’t have to collect data from each customer. The data collected from only a few customers could suffice as a sample size that can then be extrapolated towards all customers.

Data compliance

With growing concerns over data and privacy, compliance and security have become even more important topics than ever before.

Shahar mentioned he spends a lot of time explaining Sift’s data usage policy. Boundaries are set for the data Sift receives in order to be compliant with regulations.

For Alexa, Quaseer shares his motto, “Be fair, flexible and functional while collecting data.”

  • Fair — Don’t collect data if it does not offer value to your customers. There’s a lot of liability and risk by collecting sensitive data.
  • Flexible — Some customers will consent to sharing their data while others prefer to keep it private. PMs need to work with all customers and offer the option to to opt out of sharing data.
  • Functional — Educating customers on how the data will be used puts customers at ease that their information won’t be used wrongly.

The main takeaway here is that, even though data can be paramount to obtaining findings or validating hypotheses, PMs need to be cognizant to obtain consent when collecting data and also be proactive in educating customers on how the data will be used.

What skills are needed as a PM to implement AI solutions?

Palak says, “AI is more than models”. PMs who work in the AI space are to emphasize building an experience that incorporates AI features (e.g., a ‘smart’ car that can understand the user’s commands), understand market trends (e.g., where is the market moving in terms of implementing AI into products), be aware of technical limitations (e.g., what can AI do and what can it not do), and above all — think on behalf of the customer when integrating AI into the product’s feature set.

Shahar says “There is no magic around AI.” If the customer problem is well-understood, any PM can build an MVP solution to address this pain point. It is more important to understand how to build teams to execute on AI, create processes around AI, and motivate teams to build products using AI than it is to understand the fundamentals of machine learning.

Quaseer suggests finding the right use case, right away. If there is an issue, think about how AI can solve it. Sometimes, a complicated model isn’t the best answer — but rather a simple rule-based regression model. Be sure to fit the model to your use case, and not the other way around.

At the end of the day, in order to incorporate AI into your product, you don’t have to be an expert in building models. What you will need is: passion towards the domain, customer empathy along with data they are willing to share, and knowledge around AI capabilities to start building products leveraging AI.

The full panel discussion is available on AWIP’s YouTube Channel.

How to stay updated with future events:

Sign up! // Follow us on Facebook, LinkedIn and Twitter

Stay connected with AWIP — Seattle

Advancing Women in Product (AWIP)

AWIP was established to help foster and develop the next generation of women product leaders. To achieve this goal, we focus and excel on providing quality programming, for our members, that are focused on developing key skills to be a successful PM and a leader.

Alice Tang

Written by

Advancing Women in Product (AWIP)

AWIP was established to help foster and develop the next generation of women product leaders. To achieve this goal, we focus and excel on providing quality programming, for our members, that are focused on developing key skills to be a successful PM and a leader.