A letter to boards thinking about AI

Why learning, navigating substitution and complementarity and data balance sheets will be key - and why you need to start using the technology now!

Dear board members,

By now you will be weary of everyone telling you that you need to implement AI, and that AI is transformative and that AI is a strategic priority. Yes, all of those things are true - but they are not helpful, in the sense that they do not provide you with the tools you need to think and act as a board.

In this open letter we aim to set out the core things that we have learned working with boards, with the technology itself and a range of innovative entrepreneurs. The purpose is not to claim that there is one true path forward, or that these are the 10 things you need to do (although we were tempted by the attraction such lists have for all of us). Instead, what we want to do is to introduce a few mental models that may help you think things through for yourself.

The reason for this is simple: anyone that claims to know what you should do, down to exact software or solutions, is overvaluing their expertise - but more crucially undervaluing yours. You are the expert on what your business does, how it works, where the pain points are and how it breaks down (incidentally that is one of the most useful questions to ask about any business: how it breaks).

Whatever you do, then, should start in your expertise, and then you can apply that in the mental models we provide. It is our conviction that if these mental models can help you start exploring the new opportunities this technology brings, then we will all be better off: the benefits of a technology are dependent on the diffusion of that technology through all layers of society, but especially dependent on businesses adopting the new tools.

And, of course, we would love to hear from you as well – we will continue to share ideas for board members and new technologies that we run into as well as stories about successes on this webpage. Send us your ideas, thoughts and questions at [address].

Start where you are

Your organization is almost certainly already using AI, whether through formal initiatives or grassroots adoption. Your employees are experimenting with generative AI tools, automating workflows, and discovering innovative applications. This organic adoption—what we call the "shadow dividend"—represents valuable organizational learning that often goes untapped at the strategic level.

The natural impulse for many boards and executive teams is to restrict this bottom-up experimentation out of legitimate concerns about security, compliance, and governance. However, this approach risks stifling innovation precisely where it's most likely to emerge—at the frontlines of your organization, where employees intimately understand operational challenges and customer needs.

Instead, we urge you to embrace and amplify this shadow dividend while implementing appropriate guardrails. Your employees are effectively conducting thousands of micro-experiments with AI tools, discovering applications that no centralized AI strategy could have anticipated. This distributed innovation represents an invaluable source of competitive advantage—if properly harvested.

Action items:

1. Survey and learn: Conduct a systematic audit of existing AI usage across your organization. Document what tools are being used, for what purposes, with what results, and what obstacles are being encountered. This bottom-up intelligence gathering will reveal patterns of value creation unique to your organization.

2. Implement guardrails, not roadblocks: Rather than prohibiting employee-driven AI adoption, establish clear security and safety guardrails. Create policies around data privacy, appropriate use cases, and security protocols. Make it easy for employees to stay within these guardrails while continuing to innovate.

3. Create sharing mechanisms: Establish forums where employees can share their AI use cases and insights. This could include internal showcases, communities of practice, or dedicated Slack channels. When someone discovers a valuable application, ensure that knowledge spreads.

4. Scale what works: Identify the most promising grassroots AI applications and invest in scaling them. Provide resources, technical support, and strategic guidance to transform individual experiments into enterprise capabilities.

By treating employee-driven AI adoption as an asset rather than a threat, you'll accelerate your organization's learning curve and discover applications that genuinely create value in your specific context. The most effective AI strategy will combine this bottom-up experimentation with top-down direction—not sacrifice one for the other.

Learning as a Process: From Noun to Verb

The first mental model that helps us is simple - move from the noun to the verb. Artificial intelligence sounds scary, partly because "intelligence" carries so much baggage. Like all nouns, AI seems solid and real. But with technology, we should always dig beneath nouns to find the verb hiding underneath. With artificial intelligence, we find an old, rich verb: learning.

This technology helps you learn more complex things faster than ever before. So one of the first questions your board should ask is how your organization learns, and how to use AI to strengthen this learning.

Remember that you come to this challenge from a position of strength. You know your company, and you have the domain knowledge that will really help you differentiate your company from the competition. Domain knowledge beats technology knowledge - but only if you learn to use the technology well.

