Microsoft’s Vijay Mital Explains How CEOs Can Guide Their Organizations Through Radical Product Innovation with AI

What your company can achieve from GenAI is only limited by its ambition. With the right effort, AI can change how companies create new products that can transform business models and markets—and even society. Here are our learnings from such journeys.

As a CEO, you galvanize your organization around a shared vision of progress. But with AI changing the opportunity landscape beyond recognition, you are faced with some big questions. When it comes to GenAI, most CEOs recognize by now that this technology is powerful, and organizations are already using it in multiple ways across their functions. So what can the CEO prioritize and uniquely champion, fully capitalizing on these emerging capabilities?

As CEOs look to put their weight behind a few big bets that were previously unimaginable, the area receiving particular attention should be one in which they already have ideas and major ambition, but were held back by the difficulty of implementing with their existing base of people, assets and relationships. What earlier might have looked like moonshots are now coming within execution reach by a combination of GenAI advances and the creativity and culture change leadership from the highest levels.

In this discussion, Vijay Mital, Corporate Vice President & Chief Advisor for AI Transformation at Microsoft Research, shares his learnings from a decades-long career in software, and over 20 years at Microsoft, including the last five years partnering with senior leaders at companies on envisioning and applying GenAI to effect radical product innovation.

If you want to get going on this journey—and want to make some timely leaps for your company – this post may offer you a blueprint for doing just that.

What is radical product innovation?

It is best to start with a couple of synthesized examples.

Take a cosmetics company that seeks to move beyond products that make people look good, to products with verifiable wellness claims. That requires understanding not just chemistry but biochemistry, and interpreting a whole spectrum of scientific literature, lab results and raw consumer signals.

This is product innovation. It becomes radical product innovation if the cosmetics company can deliver this more complex, new product with its current research personnel, its existing data assets and at its current speed, i.e., the speed of consumer trends and tastes.

Take a reinsurance company that seeks to expand its portfolio of commercial operations with heavy exposure to environmental risk. Take one such operation, a sprawling dairy facility.

The underwriter needs to understand the type of equipment being used, and the site in which it is situated. This means interpreting satellite imagery as well as subterranean signals like ground penetrating radar. Then correlating the signals with reports of previous spills around the world and the containment efforts that were needed to prevent aquifers or nearby farming and exurban populations from being affected.

This new offering becomes a radical product innovation if the reinsurer can act on this complex information across varying types of environmental risks in diverse geolocations, upleveling and amplifying existing people and skill sets.

The product ideas don’t have to be new. In fact, they are often long-held but suppressed aspirations. What is new is the potential with GenAI to actually deliver these products at reasonable cost and at scale from an existing base of people and assets.

Once a product innovation is this radical, it opens the opportunity to change the business model, and ecosystem and market dynamics.

Cosmetics with strong wellness claims can attract new types of therapists, beauticians and fashion influencers. The ability to differentiate environmental risks can allow the insurer to become more of a partner as a company decides to locate and configure new dairy operations.

I have learnt that CEOs and senior leadership are strong at identifying such opportunities in their markets. What holds them back sometimes is the difficulty of delivering on these opportunities with their current expertise, asset base and market position. GenAI can change this.

What are your key learnings?

My learnings here started in 2019, when it first became clear to many companies that large language models (then called Foundation Models) were on a path to interpret the kinds of complex information that researchers and product innovators have to reason over, such as literature, charts and free flowing observations.

This allowed us to shift the use of AI from narrowly defined applications to broad acceleration and transformation of complex, high-stakes processes—essentially playing three roles:

• A Reasoning Engine: Read all types of information and signals and connect the dots.

• A Teaching Engine: Learn the methodologies and techniques that worked in the past, and now transfer them and adapt them for current tasks within current resource constraints.

• A Process Evolution Engine: Let people break artificial discipline and role boundaries that were enshrined in traditional application UIs. Using a conversational interface, a product formulator can look into upstream research results that were discarded and use that to optimize downstream tasks.

This gave senior leaders a simple lens on GenAI. Given this, I saw the following loose pattern succeed. For simplicity, I have formulated it as a step-by-step guide, but of course not all steps were always needed or executed in this sequence.

1. Locate your highest-reward areas 
Every company has an understanding of their value chain and what improvements bring returns. For example, in pharma, if a 10% R&D cost reduction means a 1x multiple on return, then a 10% reduction in risk of experiments and trials succeeding has a 24x return, while bringing peak sales forward has a 36x return.

Given the uncertainties and difficulties in product innovation, not to mention the effort to get GenAI right and adopted by the most creative experts, it’s natural that leaders opt to target processes with the highest multiple. With this high reward potential, it becomes easier to define interim delivery and milestones where even a partial success is valuable.

2. Create early measures that help both the business and data scientists
Identify metrics you can measure even early in the product innovation journey. The best metrics are those that are valuable both to a business owner and to data scientists.

