I’m pleased to continue our series of interviews with innovators and leaders who are building and deploying AI-driven solutions within enterprises.

This time we take a deep dive into AI deployments with Devesh Mishra, President and Chief Product and Technology Officer of CoreAI at Keystone. Devesh’s extensive prior experience of implementing AI and ML technologies at Amazon and Deliveroo gives him unique insights into the technical and commercial challenges and opportunities posed by AI.
Founded 22 years ago, Keystone has been helping organizations understand and deploy digital technologies across all industry sectors from its offices in the US and UK. CoreAI was launched in 2023 to unify their “AI and ML capabilities into a single service to help clients solve their biggest operational and commercial challenges with custom-built enterprise AI solutions.”
As Devesh explains, CoreAI’s approach is to help businesses use AI and ML technologies to create more efficient systems rather than simply looking for short-term productivity gains. This notion of “optimality” applies AI at the operating level which drives transformation across multiple dimensions of a business, not just incrementally at the edges.
What prompted Keystone to create its CoreAI offering in 2023?
From the start, Keystone has been in the business of helping clients solve business and legal challenges through a combination of technology, economics and strategy. In the early 2000s, our co-founders Greg Richards and Marco Iansiti, the longest-standing technology professor at Harvard Business School, anticipated how the internet, cloud computing and AI would reshape business and public policy and require new skills to succeed. Embracing this vision, they created a different type of advisory firm built on these multidisciplinary principles. Our team of data scientists, economists, engineers and academic experts has long leveraged these AI and machine learning capabilities to help clients by providing expert economic and technology services for their most challenging problems.
In 2023, demand for AI-driven operational and commercial solutions from clients reached a tipping point. It presented an opportunity to create a dedicated offering. We’ve seen tremendous growth within the first 18 months of formally establishing our CoreAI division and are excited to roll out new products in 2025.
The focus of CoreAI seems to be on helping businesses improve their business processes including pricing, supply chain optimization and forecasting. Can you give any examples of the types of projects you have worked on and in what sectors?
CoreAI has supported pharmaceutical leaders, e-commerce giants, CPG manufacturers and other clients in automating decision-making in supply chain optimization, forecasting, customer attribution and more.
One example is a multi-billion-dollar biotech company that hired us to solve supply and demand imbalances. We created highly accurate short- and longer-term demand forecasts that led to a less than 5% average deviation from actual administrations, as well as a dynamic supply chain control tower that automated replenishment decisions, which drove $300-$500M in free cash flow and 2% returns (vs 9-11% historically).
Our team has also been retained by a multi-billion-dollar e-commerce leader in Latin America to provide greater clarity and consistency in measuring ROI for marketing initiatives. We developed causal models to improve customer targeting, leading to 70% higher profits. We also helped to optimize the e-commerce’s assortment planning, substituting the bottom percentile SKUs to improve margins by 8%.
Do you work with specific AI technology vendors or choose solutions based on individual client needs?
We do not work with any one specific AI technology vendor. Rather, we create custom massive-scale algorithms and models that are built on clients’ internal data and existing tech stack, and recommend and engage the best fitting standard components from hyper-scalers across connectors, integration services, storage, AI/ML services, and UI/UX. What differentiates Keystone is that we have a deep bench of elite AI/ML and data science talent that allows us to do the heavy lifting of data science and model building ourselves.
Keystone builds and operates the initial solution, then transfers ownership directly into the client’s organization, eliminating the need for third-party subscriptions that restrict functionality and transparency.
What are some key issues you see clients struggling with in their use of AI?
There are four main hurdles organizations are struggling with when it comes to getting the most out of their AI efforts.
First is data readiness and an expertise deficit. Implementing AI requires a combination of applied economists, machine learning engineers, and technical product managers working in concert; most enterprises simply don’t have these skill sets in house yet.
Second is an overfocus on narrow use cases of GenAI. Many organizations are still focused on addressing productivity gains and getting stuck in “workforce AI,” which centers on using GenAI to help employees spend less time on activities or eliminate manual processes. This is table stakes now, and it doesn’t create competitive differentiation for organizations. True business transformation comes from applying AI at the operating level to drive commercial and operational efficiencies.