Start by asking better, or new, questions. Don't ask "How do we implement AI?" Ask "Where do we get stuck when making decisions?" or "What would really change things if we could learn it fast and deep in our domain?" These questions map out where learning happens in your company—who learns what, how fast, and with what results. Ask how competitors learn, and benchmark your own learning - how long does it take to implement changes? How highly do you value learning on the individual and organization level - and how do you show that?

Build places to experiment. Learning organizations create spaces where people can test ideas safely. Set up small "learning labs" where different teams can work together on real problems using AI tools. Make these labs visible to everyone—a corner of the office where curious minds from accounting can see what marketing is testing, or a digital space where remote workers can follow experiments as they unfold.

Make room for mistakes. Yes, everyone says this, but few do it well. Create simple ways to learn from errors. When something fails, don't ask "Who messed up?" Ask "What did this teach us?" Then take the crucial next step: write it down. Keep a simple "lessons learned" document that anyone can access. The best companies create "failure libraries" where anyone can search for past mistakes and avoid repeating them.

Set up regular checkpoints. Learning happens in jumps and pauses, not smooth lines. Schedule quarterly "learning reviews" where teams step back from daily work to discuss what they've learned and what they've missed. Protect these meetings, even when business gets hectic. Make them concrete by reviewing actual data and examples, not just sharing feelings or impressions.

You may now wonder what this has to do with AI. The answer is that AI can improve your learning capability as an individual and organization by orders of magnitude. You unlock the real benefits here when you document your learning carefully. Think of it as creating a map of your company's growing knowledge. When you write down what worked, what didn't, and the patterns you notice, something magical happens—you can start to learn about how you learn.

This creates a compound effect. A sales team might learn that a new AI tool helps them predict which customers will renew contracts. That's valuable. But if they track this learning over time—how accurate the predictions were, which types of customers were predicted correctly, when the tool struggled—they learn something deeper: the patterns behind the patterns.

This second-order learning is where AI truly shines. When you feed an AI system not just your business data but data about how your teams learn and adapt, the AI can spot meta-patterns. It might notice that your engineering team learns fastest when paired with customer service staff, or that market insights take three months to flow from research to product design.

Simple tools work best here. Create plain-language learning logs where teams track what they tried, what happened, and what they think it means. Make these logs searchable. Review them regularly. Look for patterns across departments.

And it doesn’t have to be hard. AI allows you to essentially ask people to dictate their daily learnings for 5 minutes a day, and these can then be easily transcribed and stored in a learning log. The mere habit of asking “what did I learn today” will change your organization profoundly.

Organizations that build this habit—documenting learning so they can learn about learning—create an ever-widening gap between themselves and competitors. While others use AI to solve today's problems faster, these companies use AI to understand how problem-solving itself works in their unique context.

The spreadsheet your team creates to track experiment results becomes as valuable as the experiments themselves. The questions that didn't work become as instructive as those that did. The timeline showing how long it took for a new insight to change behavior becomes a tool for speeding up future adaptation.

This approach transforms AI from a tool that delivers answers into a partner that helps you ask better questions. And in a world where questions evolve faster than answers, that's the ultimate advantage.

A word on experiments and learning

AI opens up powerful new ways to experiment. Think of it as having a sandbox where you can test ideas without risk. This changes how fast your organization can learn.

Start with simple simulations. Before launching a new product, you might use AI to test how customers might respond. Feed it your customer data, product details, and market conditions. Ask it to play out different scenarios. What happens if you price higher? What if a competitor responds? The AI won't predict the future perfectly, but it will help you spot blind spots in your thinking.

Master the art of vibe coding - of coding with AI - and using the power the technology has to build simulations or games that help you understand your business better.

Create "what if" experiments. A manufacturing team can ask: "What if we rearranged our assembly line this way?" An AI model can simulate the change, showing likely impacts on efficiency, errors, and worker experience. This lets you try dozens of approaches in the time it would take to physically test just one.

Build digital twins of key processes. A digital twin is a virtual copy of something real—a machine, a supply chain, or a customer journey. Once built, you can stress-test it safely. A hospital might create a digital twin of its emergency room flow, then test how different staffing levels would handle a sudden influx of patients. The learning happens without risking patient safety.

