Garbage in, garbage out. Diamonds in, diamonds out. |

Why we don’t spend money on AI guarantees, however within the layer beneath them, and why our bondholders profit.

Stroll into ten Dutch boardrooms at the moment and 9 of them can have AI on the agenda. Copilots, brokers, generative fashions in customer support. The ambition is there, and infrequently the funds too.

What’s often lacking is the muse able to supporting that ambition.

The information is scattered throughout techniques that have been by no means designed to work collectively. Definitions range by division, “buyer” means one thing completely different in gross sales than in finance. Historic information is incomplete or can’t be reproduced. Possession has not been assigned. Knowledge high quality is measured when an issue happens, not as a matter of routine. And governance exists on paper, not in follow.

Then the mannequin arrives. And the mannequin does what it at all times does: it amplifies no matter you place into it.

I’ve been working with information for greater than twenty years. Each mannequin I’ve seen, statistical, machine-learning or generative — is ruled by the identical rule:

A mediocre mannequin educated on glorious information outperforms a superb mannequin educated on mediocre information. Virtually each time. That’s not an opinion; it’s the expertise of anybody who has ever needed to clarify a manufacturing mannequin to a shopper who didn’t belief the reply.

The implications of this are broadly underestimated. Fashions have develop into a commodity: they’re out there to everybody, at falling prices and with narrowing variations between them. What isn’t interchangeable is the information layer beneath, the infrastructure, high quality, definitions, historical past and possession.

That’s the place shortage lies. That’s the place the defensible place is constructed.

In lots of organisations, information administration sits underneath IT operations. It’s handled as a value line, not as worth creation. That historic mistake is now turning into painfully seen.

Three causes we deal with it as an asset:

1. It can’t be copied. A competitor can purchase the identical mannequin tomorrow. What they can not purchase tomorrow is fifteen years of fresh, linked and documented information historical past.

2. It units the ceiling for each AI funding. Skip the information layer and also you pay twice: first for the venture that fails, then for the muse you continue to should construct.

3. It has develop into a governance crucial. Traceability, explainability and demonstrable high quality, as soon as thought of good follow, at the moment are stipulations for working in at the moment’s regulatory surroundings.

Knowledge administration isn’t merely a prerequisite for the AI technique. It’s the AI technique — in the one space you really management.

Liplyn Group doesn’t spend money on AI guarantees or in start-ups which have but to show there’s a market. We spend money on established Dutch information firms: companies producing income at the moment, with prospects who’ve stayed for years, and with providers or merchandise in information administration, information infrastructure, information high quality and determination intelligence.

The profile we search for is constant:

•     Confirmed observe file. A number of of our working firms have been round for fifteen years or extra. That’s not nostalgia, it’s collected information, buyer relationships and experience that can’t be recreated in two years.

•     Enterprise prospects with long-term contracts. Gross sales cycles of 9 months or extra are a barrier to newcomers. For the incumbent, they type a moat.

•     Untapped demand. In just about each firm we purchase, current prospects ask for greater than is at present being delivered. That’s the least costly progress out there.

•     Succession alternative. Many of those companies have been constructed by founders who’re prepared for the following chapter. We offer continuity for the corporate and its workforce, together with a structured transition.

Inside the Group, we’re constructing a portfolio whose elements reinforce each other: an data platform with broad attain and years of Dutch open and company information; an information administration firm that builds infrastructure and information high quality practices for enterprise purchasers; segmented family and deal with information utilized by public authorities and market contributors; and a property and valuation information layer at present underneath growth.

The technique has two phases. Part one is the linked information layer: turning separate information sources right into a single, coherent basis for information and determination intelligence. Part two interprets that basis into industry-specific, data- and AI-driven workflow options for patrons. Part one with out section two is infrastructure with out margin. Part two with out section one is precisely the error the remainder of the market is making at the moment.

We finance this buy-and-build technique with a mixture of fairness and bond capital. For traders in search of publicity to the information economic system with out the risk-return profile of enterprise capital, we provide bond concern collection 2026-III.

Key phrases

The curiosity isn’t funded by projections, however by the money move of firms already invoicing prospects at the moment. That’s the elementary distinction from investing within the AI hype: our underlying companies earn cash from the work that must be executed earlier than AI can create worth anyplace.

Greater than forty traders at the moment are investing alongside us.

The market at present pays an awesome deal for the mannequin and little or no for the information. That’s precisely the type of asymmetry from which returns are generated. We’re shopping for the undervalued half: the experience, infrastructure and datasets on which each AI software finally relies upon.

