Why the businesses that make data reliable, accessible and commercially useful deserve a place in the AI investment conversation
By Luke Liplijn, Founder and CEO of Liplyn Group
When evaluating an AI opportunity, the obvious question is which model will deliver the best performance. The more useful investment question may be what makes that performance valuable to a paying customer.
An impressive demonstration is one thing. A system that consistently supports decisions, fits into business operations and generates sustainable revenue is another.
At Liplyn Group, we approach this from the perspective of an investment group building data-driven businesses and financing growth through corporate bond offerings. Our focus includes data management, business information and the infrastructure that supports practical AI applications. (Liplyn Group)
Our investment thesis starts with a simple conviction: some of AI’s most durable commercial value will be built beneath the model, not inside it.
A smarter model cannot fix an undefined business problem
Consider a company building an AI assistant to help its sales team identify opportunities.
Its customer database contains duplicate records. Finance groups accounts by legal entity, while sales groups them by brand. Historical transactions are stored separately, and nobody has agreed on what qualifies as an active customer.
Ask the assistant to identify the company’s most valuable customers, and the problem becomes clear. Before choosing a model, the business needs to decide what “customer” and “valuable” actually mean.
A more capable model may help interpret information. It does not remove the need for consistent definitions, trustworthy records and clear ownership.
The same issue appears in retrieval-augmented generation, or RAG: systems that retrieve external information to inform a model’s response. Microsoft’s technical guidance identifies content preparation, access across multiple sources, retrieval relevance and permissions as central implementation challenges. Connecting a model to a document repository is not the same as giving it reliable business knowledge. (Microsoft Learn)
The old principle still matters: rubbish in, rubbish out.
Better data does not guarantee a successful AI application. But better modelling is not a substitute for knowing whether the underlying information is fit for purpose.
The asset is not the database. It is the ability to trust it.
When two competitors can access the same model, model access alone offers little differentiation.
The more interesting question is what each company can bring to that model.
Imagine one business with a collection of disconnected records. Now imagine another with consistently identified entities, documented sources, useful historical information, clear usage rights and established processes for keeping everything current.
Both have data. Only the second has the foundations of a reusable information asset.
For an investor, I would argue that the distinction is more important than the size of the database. A billion records do not automatically create a competitive advantage. The investment question is whether those records can support a valuable service that customers trust and continue to use.
The work involved extends beyond the initial build. Google researchers have described how machine-learning systems accumulate technical debt through data dependencies, feedback loops and changes in their operating environment. The maintenance burden belongs to the wider system, not just the model. (Google Research)
That is why we view the people, processes and infrastructure around data as part of the asset itself.
Data management belongs in the investment discussion
I believe data management should be assessed through the business capabilities it creates, rather than treated exclusively as an IT expense.
The first capability is reuse. If a trusted company record can support customer onboarding, sales intelligence and internal reporting, the organisation has the opportunity to avoid solving the same identity problem three times.
The second is accountability. A useful system should allow people to investigate where information came from, how it was transformed and who is responsible for correcting it. The NIST AI Risk Management Framework similarly places trustworthiness across the design, development, use and evaluation of AI systems, rather than treating it as a final approval step. (NIST)
The third is adaptability. Our preferred architecture is one in which the organisation can improve or replace its AI components without having to reconstruct its business knowledge from scratch.
This does not make data infrastructure inherently profitable. Storage costs, engineering complexity and unnecessary duplication can destroy value just as easily as poor modelling.
The commercial test remains straightforward: does the investment help deliver a better service, reduce avoidable work or create something customers will pay for?
Build the foundation around a useful workflow
Our approach separates two connected layers of value creation.
The first is the data foundation: connecting relevant sources, resolving identities, establishing definitions, documenting provenance and making information available under appropriate controls.
The second is the application layer: turning that foundation into useful products and workflows, such as business research, customer intelligence or decision support.
The order matters, but it should not become an excuse for a multiyear data-cleaning programme with no commercial outcome.
A practical approach is to start with a specific customer problem. Establish the data foundation needed to solve it, test the result and expand from there.
For example, a business-information product might begin by reliably connecting companies, directors and addresses. The next step could be a research workflow that helps a customer understand those relationships without manually reconciling multiple sources.
In that example, the product’s value would not come merely from adding a conversational interface. It would come from reducing the work required to reach a useful, verifiable answer.
The objective is not to collect more data. It is to make reliable information easier to act on.
What this means for the investment case
From our perspective, this creates an investment opportunity beyond the race to develop the most powerful foundation model.
We are interested in businesses with practical data expertise, established customer needs and a credible route from information to revenue. Liplyn Group’s stated strategy combines business information and data-management capabilities, using a buy-and-build approach supported by bond financing. (Liplyn Group)
The questions we consider most useful are commercial as well as technical.
What problem does the customer pay to solve? How difficult would the service be to replace? What does it cost to maintain the data? Can the business grow without its delivery costs growing at the same rate? And how much cash remains after the investment required to keep the product competitive?
These questions help distinguish an attractive technology story from an investable business.
Our thesis is not that every dataset is scarce or every infrastructure provider is undervalued. It is that businesses capable of turning difficult, fragmented information into dependable services deserve closer attention.
For bond investors, business quality must translate into repayment capacity
There is an important distinction between creating enterprise value and generating a return for bondholders.
A bond investment is a lending relationship. Interest and repayment depend on the terms of the instrument and the issuer’s ability to meet its obligations. A compelling AI strategy does not remove credit risk, and the ability to sell a bond before maturity may be limited. (Investor)
For that reason, we believe a data-economy investment thesis should be accompanied by disciplined scrutiny of cash flow, debt levels, acquisition execution and repayment planning.
The attraction should not be a promise that AI will make everything more valuable. It should be a clearly explained business strategy, supported by financial information that investors can assess.
The Takeaway
The question I would encourage both executives and investors to ask is not simply: How advanced is the AI?
It is: What makes the underlying business difficult to replace?
Our answer starts with trusted information, operational expertise and services that solve persistent customer problems.
Models matter. Applications matter. But the data foundation deserves equal attention—because without it, an impressive answer may remain just that: an answer, rather than a dependable business outcome.
Learn more about Liplyn Group’s data-economy investment strategy and corporate bond offerings, including the applicable investment documentation.
Disclosure: This article presents the commercial perspective of Liplyn Group. The author is its founder and CEO, and Datafloq is part of the wider Liplyn network. This is not personalised investment advice. Capital is at risk; participation is subject to the applicable offering documents and investor eligibility requirements.