Use AI to spot natural experiments already happening in your business. Companies often run accidental experiments—a manager in one region tries a new approach while others stick with the old way. AI can sift through your data to find these natural tests and measure their results objectively.

Test competing ideas simultaneously. When teams disagree about the best approach, don't waste time arguing. Use AI to help run quick, parallel tests. A marketing team debating which message will resonate can test all the options in small, targeted campaigns. The data settles the debate, and everyone learns from seeing which assumptions proved correct.

Make counterfactual thinking routine. After important decisions, ask AI: "What might have happened if we'd chosen differently?" This builds a habit of exploring branches not taken, making future decisions richer.

The beauty of AI-powered experimentation is its speed and safety. You compress years of trial-and-error - of learning - into months or weeks. You test risky ideas without actually taking risks. You explore extreme scenarios that would be irresponsible to try in reality.

This doesn't replace real-world testing—it makes it more targeted. When you finally implement changes, you do so with eyes wide open, having learned from dozens of virtual attempts. Your first real try benefits from all the learning gained in the digital sandbox.

Navigate substitution and complementarity

Complementarity and substitution are two fundamental economic concepts that describe how goods or inputs relate to each other in production and consumption.

Complementarity occurs when two items work better together than separately—their combined value exceeds the sum of their individual contributions. Think of coffee and cream, hardware and software, or AI and human judgment. When items are complementary, using more of one increases the value of using the other. This relationship creates powerful synergies in business contexts. For example, when digital technology complements skilled labor, productivity can rise dramatically as each enhances the other's effectiveness. The hallmark of complementarity is that demand for one item rises when the price of its complement falls.

Substitution, by contrast, happens when one item can effectively replace another, serving essentially the same function. Examples include butter and margarine, different brands of smartphones, or automated systems replacing manual labor for routine tasks. When goods are substitutes, they compete with each other, and increasing the use of one typically means decreasing the use of the other. The defining characteristic of substitution is that demand for one item rises when the price of its substitute increases. In business strategy, understanding whether technologies or workers function as substitutes is crucial for making sound investment and staffing decisions.

AI naturally pushes organizations toward substitution. The technology's ability to handle routine tasks creates an obvious, gravitational pull: replace humans with algorithms, cut costs, improve efficiency. This path of least resistance is tempting but limited. It follows a predictable pattern we've seen in previous technological shifts—automation replaces labor, profits temporarily rise, and capabilities eventually standardize across competitors - and this is where it gets dangerous, because now your entire business model is commoditized.

The innovative board, however, recognizes a different possibility. Rather than merely substituting AI for human roles, they can guide their organizations toward complementarity by fundamentally reimagining business functions when certain components are automated. This isn't just about preserving jobs—it's about transforming them into higher-value activities that AI can't easily replicate.

Consider customer service. The substitution approach simply replaces front-line agents with chatbots and voice assistants. The complementarity approach transforms the entire function. When AI handles routine inquiries, human agents can evolve into relationship managers or sales consultants with deep product knowledge. Companies like American Express navigated this shift successfully during the digital banking transformation—as online banking automated transactions, they retrained branch personnel to become financial advisors who could build deeper customer relationships and sell more complex products.

Manufacturing offers another example. When robots first entered factories, many companies saw them purely as labor substitutes. Toyota took a different approach with its "autonomation" concept—automation with a human touch. They used machines to handle repetitive physical tasks but elevated human workers to become problem-solvers and continuous improvement specialists - or learners! This complementary relationship between humans and machines helped Toyota maintain its competitive edge for decades while competitors who simply substituted machines for people saw their initial productivity gains quickly matched.

Channeling the resources freed up in substitution into learning accelerates the pace of learning overall.

IBM's transformation provides perhaps the most instructive case. As computing hardware became commoditized in the 1990s, IBM faced intense substitution pressure. Rather than competing in a race to the bottom, CEO Lou Gerstner recognized that IBM's human capital could be redirected toward higher-value services. The company pivoted from hardware manufacturer to solutions provider, with AI and automation handling routine aspects while IBM consultants tackled complex business problems. This dramatic business model reinvention turned potential substitution into powerful complementarity - and again, these high-level services were dependent on consultants learning more complex tasks.

