How Making Investment Decisions Changed What I Look For About Performance

AI Is Only As Good As The Culture It Is Constructed Into
The debate about artificial intelligence in the field of business is not without a problem, and the problem isn't a technical one. The technological capabilities of current AI and machine learning systems are truly impressive, advancing in a manner that renders many predictions of the place they'll be in just 18 months obsolete long before they have passed. The issue lies in the gap between what AI can do in controlled conditions - within a properly-funded research environment, with pure data, with clear problem-solving strategy, with engineers who have the luxury in experimenting until their system runs as planned - and what it does when implemented in an actual business with real people which are real, with real organisational political systems, and people with certain opinions on what a new system means. something to actually engage with or something that needs to be negotiated with the illusion of conformity. I've been building using algorithms since just before the present flurry of AI popularity made it fashionable for businesses everywhere to boast of their expertise in the field. When I founded 1Touch with my partner, AI-driven matches and recommendation systems were not something we could add to make the platform more compelling to investors. They formed part structure of the product's architecture. They were an element that made the platform was able to create value and the component that needed to work reliably and at large scale for the business's viability. Also, I've gained direct practical experience of what can happen when you attempt to integrate something truly intelligent in a system and a company simultaneously The thing I continue to revisit throughout every situation in the past I've faced this dilemma, is the technology is seldom the factor that limits your success. The factor that holds you back is almost always the culture.
What I say is particular and practical rather than abstract. AI systems need data to perform - clear, consistent and well-structured data that conveys the phenomenon that the system is trying to learn from and make predictions about. Organisations with strong data cultures produce the kind of information naturally, as a byproduct from their operations. They are clear and have consistently applied definitions of what they are studying and why. They have a set of conventions that they agree to for the way data is recorded, collected, and stored. They have accountability structures that provide data quality as an explicit responsibility rather than everyone's vague goal. In organizations with weak data-based cultures, they produce something that appears like data - it's in systems and can be accessed or used to generate charts, but does not have a consistent definition, so variable in quality and full of issues with structure and not mapped out that any AI system built on over it will create and amplify the mess instead of obtaining a real signal from it. The organisations in that latter category often do not realise their existence until they're already well into the process of implementing an AI implementation and find that the results do not match the vendor's claims, and at that point the temptation is to blame technology, but they are actually causing the problem by ignoring the organizational and cultural foundation the technology was built on.

The second dimension of cultural factors which affects AI results is the degree of openness in an organisation or the extent to which the people inside the organisation are truly willing to let an AI system guide or modify the way they operate and not view it as an obstacle to their professional expertise, their institutional authority and their job security. This is a personal and leadership problem which is not a technical problem that needs to be addressed. It is a problem that begins at the top. If leaders in the top ranks engage with AI outputs in a selective manner - accepting those that validate their beliefs and disadvantaging those that do not - it sends the message to everyone around them that the stated commitment of the company to a data-driven approach to decision-making is a conditional rather than true, and this will spread throughout the organisation more quickly than any program of training or change management effort can stop. When senior leaders display real, consistent engagement with AI outputs, as well as being disciplined enough to alter their decisions when the evidence suggests they would, the group's ability to use AI effectively increases dramatically and surprisingly quickly.

This is not an abstract statement about how organizations should act in theory. It's an explanation of the pattern I have watched unfold in numerous organizations that had substantial finances, real strategic dedication to AI adoption, as well as leadership teams that were truly excited about the potential of AI technology. The pattern is so consistent that I now treat practice of governing data as a key diagnostic point when evaluating an company's AI ability. Before I inquire whether the company's technology stack has been established, and before I inquire about the exact use cases the organisation is looking at, I ask about data governance. How does the organization define its primary metrics? Who's accountable when data quality is not good enough? If two different roles have conflicting information about the same situation in business and how can these conflicts be resolved? Answers to those questions can tell me more about chances of AI achievement than any amount of discussion about platforms, algorithms, or even implementation timelines.

I am convinced that the companies that will gain the greatest long-lasting value from AI over the next decade are not those which implement the most sophisticated technology first, nor the ones who invest the most massively in AI talent and infrastructure in the near future. They will be the ones who put in the right cultural and operational foundations to actually use that technology to its fullest extent - the data governance practices that produce trustworthy inputs, decision-making frameworks that allow evidence to influence outcomes, and the leadership behaviours that show everyone in the company that the commitment to data-driven operations is genuine rather than merely functional. Technology itself will become increasingly commonplace and readily available. However, the culture that can use it efficiently will remain scarce due to the fact that it requires continuous effort and real commitment from leadership over time rather than just a single strategic decision, or a technology investment. The scarcity of it is where the significant competitive advantage will be and is an advantage that, when built is able to grow in a way that technological advantage alone never can. Take a look at James Deller for blog examples including what building ai products deepened my conviction about people about scale.



