The Second Business

The product is not the software

The gap is easy to underestimate because most of the new work does not improve the original internal workflow. Multi-tenant security does not make an employee finish the task faster. Contracting does not improve model quality. Customer onboarding does not make the feature more elegant.

They are still part of the product once somebody outside the company pays for it.

The internal users may also share terminology, processes, data structures and institutional knowledge that the application quietly assumes. The apparent product is often software plus a large amount of invisible organizational context.

Commercialization is the work of making enough of that context explicit, dependable and economically deliverable that the capability can travel.

Choose the business you are actually building

Not every externally valuable capability should become SaaS. The same underlying expertise can become software, a maintained knowledge product, or a technology-enabled service. Those models can all be good businesses. They simply scale differently.

This distinction matters especially in expert businesses. aosphere, created by Allen & Overy in 2002, became a subscription platform for legal and compliance information. When outside investors partnered with the firm in 2023, aosphere described itself as a profitable business serving more than 725 blue-chip clients.³

Linklaters takes a different approach with CreateiQ, which the firm describes as a “tech company housed in Linklaters” focused on capital-markets contracting.⁴ The organizational form is different because the underlying operating model is different.

The first commercialization decision is not whether to spin something out. It is what recurring obligation you are willing to own.

Internal adoption is not market validation

Internal software has something commercial products do not: captive distribution. Employees may be expected to use it. The company controls the surrounding process. The person benefiting from the product and the person funding it can sit inside the same planning cycle.

The external market is less forgiving. A prospective customer can keep doing the work manually, buy an incumbent, wait for a platform vendor to add the capability, or increasingly build a credible first version itself.

So the external test is not whether another company likes the demo. It is whether a specific buyer will allocate a specific budget to solve the problem at a price capable of supporting the business behind the product.

“Large market” has now become a discussion about penetration, competitors, sales cycles, implementation capacity and why one in every five viable organizations would choose this product. That is a much more useful strategy conversation.

Seven conditions have to become true

There is no universal ARR threshold that makes commercialization sensible. There are, however, conditions that a sustainable external business eventually has to satisfy.

Companies usually do not disagree with this list. The risk is that they hide the requirements inside optimistic shorthand.

“We already know the customers” may mean partners can make introductions. That is not yet a repeatable sales motion. “The technology is built” may mean the core workflow works. That is not the same as external identity, observability, configuration, documentation, support and security review.

“Our experts are the moat” may also be true. But if every implementation requires ten hours from a scarce partner, that expertise is simultaneously differentiation and a constraint on scale.

A spinout is an organizational choice, not a business model

Some of the strongest examples of an internal capability becoming an external business use very different structures.

BlackRock's Aladdin began as technology at the center of the firm's own investment process and became a platform for outside institutions. In 2025, BlackRock reported $2.0 billion of technology services and subscription revenue, with Aladdin representing the majority; organic annual contract value grew 16% excluding Preqin.¹

Arcesium went the other direction structurally. Its history traces back to technology and operating capabilities at the D. E. Shaw group, with Blackstone Alternative Asset Management as seed investor and anchor client when Arcesium launched independently in 2015. Arcesium reports that it began with 436 employees and $152 billion in gross AUM and today supports $6.4 trillion with more than 2,400 employees.²

aosphere eventually moved toward standalone ownership with outside investment.³ CreateiQ remains explicitly housed inside Linklaters.⁴

The lesson is not that one structure wins. A separate company can create clearer capital allocation, compensation, governance, customer neutrality or access to outside investment. Keeping the product inside the parent can preserve distribution, expertise and shared infrastructure.

The better questions are operational: Who owns the roadmap? Who owns the P&L? Who chooses between an internal request and an external customer requirement? Who funds a security feature that matters to customers but does nothing for the core business? Who carries the sales target? Who can stop the product?

If those decisions continually return to the economics and priorities of the parent, the software may have customers while still operating like a project.

What does the second business have to earn?

Commercialization discussions get abstract once revenue enters the room. I prefer to make the P&L deliberately simple first, then add complexity only when it changes the decision.

Start with a hypothetical workflow product sold to regulated enterprises. At 40 customers paying $100,000 annually, the business produces $4.0 million in revenue. At a 75% gross margin, that leaves $3.0 million after hosting, model usage, direct implementation and support. If product and engineering, sales and marketing, administration and security total $3.7 million, the operating result is a $700,000 loss.

The useful part of the model is not the base case. It is how quickly the decision changes when the assumptions move.

A longer sales cycle delays cash and may force hiring before revenue appears. Lower pricing increases the required customer count. More implementation work can reduce gross margin while simultaneously limiting how many customers the team can onboard. Enterprise payment terms can create a different cash profile from recognized revenue.

Then there is the parent-company subsidy.

Suppose the product reports a $700,000 operating loss but receives another $600,000 of security, legal, executive, infrastructure and specialist time from the parent without charging it to the product. The cash view says the business lost $700,000. The economic view says supporting it consumed $1.3 million.

Shared capabilities can be one of an incumbent's greatest advantages over a startup. They become dangerous only when they disappear from the economics.

Buy evidence before certainty

The commercialization decision is often framed as binary: are we doing this or not? I think that is the wrong unit of decision.

Early investment should buy evidence that allows the next investment to be made with less uncertainty.

That sequencing changes the quality of the conversation. Independent discovery tests whether the problem travels. A design partner tests whether the workflow travels. A paid deployment tests whether budget exists. A second and third implementation test whether delivery can repeat. Renewal tests whether value remains after novelty disappears.

It also makes stopping less dramatic. A company has not “failed to become a software company” because an external design partner exposes a workflow that does not generalize. It bought useful evidence before funding the expensive version of the assumption.

Traction is not permanence

A software business embedded inside a larger company competes for two markets at once: customers outside the company and capital inside it.

Meta's Workplace is a useful example. The product grew out of software used internally at Facebook and reached millions of paid subscribers. Meta later decided to discontinue it; customers had normal access through August 2025 and the product fully shut down on May 31, 2026.⁵

That does not mean Workplace had no value. It means external adoption did not eliminate the need for continued strategic commitment from the parent.

GE's digital strategy illustrates a different problem. In 2018, GE announced that it would assemble Predix and other industrial software assets into a separately operated business starting with roughly $1.2 billion in annual software revenue.⁶ GE's 2018 filing also reported $3.9 billion of broader digital revenue while explaining that those revenues and associated costs were largely generated inside its operating businesses.⁷

The distinction is important: software attached to equipment or services, internal consumption, external recurring revenue and standalone profitability can all be valuable measures. They are not interchangeable measures.

AI raises the bar on differentiation

Generative AI is lowering the cost of creating useful internal software. More companies will therefore find themselves asking whether an internal capability should be commercialized.

Lower creation cost does not remove the commercialization gap. In some respects it widens it.

A production AI product may need continuous evaluation across changing models, controls around sensitive data, model-routing and cost management, auditability, human escalation, observability and a process for failures that are not deterministic. At the same time, the underlying model capability may become easier for competitors to reproduce.

For an AI product, “we built this” is especially weak evidence of durable differentiation.

The strongest asset may still be the parent company: domain expertise, customer relationships, proprietary data, a demanding internal production environment and a trusted brand. But each advantage has to cross the organizational boundary. Expertise that cannot be encoded or delivered economically may remain a service advantage. Relationships that produce introductions but not purchases are not distribution.

The question is always the same: can the advantage travel?

What leadership should be able to answer

Before calling an internal tool a software business, I would want leadership to answer the following questions with evidence rather than adjectives.

Those answers do not automatically produce a decision to commercialize. They produce something more useful: a decision that understands what commercialization actually means.