The last decade was spent building products that were cloud native and mobile native, and this disrupted several industries and changed the way we live. Mobile made it possible for entrepreneurs to use camera and location to build products such as Instagram or Uber. As we look to the future, it is clear that something very exciting lies beyond cloud-native and mobile-native, and that is AI-native. When you’re building “AI-first”, you’re taking AI as the starting point of the design process. It’s no longer about adding cool AI-powered features, it’s about removing pre-AI legacy features and creating an entirely new, AI-centric product experience.
There are several factors that are making the movement to AI Native possible:
This also means that entrepreneurs and product managers need to rethink both products and the way products are built to make them AI-native. Some of the features and trends that AI native products will encompass are:
AI-first products are closer to living things. Thanks to machine learning, for the first time ever computers are adapting to humans rather than the other way around. Users put much less effort into interacting with them, resulting in a core product brief that is very simple: “it should just work”.
One way this will happen is in the way we interact with machines. PCs had a disjointed experience with separate keyboard, mouse etc. Mobile made the experience more direct and “human like” with touch and the ability to for example pinch and zoom. The push to interact naturally with computers will only get stronger. There is already a shift to “conversational” user experiences as opposed to interactions using screens. Systems will have to leverage the power of AI to enable those conversations with users. We are already seeing this with speech and possibly very soon we will also see gestures.
We are also witnessing AI being used to make user interface more efficient. Android is doing this really well. The previous version of Android (Oreo) tapped into machine learning to make proactive decisions, like sort the apps that you use frequently. The new version of Android goes one stage further. App Actions uses machine learning to suggest apps to users based on their habits. It can automatically anticipate your next action based on the current action and show you a shortcut to take you straight to the section of the particular app that deals with it. For instance, if you book a cab to work on Uber every day at a particular time of day, Android P will start suggesting that action.
These ofcourse remove friction and make the product easier to use.
Product fail all the time, and they will fail all the time in the future as well. Product managers and entrepreneurs will need to ensure their AI products gracefully handle low performance scenarios or failures. One way is to add better rail guards and add much more intelligent cues to set user expectations and allow products to fail more gracefully.
Product creators should also use this opportunity to provide a method for users to immediately relabel data to further improve the model. For example, when a user clicks the thumbs down button the chatbot should respond by not only apologizing but also asking for user feedback. This feedback could easily prevent a similar failure from occurring again in the future.
Besides with AI, prediction of product failure will become easier. AI algorithms will also be able to manage uncertainty, which means that the algorithms know if they’re facing a part of the problem space they’re inexperienced in or not good at solving — they can then farm out to a human or revert to a set response. Ultimately this will give the user a far more satisfying and genuine experience.
As we develop systems that operate dynamically, we’ll also need to rethink Q&A. We will need models for real-time error detection so that we can fail gracefully or have the system jump into another path of action. One way this could be done is by getting feedback from an independent application that constantly observes the main system and looks for abnormal or inaccurate behavior. Once it detects this type of behavior, it would give feedback to the main system so it can adjust its actions.
The best AI products make the user more capable. And the very best AI completes a task start to finish for the user; the user should have to do as little work as possible to make it happen. These are truly autonomous intelligent agents.
For example AI-native products will understand their users better, and help them onboard faster. This is what recommender systems do. They allow users to get started very quickly. These products will shorten the user journey to the wow moment and help beginner users learn advanced skills without much effort.
Another example are chatbots, such as 1–800-FLOWERS Facebook Messenger chatbot which picks up conversational cues to help customers order flowers without the customers necessarily going to their website and sorting through several products to find the right one.
A truly great application powered by AI will deliver that wow moment, the magical feeling of accomplishing something that wasn’t quite possible before, right out of the box. Take Google Photos, which can retrieve images matching a search query. Or SwiftKey, which predicts the words you’ll type next.
We all do some A/B testing and measure some key things. But this is not baked into the system grounds up. The reason this is important in an AI based tool is because the system gets better the more training it has had and the more feedback loop it has.
