Mayank Kejriwal – CEO and Co-Founder of GRAIL.
getty
When business leaders talk about AI, the conversation usually centers on automation. The assumption is that if a system can draft, summarize, analyze or answer questions faster than a human, then the main value lies in labor savings. I think that framing is too narrow. In many organizations, I believe the deeper bottleneck is not a shortage of tools, but weak institutional memory.
The Key Problem With Fragmented Knowledge
By institutional memory, I mean the accumulated knowledge that helps an organization actually function: how decisions were made, which proposals succeeded, where key documents live, which experts know what and what the real workflow looks like rather than the official one. It affects how policy, judgment and precedent interact in practice. In too many enterprises, that knowledge is scattered across inboxes, PDFs, spreadsheets, Slack threads, legacy SaaS platforms and the heads of a few experienced employees.
That fragmentation is expensive. McKinsey has estimated that knowledge workers spend a large share of their week writing emails, searching for information and collaborating internally, and that improving communication and knowledge-sharing workflows could raise productivity by 20% to 25%.
More recently, Deloitte warned that the coming retirement wave could trigger a massive institutional knowledge exodus, noting that many organizations still fail to capture critical knowledge before experienced employees leave. Far from being side issues, these are critical operating constraints for entire industries.
Memory As A Strategic Asset
This matters even more in knowledge-intensive sectors such as research, healthcare, defense, advanced manufacturing and professional services. In those environments, the real cost of weak institutional memory is slower execution, inconsistent decisions, harder onboarding, and avoidable dependence on a handful of people who “just know how things work.” When those people leave, or when the organization tries to scale, the system becomes fragile.
That is one reason I think AI’s most durable enterprise value may come not from replacing people, but from helping organizations turn scattered memory into usable infrastructure. Used well, AI can help surface prior work, connect related documents, summarize history and make institutional knowledge available at the point of work. But that only happens if leaders treat memory as a strategic asset rather than a by-product of operations.
Why AI Is Not An ‘Answer Layer’
In higher ed, research administration offers a useful example. At research-intensive institutions, critical knowledge is often distributed across specialized offices, sponsor websites, internal guidance, prior submissions and experienced staff. A paper in the Journal of Research Administration describing Duke University’s myRESEARCHpath noted that when information from research support offices sits in disparate locations, investigators and administrators struggle to find what they need when they need it. That problem is not unique to universities. It is a familiar pattern across enterprises: The information exists, but not in a form that is easy to trust, navigate or reuse.
This is why I would caution leaders against treating enterprise AI as an “answer layer” floating above organizational chaos. If the underlying knowledge environment is fragmented, outdated or poorly governed, AI will often amplify those weaknesses. It may retrieve the wrong policy, miss key context or produce a plausible-sounding summary of incomplete information. In other words, AI readiness is often a knowledge architecture problem before it is a model problem.
Asking The Right Questions
The practical implication is that leaders should start asking better questions. Not just: Which model should we use? But also: Where does our critical knowledge live? What proportion of it is actually accessible? Which workflows still depend on tribal knowledge? What happens when a senior employee leaves? Where do users go when they need context, not just data? Those questions get much closer to whether an AI initiative will create durable value.
I also believe this is why the best AI deployments tend to start with narrow, high-value knowledge workflows. Instead of trying to build a universal agent for the whole company, it is often wiser to begin with a bounded use case where fragmented memory is already hurting execution. That might mean proposal development, technical support, internal research, onboarding or portfolio reporting. In each case, the win is not just faster output but stronger organizational recall.
In the research office, the challenge is rarely that institutions lack information. It is that critical knowledge is spread across too many systems, documents and people to be used consistently at the right moment. Rebuilding that layer should not be thought of as a software problem, but as an opportunity to make institutional memory more searchable, traceable and operational.
The Big Idea
My takeaway for executives, especially in higher ed, is simple: Before asking how AI can automate more work, ask whether your organization can remember what it already knows. In many enterprises, that is the more important problem. The organizations that win with AI will not just be the ones that generate faster answers. They will be the ones that build stronger institutional memory and turn it into a real competitive asset.
Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?