The pattern is clear across these examples: innovative boards don't fight the substitution tide—they redirect it towards learning and invention. They ask: "If AI handles this function, what new, more valuable activities can our people pursue?" This approach transforms potential threats into opportunities for differentiation. While competitors use AI to do the same things more efficiently, these companies use AI to do entirely new things that create unique value.

And this question can be further refined: it is not just about valuable tasks, but tasks that are clear complementarities of what the business already does and improve its pace of learning. This may well mean that whole industries will re-structure, as you realize that the resources and capital freed up by substitution can be used to expand vertically or horizontally in your value network.

The data balance sheet

Just as your careful oversight of the financial balance sheet has guided this organization through economic cycles, we think that similar scrutiny must now be applied to what we call the "data balance sheet."

As a board, you already understand how assets and liabilities shape an organization's financial health. The same principles apply to data, though with important nuances that make this oversight both more challenging and potentially more rewarding.

Your organization possesses vast data assets scattered across systems, departments, and external partners. Customer interactions, operational metrics, employee performance, market signals – these represent the raw material from which AI can extract unprecedented value. Yet how many boards can answer basic questions about these assets with the same confidence they bring to financial discussions? Do you know which datasets drive the most value? Can you identify your most significant data gaps? Would you recognize if your organization were accumulating unsustainable data debt?

This last concept – data debt – may be unfamiliar but is profoundly important. When an organization postpones necessary investments in data infrastructure, accepts poor quality information in critical systems, or fails to document key datasets, it accumulates debt that compounds over time. This debt doesn't appear on financial statements, but it silently erodes your capacity to innovate, just as surely as financial debt can constrain investment. And this debt slows down your ability to learn and adapt in a quickly shifting landscape.

There is also another kind of debt here: data you do not control. If we agree that domain knowledge beats technology knowledge, we realize that domain data is a key differentiator and a competitive advantage. Do you control the data you need to learn in the right way, or is it owned or co-owned by someone else?

Consider how your suppliers and partners fit into this picture. Modern organizations operate in complex ecosystems where critical data flows across organizational boundaries. Your customer insights might reside partially in a CRM system, partially in a marketing platform managed by an agency, and partially in service databases maintained by support partners. A complete data balance sheet accounts for these extended assets and the complex rights and obligations that govern them.

FinTech Klarna provided an interesting example of this recently when they declared that they had brought their data inhouse to ensure that they could use it properly with their new AI-infrastructure. Other companies are surely preparing to do the same thing.

Perhaps most important are the empty spaces on your data balance sheet – what I call the "collection gap." These represent information that should be captured but isn't. In traditional accounting, if you're owed money, it appears as an account receivable. But who tracks the insights you're owed by your own operations? For example, if you're not systematically capturing why customers abandon purchases, or how specific operational changes affect quality, these represent uncollected data assets that should appear on your balance sheet.

AI thrives on data, but not just any data – it can really extract value from complete, representative, well-structured information. It may well work with unstructured data as well for some use cases - so do not discard that - but the structure helps. The organizations that will extract the most value from AI investments are those that have meticulously managed their data balance sheets long before implementing sophisticated AI-solutions.

Here it will really help if you begin asking different questions in your oversight role on the board beyond the traditional "How much are we investing in AI?". Consider questions like:

"What is the current state of our data balance sheet?" "Where are our most significant data debts, and what is our plan to address them?" "How do we value our data assets, and which create the most strategic advantage?" "What critical information should we be collecting but aren't?"

By expanding your governance approach to include the data balance sheet, you'll ensure that AI investments rest on solid foundations rather than shaky assumptions. You'll see more clearly where strategic data investments should precede technological ones. And you'll develop a more sophisticated understanding of the true drivers of value in an increasingly digital world.

The organizations that master this mental model won't just implement AI more effectively; they'll fundamentally transform how they create and capture value. Just as financial capital was the defining resource of the industrial age, data is becoming the essential resource of the AI era. Your careful stewardship of this resource will determine whether your organization thrives or merely survives in the years ahead.

And yes, we also believe that ultimately your data balance sheet will merge with your financial balance sheet - this is already happening today, as companies collect and shape up their data and use it as collateral for growth loans.

