This presentation was delivered by Abhishek Ratna, Head of AI & ML Developer Marketing & Olivia Burgess, Product Marketing Leader, Google Workplace Security at Google, at the Product Marketing Summit, Seattle 2022. Catch up with a variety of talks with our sister community, PMA's, OnDemand service.

Abhishek Ratna: Unless you've been living under a rock, you probably know what artificial intelligence (AI) is, so I won't belabor the point. We’re living in very exciting times, and Olivia and I have been lucky enough to have front-row seats to see how AI is being used in enterprises.

We’ve also had the privilege of working with product marketing, product, and engineering teams as they bring machine learning products to market.

We’d like to share some of our insights into why AI matters, why you should care, and how you should think about bringing AI-based products to market.

Olivia Burgess: If you're thinking, “but I market dog food, or shampoo,” don’t worry. We're also going to share some general principles that you can apply to your marketing and that will help you articulate value back to your broader business. No matter what type of product you market, this article will be helpful.

A new world of possibilities

Abhishek Ratna: Lost Tapes of the 27 Club is an album developed by a UK-based organization that focuses on mental health issues amongst the musician community.

They wanted to raise awareness of these issues, so they partnered with a marketing agency that trained an AI on thousands of hours of tracks by musicians who lost their lives to mental health issues at the age of 27 – people like Jimi Hendrix, Kurt Cobain, Amy Winehouse, and Jim Morrison.

That AI essentially spun out a brand new album that was picked up by 50-plus publications in 190 countries, including Rolling Stone magazine. At the Drum Awards, the album won the Grand Prix for its innovative use of technology, as well as the Ad Tech for Good prize. More importantly, it helped a lot of musicians around the world find resources for mental wellness.

A title that says "a new world of possibilities" then an image of the lost tapes of the 27 club, and then a grid with 15 images of avocado chairs.

What stands out to me, besides the feel-good factor, is the generative power of AI. It can create original music using just a thought. Other AI technologies can do the same with images. If you type in “avocado chair” as a prompt, you can get a dozen renderings in a few seconds.

You can tell language-based models to create poetry in the style of William Shakespeare, and they’ll automatically do that for you. is another great example; it uses advances in natural language processing to understand human conversation in real-time and take action on that.

There aren’t only tremendous creative possibilities for AI; there are also significant business implications. If your product doesn’t already incorporate AI in some way, shape, or form, you might be in trouble.

AI has become table stakes. Most products today innovate and compete on the basis of their AI superpowers. AI has gone from being a nice cherry on top to the next competitive differentiator. Boards around the world are talking about how to better incorporate AI.

As Deloitte puts it, we’re in the age of “with”. We’ve moved from the time when machines were used for carrying out our day-to-day tasks to being in a space where our capability is augmented with smart machines.

Olivia Burgess: There’s been a kind of hockey stick of innovation in how AI and machine learning have developed over time. Where before there were just a handful of niche solutions for specific utilizations, today we’re seeing more and more enterprise-ready AI tools. That means more organizations feel ready to use AI in their general operations.

Title which reads Nascent innovation to enterprise-ready tools - then two graphs, one titled Hype cycle for Artificial Intelligence, 2021 by Gartner, and the other a pyramid also by Gartner titled Popular use cases with obvious value.

There are also more companies – key players like Microsoft, Databricks, Google, and Amazon – bringing products to market that can be used very easily by data scientists and machine learning engineers within their organizations.

How to approach AI projects incrementally

Abhishek Ratna: Before we dive into the specifics of using AI, there are some principles we’d like to share. We’ve found these helpful as we’ve rolled out AI projects.

Rule number one is don't be afraid to not use AI. It's treated as this magic cure-all, but if you can get good results without it, I’d recommend taking that route.

Rule number two is to embrace data and measurement. Marketers who understand their customer data, segmentation data, and metrics very well are the ones who typically drive the most value, and best articulate the value of their projects.

Rule number three is to grow incrementally. As someone who's worked on deploying AI-based marketing projects at Facebook and Databricks, I can share that the big bang approach rarely works well. It's better to start small. Start by automating one task or doing some sort of bespoke analysis of your customer interactions. That will start giving you some value.

If you want to do something as powerful as generating product recommendations or building real-time personalization, it's wise to take a successive path and graduate to that level of maturity.

The user-first domino effect

Olivia Burgess: To differentiate the AI or ML product you're bringing to market, there are two key personas you need to speak to, and there needs to be a symbiosis between both.

The first key persona is the end user. That might be the data scientist who’s building the algorithm. It could be the machine learning engineer, who's responsible for implementing that algorithm into the business and showing value from the predictions it makes.