Why we don’t spend money on AI guarantees, however within the layer beneath them, and why our bondholders profit.

Stroll into ten Dutch boardrooms at the moment and 9 of them can have AI on the agenda. Copilots, brokers, generative fashions in customer support. The ambition is there, and infrequently the funds too.

What’s often lacking is the muse able to supporting that ambition.

The information is scattered throughout techniques that have been by no means designed to work collectively. Definitions range by division, “buyer” means one thing completely different in gross sales than in finance. Historic information is incomplete or can’t be reproduced. Possession has not been assigned. Knowledge high quality is measured when an issue happens, not as a matter of routine. And governance exists on paper, not in follow.

Then the mannequin arrives. And the mannequin does what it at all times does: it amplifies no matter you place into it.

I’ve been working with information for greater than twenty years. Each mannequin I’ve seen, statistical, machine-learning or generative — is ruled by the identical rule:

A mediocre mannequin educated on glorious information outperforms a superb mannequin educated on mediocre information. Virtually each time. That’s not an opinion; it’s the expertise of anybody who has ever needed to clarify a manufacturing mannequin to a shopper who didn’t belief the reply.

The implications of this are broadly underestimated. Fashions have develop into a commodity: they’re out there to everybody, at falling prices and with narrowing variations between them. What isn’t interchangeable is the information layer beneath, the infrastructure, high quality, definitions, historical past and possession.

That’s the place shortage lies. That’s the place the defensible place is constructed.

In lots of organisations, information administration sits underneath IT operations. It’s handled as a value line, not as worth creation. That historic mistake is now turning into painfully seen.

Three causes we deal with it as an asset:

1. It can’t be copied. A competitor can purchase the identical mannequin tomorrow. What they can not purchase tomorrow is fifteen years of fresh, linked and documented information historical past.

2. It units the ceiling for each AI funding. Skip the information layer and also you pay twice: first for the venture that fails, then for the muse you continue to should construct.

3. It has develop into a governance crucial. Traceability, explainability and demonstrable high quality, as soon as thought of good follow, at the moment are stipulations for working in at the moment’s regulatory surroundings.

Knowledge administration isn’t merely a prerequisite for the AI technique. It’s the AI technique — in the one space you really management.

Liplyn Group doesn’t spend money on AI guarantees or in start-ups which have but to show there’s a market. We spend money on established Dutch information firms: companies producing income at the moment, with prospects who’ve stayed for years, and with providers or merchandise in information administration, information infrastructure, information high quality and determination intelligence.

The profile we search for is constant:

•     Confirmed observe file. A number of of our working firms have been round for fifteen years or extra. That’s not nostalgia, it’s collected information, buyer relationships and experience that can’t be recreated in two years.

•     Enterprise prospects with long-term contracts. Gross sales cycles of 9 months or extra are a barrier to newcomers. For the incumbent, they type a moat.

•     Untapped demand. In just about each firm we purchase, current prospects ask for greater than is at present being delivered. That’s the least costly progress out there.

•     Succession alternative. Many of those companies have been constructed by founders who’re prepared for the following chapter. We offer continuity for the corporate and its workforce, together with a structured transition.

Inside the Group, we’re constructing a portfolio whose elements reinforce each other: an data platform with broad attain and years of Dutch open and company information; an information administration firm that builds infrastructure and information high quality practices for enterprise purchasers; segmented family and deal with information utilized by public authorities and market contributors; and a property and valuation information layer at present underneath growth.

The technique has two phases. Part one is the linked information layer: turning separate information sources right into a single, coherent basis for information and determination intelligence. Part two interprets that basis into industry-specific, data- and AI-driven workflow options for patrons. Part one with out section two is infrastructure with out margin. Part two with out section one is precisely the error the remainder of the market is making at the moment.

We finance this buy-and-build technique with a mixture of fairness and bond capital. For traders in search of publicity to the information economic system with out the risk-return profile of enterprise capital, we provide bond concern collection 2026-III.

Key phrases

The curiosity isn’t funded by projections, however by the money move of firms already invoicing prospects at the moment. That’s the elementary distinction from investing within the AI hype: our underlying companies earn cash from the work that must be executed earlier than AI can create worth anyplace.

Greater than forty traders at the moment are investing alongside us.

The market at present pays an awesome deal for the mannequin and little or no for the information. That’s precisely the type of asymmetry from which returns are generated. We’re shopping for the undervalued half: the experience, infrastructure and datasets on which each AI software finally relies upon.