Practice Makes Possible: The Board Member's Learning Journey

As we've explored how organizations can harness AI through learning, navigating substitution and complementarity, and managing the data balance sheet, we must now turn to perhaps the most fundamental question: how do you, as a board member, develop the personal literacy and experience with AI that will enable you to guide these transformations?

The answer is deceptively simple yet profound: daily practice. This isn't merely about staying informed—it's about developing a new way of thinking, one where AI becomes an extension of your cognitive process.

Anders Ericsson's groundbreaking work on deliberate practice reveals that expertise doesn't emerge from occasional engagement but from consistent, focused practice. Just as learning to play the piano requires daily discipline, developing AI fluency demands regular, hands-on experience. This isn't a theoretical understanding you can delegate or outsource—it's a practical wisdom that must be personally cultivated.

Begin by making AI part of your daily workflow. Start conversations with it in the morning to organize your thinking. Use it to summarize board materials, extract key questions, or simulate different scenarios before making decisions. Draft communications with its assistance. The goal isn't to automate your thinking but to extend it—to create a partnership between your judgment and the technology's capabilities.

As you practice, embrace what might seem counterintuitive: be unreasonable in your expectations. We consistently underestimate what these systems can do because we approach them with preconceived limitations. Ask the impossible. Request analyses you believe are beyond its reach. Pose questions that seem too complex or nuanced. You'll be surprised how often the technology exceeds your expectations, revealing capabilities you hadn't imagined. This "productive unreasonableness" stretches both the tool and your understanding of it.

Experiment with new workflows that challenge traditional patterns. Dictation offers a compelling example—most professionals can speak three to seven times faster than they type, yet we remain locked in typing-centered workflows. Try dictating your thoughts and using AI to refine them. This simple shift can dramatically accelerate your ideation process and open new cognitive pathways. The technology becomes not just a tool but a thinking partner that helps you access your own insights more efficiently.

Perhaps most uniquely, use the tool to learn the tool. AI is the first technology that can teach you how to use itself more effectively. Ask it how to formulate better prompts. Request guidance on complex tasks. Have it suggest workflows you haven't considered. This recursive learning—using AI to learn about AI—creates an accelerating cycle of improvement that is unique to cognitive technologies.

As you develop this personal practice, you'll notice something remarkable: the boundaries between your thinking and the technology's capabilities begin to blur. You'll develop intuitions about when to lean on AI and when to rely on your judgment. You'll recognize patterns in how it responds to different types of questions. You'll build a working relationship with the technology that transcends superficial understanding.

This daily practice serves another crucial function: it models the behavior you want to see throughout your organization. When board members engage directly with AI rather than treating it as something to be understood only in the abstract, it sends a powerful signal. It demonstrates that AI literacy isn't a specialized technical skill but a fundamental capability for modern leadership.

The organizations that will thrive in the AI era won't be those with the most sophisticated algorithms or the largest data repositories—though those matter. They'll be the ones where AI becomes woven into the cognitive fabric of the organization, starting with the board. They'll be places where learning happens not just about AI but through AI, where the technology becomes a catalyst for human potential rather than a replacement for it.

As you leave this letter and return to your responsibilities, I offer this simple challenge: make AI part of your daily practice for the next 30 days. Use it not just for obvious tasks but for stretching your thinking in new directions. Be unreasonable in what you ask of it. Experiment with workflows that might initially feel awkward or unfamiliar. Ask it to teach you how to use it better.

At the end of those 30 days, reflect on how your understanding has evolved. Notice the questions you've learned to ask that weren't obvious before. Observe how your relationship with the technology has changed. Then bring those insights to your board discussions, not as abstract concepts but as lived experience.

The future of your organization in the AI era won't be determined by the technology itself but by your collective ability to think with it, learn through it, and imagine possibilities beyond it. That journey begins not with grand strategies but with daily practice—with the seemingly small decision to make AI your cognitive companion rather than just another item on your agenda.

The piano doesn't play itself, and neither does AI reveal its true potential without your engaged participation. Begin practicing today, and watch as new possibilities emerge—not just for your organization, but for how you understand your role within it.

And tell us what you learn, because we want to learn with you.

Petri Kokko / Nicklas Berild Lundblad