Or, it may be a developer who wants to implement AI and machine learning in a solution they're building. In my experience, these are the folks that you want to spend a lot of time thinking about.

We usually reach out to this audience with things like white papers, technical demos, analyst reports, and customer references. Another great way to find them is through grassroots communities, places like Kaggle, the contest platform where data scientists can show their prowess in building accurate models, or Stack Overflow, where people go for community-based support.

The user-first influence domino - two people, the first is a "Machine Learning Practitioner (Data scientist/ML Engineer) - underneath it says "Grassroots communities, free trials/demos, ML contests, technical how-tos, whitepapers" - then the next person is a CXO/Head of ML/Director of R&D and underneath it says "Analyst placement, total cost of ownership/total economic impact, customer references, and propensity modeling."

However, you reach out, engaging at the user level and understanding their needs deeply will be critical to getting you towards your second key persona: the head of R&D or the CXO who's responsible for buying.

Of course, you need to take the needs of the head of R&D and the CXO into account too. They want to make sure that your product is going to provide a high return on their investment without disrupting their organization too badly. They want to be able to use your solution’s predictions meaningfully within their organization and keep the train on the tracks.

Achieving differentiation in an AI future

Abhishek Ratna: What's interesting about AI is that it's a fundamentally different way of thinking about products and how they compete. Up to now, we've been building products with great features, touting their ease of use, and talking about how capable they are in terms of the feature set.

In an AI-enabled future, that might change slightly. We feel there are three or four common traits or characteristics upon which AI-based products of the future may compete, and these traits will inform our GTM work.

The first of those traits is intelligence. We’re used to smart devices that are connected, and transmit and sync data beautifully. The next level of innovation is creating intelligent applications, which are not order-takers but decision-makers.

These intelligent applications can understand context, sentiment, and intent, and they can facilitate our decision-making work with their intelligence capabilities. This intelligence is sure to evolve as time goes by.

The AI of the future will also be adaptive and able to respond to new challenges in real-time. This is already happening with self-driving cars and industrial robots, for example, where meta-reinforcement learning allows robots to respond to previously unseen situations.

Four icons - one with a head labelled intelligent, one with a clock labelled adaptive, one with a person branching off labelled ambient, and the last is hands shaking labelled ethical.

Up-and-coming AI will also be ambient. Right now, AI is generally deployed through large servers, hosted in a data center or the cloud, then accessed through a phone or a browser. That's changing very quickly.

There’s going to be a diffusion of technologies, thanks to which AI’s computing power will percolate through buildings, smart devices, and microprocessors, so a technology’s ability to be embedded in your environment could become a competitive factor among AI-based products of the future.

Sadly, it's not all smooth sailing. We're already seeing a lot of challenges with deploying AI in real-world situations, and there are genuine concerns around privacy, trust, and ethics.

The companies and products that will become successful with AI in the future will be those that put users’ trust, safety, and well-being at the forefront. Responsibility and ethics will be competitive features for many of these products.

Olivia Burgess: Prediction engines should prevent bias and promote fairness. I think everyone recognizes that moral imperative – we don't want to become the Terminator. However, that’s not the only reason to make sure you’re building responsible and ethical AI solutions.

In a study we ran in collaboration with The Economist, we found that most executives believe that ethical AI will help them get a competitive edge and be able to use fair predictions in more meaningful ways. There’s a huge amount of business value to be gained from making sure that you have tangible principle-based operations that promote fairness within your models.

Data-based differentiation

Olivia Burgess: The evolution of AI and machine learning is opening up a world of opportunities to implement intelligent solutions. But how can you differentiate your product?

How do you show the power of your platform versus another? It all comes down to data and showing users that you can give them what they need. It will become table stakes to prove to users that they can deploy more models and experiment faster thanks to the solutions we provide.

As an example, we at Google recognized that data scientists wanted to be able to build more models with fewer lines of code; those who are not super technical wanted to run models more easily. We launched Vertex AI last May with this in mind. Our benchmarking and data were all focused on being able to build a model with 80% fewer lines of code.

The title "Data-based differentiation" with screenshots of 2 articles - the first titled "Google Cloud unveils Vertex AI, one platform, every ML tool you need." and "Appliced ML summit".

That one data point popped up again and again in the press. All of our subsequent blogs and content have included customer references or internal benchmarking statistics about how quickly and easily you can use this technology, and how it’s helped organizations to cut costs or mitigate mundane tasks through automation.

When you want to differentiate an AI product in a crowded space, where so many standard capabilities are table stakes, it’s vital to first determine the tasks your users want to do and identify their pain points.

Then you can provide data points that show how your product can solve those problems. This approach makes AI’s possibilities real for the customer and the end user.