One example of such a metric is information coverage. How many sources of information did the decision maker take into account? Did they learn from previous focus groups, or social postings by consumers, or experiments in adjacent product lines? Often, high information coverage in critical decisions is a proxy for risk reduction. This is a business metric: reduction of risk of late failures, thus acceleration of getting product to market. It is also a metric that data scientists can tune the AI on.

With GenAI, for the first time, we have systems that can continually improve – provided they are told what being better means!

3. Bring in multiple business owners from day one
Traditionally, software projects like focusing on one defined problem or process at a time. Easy to manage, easy to measure. Instead, perhaps counterintuitively, what we saw succeed is a “foundational” approach: work on multiple product innovations at the same time, let the AI core learn from different types of information and different types of users. That created a robust base with sufficient flexibility to let people discover new ways of working and stretch artificial boundaries.

4. Empower champion users to share their reasoning and create trails for others
It took a long while for enterprises to become data-driven. Multiple years to move people from relying on hunches or private knowledge to using tools like spreadsheets, data lakes and business intelligence effectively. With GenAI, we are asking people to move from data-driven to learning-driven: how did others solve an analogous problem, what was different, and how to adapt that learning for my own situation.

Even with the best implementation of GenAI, there is a learning curve—not a UI learning curve, because everyone can talk to a bot, but a reasoning learning curve. How to pose questions, how to break them down into pieces that can be cross-checked by a colleague, how to bring existing tools like evaluators and predictors into the reasoning. The key is the observation that a small percentage of people always figure this out really fast—and what accelerates everyone else is if these champions can easily and safely share their reasoning chains, their methodology as templates for others.

5. Learn to start from a low-data environment and create a robust data estate
Radical product innovation often means that a company is going into an area where they do not have as much data as they have for traditional product lines or service lines. The reinsurer above does not already have a large dataset on environmental inspections, claims, damage mitigation and dispute resolutions. The cosmetics company can’t build on a body of existing focus group reports and influencer videos from people who have already tried products with both beauty and wellness claims.

What we saw succeed was a two-pronged approach. In one prong, get value where few-shot learning works, i.e, give Large Language Models or Large Multimodal Models (both called LLMs) a small number of carefully chosen and commented examples and let the LLMs use their pre-training and world knowledge to generalize. Interestingly, recent advances in LLMs, like GPT o1’s delivery of PhD level reasoning, will make this generalization even more powerful.

In the second, identify those types of data gaps that can be filled with data synthesis. Sometimes, LLMs themselves can add missing attributes or generate new hypothetical scenarios and case studies. At other times, and this technique is still emerging, simulators and emulators can create variations of existing data or generate entirely new records.

The important part was a very deliberate exercise of creating a data estate, with an understanding of what gaps LLMs can fill and what data can be synthesized by other computational methods, before resorting to manually gathering or licensing new data.

What is the CEO’s role in ensuring innovation?

GenAI is a very significant technology breakthrough. But it is not the first technology that companies have used to radically transform their products, or change their ecosystems and markets, or even benefit society.

Let’s go back to the dawn of the internet. At that time, perhaps the most radical product innovation I saw was in heavy industry, from generators to recycling plants to gas turbines. Seemingly overnight, the industry became decentralized and asset light. Logistics providers could offer services in far flung locations without adding people and facilities. Small manufacturers could now become prime contractors by the ability to co-opt engineers spread across the global supply chain, each an expert on a component of the overall solution. These internet-native companies created value for themselves and also achieved a substantial shift in industry and society as a whole.

In the cloud plus mobile era, it was e-commerce companies who completely reimagined the relationship between products and people. From how desire and demand was created, to the act of browsing and selecting, to increasingly rapid delivery, and even continuous delivery on a subscription model.

Clearly, the idea of CEOs and senior leaders seeing new technology as a way to create radically new products and services is not new. In fact, we could say that encouraging and effecting such a change is senior leaders’ most important job. What is different with GenAI is that for the first time we have the ability to deal with complexities of information, expertise and reasoning that no previous era of technology was able to. This means that it can reach into almost every R&D process and every customer signal channel for every type of product. Now, the opportunity spectrum is wider, which is why it becomes even more of a CEO concern.

It’s not about pivoting to an entirely unknown business, but rather leveraging existing technical know-how, people, distribution chains and data to then apply a radical product innovation approach.

Every CEO is inundated with a thousand ideas to improve their business strategy. So how do you know what to focus on?

Radical product innovation may well be the biggest tool at your disposal—if you want to think big and pull the biggest lever for your business, then it’s a great way to transform and leverage what you already have. This means you can benefit from your existing place in the market and launch into a new era.

But it’s only possible through leadership that creates a culture of curiosity and innovation. As I saw AI shift to more high-stakes processes, I also saw senior leaders celebrate experimentation. As new AI capabilities came in, some at an immature stage, users discovered them and brought them into their reasoning flows with care, often cross-checking them against prior models and other ground truth. In other words, as always, more than technology, it is about ambition, leadership and culture.

Authors

Vijay Mital, Microsoft
Fawad Bajwa, Russell Reynolds Associates