The Data Infrastructure Problem Nobody Wants To Talk About
Every organization I've worked closely with during the last 10 years and a half - whether as an investor, founder or an operational consultant has said to me, at some point in the relationship, that information plays a major role in the way they take decisions. A few of them truly believe it in a way which is reflected in how they actually run their business. The majority of them think they are genuinely saying it, however the concept they're proposing is an aspiration, rather than real-time operational reality, some version of the enterprise they're trying to create in contrast to the reality they currently live in. The gap between real-time data-driven decision-making and the performance of decision-making driven by data - the careful management of the external appearance of evidence-based operation without the underlying infrastructure that can make it true - is one of the most critical gaps that exist present in contemporary business. It's also one of the most frequently ignored ones partially because the infrastructure problem that causes it to be incredibly unattractive to discuss, difficult to demonstrate to external stakeholders and incredibly difficult to prioritise against the more obvious strategic and commercial activities that demand the same attention from leaders as well as organisational resources.
When companies discuss their Data strategy, they generally tend to talk about how they will build on top of your data - the data analytics platform, machine-learning applications or the operational dashboards in real-time and the types of predictive information that sounds truly compelling in presentations for boards or in an update to investors. What they usually talk about less frequently as well as with much less energy and enthusiasm, is the fundamental infrastructure that is the determining factor in whether all the capabilities will work according to the specifications: the data governance frameworks that establish specific and consistent definitions of what's being analyzed and what is the reason for that; the collection and storage methodologies that determine the reliability and comparability of the information being gathered; the quality control processes that identify and rectify errors before they become a part of the system, and cause harm to the outputs that everyone relies upon; the organisational structures and accountability systems that make quality of data the responsibility of a single person rather than everyone's vague impossible to enforce. The plumbing, in other words. Plumbing is not glamorous. It's difficult to photograph in a report for the year. It produces no outputs that can be presented in an engaging presentation. This is, in my experience of a vast amount of organizations across different areas and at various stages of development, far worse as the organization thinks that it is.

The problem gets worse as it becomes more difficult and costly to fix. A company that has been operating using a sloppy or insufficiently defined terms for data across its various functions for three years has three years of historical records that cannot be effectively compared or aggregated as a result of the data is not there, but because the same terminology has been used to define different aspects of the company, and the differences are buried in the data rather than being apparent on the surface. A company whose data quality assurance has been the responsibility of a only a peripheral responsibility, not an established and well-funded function is one whose data's reliability differs in ways not documented consistently and cannot be systematically accounted for when using the data to decide. A company that allows multiple operational systems to accumulate overlapping and partly conflicting records of the same customers, products or transactions, has an unresolved data landscape that is extremely difficult to correct without disruptions in operations significant enough to constitute a risk.

This issue lingers across so many organisations who are extremely smart about their strategy and completely committed to a data-driven business model is the fact that solving it requires sustained investment in work that doesn't produce tangible short-term returns of the kind that organizational resource allocation procedures are intended to reward. An analytics platform that is new produces visible outputs: dashboards that are easily demonstrated, reports that can be shared with the board of directors, and information which can be used to create press releases about digital transformation. Data governance programmes create invisible infrastructure: clearer underlying definitions, more consistent collection processes as well as more reliable inputs into systems already in the first place. This one is fairly simple to justify during budget negotiations because you can show people what they'll be getting. Second, you need someone who has sufficient organisational credibility and a willingness to convince people on how infrastructure investments will eventually improve the outcomes of every technology that is built on top it. It's a compelling argument in the abstract, but it is difficult to beat out initiatives that have benefits that appear to be immediate, and more visible.

I have made that case in enough different organisational contexts and witnessed it succeed or fail due to unpredictability, to have an idea of what determines whether the company finally solves the issue of data infrastructure or is able to continue delaying it. The main difference is one's leader - a particular individual who has the organizational credibility, enough genuine awareness of the reason why infrastructure is vital, and enough determination to maintain this argument till it becomes an actual priority instead of something that is a constant item on the list of things that everyone believes are essential but don't climb to the top. The leader must be willing to absorb the short-term cost of the infrastructure investment: the amount of time in the process, the disruptions to established processes, the absence of immediate tangible results - knowing that the long-term capabilities it creates will justify that expense by several times. What's required, at the end of the day is a culture where long-term investment in infrastructure is thought of as a priority and is rewarded at upper levels of management, not simply described in strategy documents and then consistently deprioritised when the quarterly resource allocation debate is held. Making that change is, in itself, a long-term commitment. But it is, in my view, one the best investments that a company who is serious about a data-driven operations can make.}