Can AI make a product hit?

When we hear people talking about AI all the time, they make it sound like a magic wand. A magic wand that can do anything from piloting spaceships to killing Voldemort.

But can AI truly bring value to the real world? Can it make you a product hit?

We will show you how you could create a winning AI product for your company today.

Artificial intelligence vs. Machine Learning vs. Deep Learning

We must stop referencing human intelligence to artificial intelligence.

We say this because we often see defining AI as some kind of Robocop-like intelligence though exhibited by computers. Let's make it clear - we don’t have that (yet).

Artificial intelligence is the general field of knowledge that covers machine learning and deep learning. Machine learning and Deep Learning are a specific subset of methods that everybody talks about because they show astounding results for multiple tasks.

The thing is that today’s predictive algorithms are as intelligent as we train them to be. There are some problems that predictive algorithms are remarkable at doing, some others that simply can’t. This video from Carl Betzler from Amazon will help you understand why.

So, before you build a prediction algorithm inside your organization, you must consider the problem you want to solve and the available data and knowledge resources.

We often see a trash-in, garbage-out scenario where data is unavailable or there is a knowledge gap to use that data for the custom purpose your company needs. Remember that your model will underperform if you train an ML model on inaccurate or irrelevant data.

Is it AI?

Forbes reported in 2019 that nearly 40% of European companies classified as AI startups don’t use AI in any material way for their business. Looking at AI's promotion today, I wouldn't be surprised if the number doubled

The problem is that they use the label as a marketing maneuver and not to provide value. They believe labeling their product as an AI solution gives them the credibility needed.

We often see companies doing AI for no reason. There are a lot of cases where simple heuristics could do the work much cleaner, faster and better.

This might be shocking to a lot of you, but before building anything, it is always important to know whether it makes sense to use AI or not. We work with our partners to define the best strategy and timing to implement AI across multiple use cases.

Big tech's secrets for implementing AI

When you think about AI, you might think about big tech companies.

These companies define their user journey, plan their team’s efforts, and launch their ML product while the user's pain point is still real. The key here is to consider data scarcity and the expensive expertise needed to build it out and deliver the features with a timely launch.

In this video, Ria Sankar, founding member of the Microsoft AI for Good Research Lab, will help you understand the process of building AI products at big tech firms like Amazon and Microsoft.

We divide the entire process of building a product using ML into the following steps:

1. Define the pain and arrive at a Minimin Viable Product (MVP): The product leader must choose the right target based on the problems. Bringing experts to outline the user journey and its problems is highly recommended. Then craft a Minimal Viable Product (MVP) by choosing the minimum expected outcomes to cover the investment and provide a return to the organization.
2. Gather, clean transforms, and prepare your data: You must first identify the data source and choose the correct column and fields relevant to your target. Explore the data, find potential problems within the structure and define the opportunities and challenges you will have. The next steps include cleaning and pre-processing the data and preparing it for training and testing.
3. Train, test, and select your algorithms: In this step, you will search for multiple models and methodologies until you find the right candidates. Exploratory analytics — both univariate and bivariate — should be carried out to summarise the model’s main characteristics. The next steps include feature engineering and feature selection to extract features from the data. Finally, model selection can be made using ensembling methods like neural networks. Once you have a list of algorithms, you will compare their performance metrics (accuracy, sensitivity, F1-score, etc.), speed, biases, and opportunities and choose your model. Here models will interpret the data and output the result.
4. Integrate your project with your workflow and make it sustainable over time: This step involves bringing all the pieces of a puzzle together to solve the problem. In a nutshell, you will need to involve your software team to choose the right framework (using flask, Django, node), build a user interface (using HTML, CSS, and Bootstrap), and integrate it with the backend framework chosen. The final steps entail testing it with the end user and validating the workflow.
5. Deployment and monitoring for possible data drifting: The final step is to deploy and monitor the model in a sustainable and self-sufficient way (that allows the model to learn over time). It can be deployed as a web app on a cloud or on-premise, but remember to find ways that allow your model to learn over time. Additionally, placing a monitoring mechanism is important because it will allow you to know when your model starts to face data drifting or other problems. Think about control methods in the event of disasters.

How can your AI product make a hit?

Now that you know what it takes to build an AI product, we will give you the hacks that we use to make it a product hit. We follow this framework to do it:

1. Identify the opportunities

As we mentioned before, the first step is to understand the current state of your company and the bottlenecks you have. Lay down each bottleneck and connect them to potential prediction scenarios that could radically impact your business (it should be 5x or better). We recommend segmenting those opportunities by company areas. For example, if you are a pharmaceutical company, you might be interested in segmenting opportunities across the following units:

• Operational excellence: Improve operational performance by creating tailored algorithms that use current and new data to enable improved decision-making powered by AI / ML.
• Provider experience: Facilitate current or future services and accelerate the impact on programs, initiatives, healthcare professionals, patients, and other stakeholders with deep transparency.
• Patient experience: Enable a holistic and seamless approach to shorten the diagnostic and treatment journey and provide appropriate and personalized patient care.

2. Quantify the value you expect to bring with the opportunity

This is one of the most important aspects of your product planning because it will allow you to understand whether this is a project you should do now, later, or not. These are the things you must define:

• The main objective of the product. For example: Using ML to predict which patients from a cohort of 300,000 patients will not comply with therapy and increase compliance from 3.42% to 15.25% in 12 months.
• Quantify the added value for all stakeholders. For example, Patient: Increase patient outcomes to reduce mortality rates with personalized strategies in the next 24 months.Provider: Outperform last year's high-cost oncology benchmarks to increase the number of payers sending me cohorts of patients in the next 12 months.Payer: Reduce overall expenditure of oncology patients in the country's northern region.You (pharma): Increase therapy outcomes, increase therapy compliance, and create better goodwill with all stakeholders in the next 16 months.

The value should always be calculated numerically and within a specific time frame. Some common metrics for the main objective are cost reduction in dollars, the added speed in hours, and the operational capacity you bring several services.

Finally, depending on your chosen value, you must calculate how much money you leave on the table for not achieving that goal today. This, in turn, lets you know the minimum outcomes you have to meet to pay for the AI/ML investment.

3. Identify the resources you have and the ones you don’t

Ask yourself where you stand today and what is missing to get the opportunity done.

Resource checklist:

• Step 1 - Data check: Do you have historical data to show a model of what you want to predict in the future? If not, can you access some partner’s data or publicly available data to get your MVP done?
• Step 2 - Distribution: Do you have a distribution partner that will allow you to take this to real life? If not, can you find one while you develop your AI project?
• Step 3 - Domain expertise: Do you have the domain expertise to guide the models' predictions and judge whether the results are correct? If not, will you be able to bring a consultant or a partner to help you understand the bottlenecks and strategy?
• Step 4- ML expertise: Do you have the AI/ML expertise to do this? If not, what are the tools you know or the teams you will need to hire to make your idea a reality?

We often see projects fail because they lack one of these resources. You must be strategic in choosing the project based on the results across the framework and prioritize to allow one project to fund the others with the value it brings.

Whenever you're ready, there are three other ways we can help you:

1. Work with us 1:1: to bring the power of AI to your business.

2. The Ark Library: Take a look at the free resources we have to bring AI to your day-to-day.

3. Join our community for short and sweet tips: If you are the type of person who finds it helpful to receive short, daily tips, then follow us on LinkedIn. We post a few short tips each day.

‍