Third, we’re seeing organizations struggle with adopting and aligning to this new mental model. Leadership must first articulate a bold vision and audacious goal, then allocate the resources needed to drive and be accountable for achieving these goals. These goals must be jointly owned between regions and a center of excellence. We’ve seen over and over that a centralized center of excellence is crucial in making AI efforts successful. Organizations need the deep involvement of the business units and divisions that will benefit from the insights that AI provides, as they will work closely with the centralized team to provide data, local insights, and make high value judgments on AI output.
And finally, securing buy-in from senior leadership can sometimes be a challenge. More studies are pointing to AI fatigue and skepticism about its output. That said, enterprise investment in AI is projected to increase in 2025. There is an opportunity for businesses to gain competitive advantages if they move away from generic and narrower AI applications and focus more on custom-built AI models and algorithms that can fundamentally transform enterprise decision-making.
How do you see the types of assignments that CoreAI work on evolving over the next 2 to 3 years?
We partner with our customers on a journey to transform how they operate, using the most advanced AI, science and engineering in the world. After starting with high impact, rapid ROI point solutions such as forecasting and downstream impact, we fundamentally transform the way their organizations operate by evolving them to put human-first AI systems at the core of their prediction and decisioning operations. Humans make high value judgments while AI makes high volume decisions. The outcome goes far beyond improving returns on individual investment decisions; it drives a flywheel that fundamentally increases enterprise value.
Gartner sees GenAI moving from the Peak of Inflated Expectations into the Trough of Disillusionment on their Hype Cycle model. What are your thoughts on this and whether some of the expectations around AI-driven change in the enterprise have been unrealistic?
AI has completely altered industries, and the speed at which it is advancing is also creating a level of uncertainty and sometimes confusion among businesses that often drives them to invest in multiple pilots simply to try and understand it and stay ahead of competition. The trouble with this approach is that a very small minority of these projects reach production, and they’re costly. The other issue is that most of the pilots focus on productivity through GenAI and large language models versus true business transformation at the operating level. This hyper-focus on productivity instead of optimality has been one of the biggest blind spots for enterprises.
There is a much broader generation of advanced AI models that have been left out of the conversation, often called “operational AI” or “core AI,” and it is the best way for businesses to build a sustained competitive advantage. It applies AI at the operating level, through algorithms and models that are custom built from internal data and econometric techniques rather than external, generalized data from the internet or third-party applications. It drives transformation across multiple dimensions within the business.
I think business leaders’ expectations are starting to shift; more are re-evaluating their investments and starting to consider different metrics for their AI projects. It’s important to recognize that real ROI takes time – in technology, skills, and organization.
We hear that agenticAI will be a theme of 2025 – what demand are you seeing on this front?
At Keystone, leveraging AI to automate decision-making has been central to our mission long before the term ‘Agentic AI’ gained traction. For example, during my decade at Amazon, I built AI systems that automated retail demand forecasting and inventory planning, enabling these systems to place purchase orders for hundreds of millions of SKUs autonomously. Today, we are already leveraging advancements like the Transformer architecture which are at the core of large language models to transform enterprise decision-making by enabling AI systems to operate with greater autonomy in dynamic environments. We see growing demand from businesses looking to harness this next generation of AI for more adaptive and self-directed decision-making, and Core AI is well-positioned to lead this evolution.
The nature of consulting and its reliance on human inputs raises scaling challenges. To what extent is Keystone/CoreAI able to overcome these challenges with its offerings?
Rather than “consultants,” we consider ourselves builders who are working in partnership with the enterprises we serve. Our approach employs a ‘build, operate, and transfer’ framework led by two cross-functional teams – our Keystone CoreAI project team and the client team that’s made up of information tech, developers and data scientists. We partner with clients to build, train, deploy, and finetune solutions, then train clients’ internal talent to lead long-term AI initiatives and take over operational responsibilities. This allows clients to ultimately operate independently.
The other thing I’ll say about scaling is that with AI, the value of scale never stops climbing, whereas traditional models will eventually reach a point of diminishing returns. We also design our solutions so that humans and machines work seamlessly together, meaning we build algorithms and models that focus limited human resources on the high value judgments they’re best at with the full-scale potential of things AI is good at, which are making high volume decisions. It fosters a culture that is scientific in its application of continuous improvement. It positions enterprises to automate and scale end-to-end decision-making across the business.
As companies transform themselves into more digitized organizations and begin to put more of their operating model in digital form, they will see greater growth and differentiation. AI is becoming core to business strategy.