Once you put AI into a product you should see an ongoing improvement of that product through the learning that the system can now generate. This leads to more people using the product which means the data that it has at its disposal to optimise itself increases, and you are hopefully set up with a self fulfilling cycle of improvement that leads to exponential success. The feedback system will allow products to learn and get smarter.
This would also mean that product managers and entrepreneurs educate the user that their model will improve over time. The more users invest into it, the better predictions or recommendations the product can provide.
When building AI products, you will often build a powerful product that will do 90% of a task well enough, but then you’re left with this remaining 10%. The question is, what happens with this remaining 10%? If you decide you want to build your machine learning system to handle it, you typically face diminishing returns. It can typically take a month to get 90% accuracy, a year to get 95% accuracy and a decade to get 99% accuracy.
If you’re trying to ship a machine learning product, you really want to ship one where there’s a good tolerance for occasionally getting things wrong.
For example, Google has these smart replies for Gmail. They unintrusively provide suggested replies at the bottom of your email. If one of the replies isn’t very good, it doesn’t matter. If one of the replies is good, the user clicks on it and it saves some time. That’s a really nice way to deploy a machine learning product. Rather then responding on the user’s behalf, it simply suggests options.
A successful machine learning products picks its battles carefully. It’s about choosing to ship something that has a high tolerance for occasional errors, baked into the nature of the product. Even if you want to ship something that does something on the user’s behalf, getting manual approval is a sound approach.
AI products can differ significantly from traditional products. In traditional products, success is usually measured through delivery of a ‘deterministic’ product that always delights customers — a hardware product has the same behavior under the standard conditions, the same user actions in a software product results in the same expected response.
An AI-driven product, however, may not always have a deterministic behavior and may in fact produce counter-intuitive results — a personalized recommender system may produce different results to a user action after learning additional preferences. The matrix for success will look different in an AI-based product as probability and statistical terms may be introduced.
Currently when we design products, we think about screens and flow of the app. Increasingly we will need to factor in what goals we are trying to achieve. For example, in a news app, the amount of time a user spent may not be a good metrics, especially when the bulk of the time was spent in searching for interesting content. The app itself should be able to suggest the relevant content for the user.
Turning raw data into things that patterns can be derived from is the process of creating features. A feature is simply an individual measurable property of a phenomenon being observed. The generation and selection of these features is a combination of art, science, experimentation and learning. Domain expertise is required for this initial generation of features. If you know a lot about the data you’re looking at then your starting point for features will likely be stronger than those without any knowledge of what the data might show.
Besides as technology gets horizontal, products get more vertical. A lot of tech will be open source, algorithms themselves are domain independent and work across different problem areas. That means that proprietary code is no longer a defensible asset when it’s in the path of the mainstream AI train. So for companies, the differentiation will come by the fact whether they are choosing the right problem to solve. And this will come with domain expertise. In this wave of AI, companies will find very vertical problems, where they really own a narrow vertical.
Just like mobile’s location feature led to creation of Ubers of the world, AI will also bring in fundamental capabilities that will allow entrepreneurs and product managers to create new features and products.
For example, AI systems do classifications really well. This could be used in the way users are storing things or searching things.
Retail companies are using AI-driven visual search to help shoppers to promptly find their desired fashions. Pinterest recently launched a visual search tool called Lens, which uses machine vision to detect items on the web or in the Pinterest library and suggest related items.
Another core capability is prediction. We are seeing this already in marketing automation where being able to predict the success of an email campaign or marketing initiative can help companies continuously improve marketing efforts (in display, text, video, or even direct mail).
Another example is authentication. Artificial Intelligence will allow entrepreneurs to incorporate much more frictionless ways to authenticate users. Face detection, fingerprint and voice-enabled services will become a common way for individuals and businesses to access information and data.
When thinking about building products for AI-native, entrepreneurs and product managers have to think about 3 different things. The essential ones are:
They have to make these three things work together and in such a way that the user who is using the product finds it really useful and actually falls in love with the product.
Whether you are an industry giant or an emerging startup, now is the time to embrace AI-native and begin defining your future. The next generation of trillion dollar products and services will be built for AI-natives and emerge even faster than the last generation of billion dollar ones that rode on mobile and cloud.