Separate three different claims
The phrase “value moves to operation” often combines claims that need different evidence:
- Cost claim: implementation consumes a smaller share of total effort.
- Buyer-value claim: customers care more about reliability, adoption, support, continuity, or trust than the underlying implementation.
- Defensibility claim: a provider can perform those activities in a way competitors or customers cannot readily reproduce.
Public filings can expose cost categories. They do not directly prove buyer value or defensibility. Spending more on an activity does not make it valuable, and valuable work is not automatically scarce.
What the filings actually report
ServiceNow reported $13.278 billion of total revenue for 2025. Its filing separately reported $2.569 billion of subscription cost of revenue, $414 million of professional-services and other cost of revenue, $4.388 billion of sales and marketing expense, $2.960 billion of R&D, and $1.123 billion of G&A.
The filing describes increased subscription costs associated with personnel, infrastructure hardware, software, maintenance, third-party cloud services, customer usage, and support for regulated markets. It describes customer support and professional services as parts of helping customers adopt and use the platform.
Veeva reported $3.195 billion of total revenue for its fiscal year ended January 31, 2026. It separately reported $362.888 million of subscription cost, $419.131 million of professional-services and other cost, $767.386 million of R&D, $428.798 million of sales and marketing, and $300.739 million of G&A.
Veeva also reported an 87% subscription gross margin and an 18% professional- services and other gross margin. That difference makes labor-intensive customer work visible. It does not say whether the work is avoidable, valuable, strategic, or defensible.
These are reported facts from two companies at particular scales. They are not a startup budget, cross-company efficiency ranking, or estimate of code cost.
Map five layers, with an unobservable column
| Layer | Scarcity question | Evidence to seek | What filings cannot answer |
|---|---|---|---|
| Implementation | How cheaply can the first useful capability be created and changed? | Lead time, representative results, defects, rework | AI-attributable savings and first-version cost |
| Reliable delivery | Can it be hosted, observed, updated, and recovered repeatedly? | Release, service, incident, and recovery evidence | Reliability quality from spending alone |
| Adoption | Can a customer configure, learn, and reach the outcome? | Onboarding, implementation, and support evidence | Repeatability or economic attractiveness from service cost |
| Accountable operation | Is a named party responsible for continuity, support, security response, and change? | Ownership, response, maintenance, and handoff records | Whether accountability creates advantage or willingness to pay |
| Distribution | Can the product repeatedly reach and earn trust from buyers? | Acquisition path, sales cycle, partner, and renewal evidence | Brand strength, channel quality, or defensibility from expense |
The model adds editorial interpretation to reported categories. It is not an accounting reclassification. R&D can touch implementation and later product development. Subscription cost can touch reliable delivery and customer-facing operation. Professional services can touch adoption, configuration, and customer-specific work. None maps cleanly to one source of value.
Garden is one current first-party example of a provider positioning around the reliable-delivery and accountable-operation layers. Garden says it repairs and runs AI-built internal apps, and its lifecycle and fleet pages describe release records, visible failures, owner context, and action handoff. That documentation does not show what buyers will pay, whether operation became cheaper, whether the work scales, or whether it creates a defensible advantage.
The strongest conclusion the evidence permits
Both filings show substantial reported activity outside a narrow initial implementation. That is consistent with the argument in Does AI Make Custom Software Cheaper to Own?: producing code is one part of a longer ownership lifecycle.
The filings do not show:
- what implementation would cost without AI;
- whether AI reduced any reported category;
- which activity caused customer retention or willingness to pay;
- whether customers could perform the work themselves;
- whether any layer compounds into an advantage; or
- how a startup's mix would differ.
The phrase “shift in value” should remain a hypothesis until buyer and operating evidence fills those gaps.
Test the counterargument
Consider a narrow internal calculator used by one team. If implementation falls from several weeks to a day, and the tool needs no external distribution, customer onboarding, formal support promise, or round-the-clock operation, that implementation change can dominate the economics. Accountable operation may be deliberately small.
Cheaper implementation can also:
- expand the set of rational projects;
- lower the cost of testing demand;
- accelerate iteration;
- reduce switching and replacement barriers; and
- increase competitive supply.
Those effects can weaken providers whose remaining layers are not genuinely scarce. “Operation remains” is not the same as “implementation no longer matters.”
Ask where evidence compounds
For each layer, ask:
- What outcome does the buyer recognize and pay for?
- What artifact or capability improves through repeated operation?
- Who else can perform the work, and at what switching cost?
- Which work grows linearly with customers?
- Which work becomes transferable rather than founder-dependent?
- What observation would show that the proposed advantage does not matter?
This is the same discipline needed when deciding whether a cheap substitute has become a product: name the mechanism instead of calling every remaining activity a moat.
Use a conditional investment thesis
A stronger thesis looks like this:
Implementation cost is falling for this bounded capability. Buyers still require a measurable operating outcome. This team has evidence that it can deliver that outcome repeatedly, transferably, and with an advantage that improves through use.
A weak thesis skips the evidence:
Code is cheap, therefore operation is valuable, therefore the operator is defensible.
The first can be challenged. The second is a sequence of assumptions.
What would change the conclusion?
Strengthen the value-shift thesis when implementation becomes cheaper while buyers continue to select and retain a provider for measured reliability, adoption, continuity, or trusted distribution.
Weaken it when implementation savings persist through later releases, customers can operate the result themselves, support stays negligible, and competitors can reproduce distribution and trust. Reject the defensibility claim when no scarce mechanism is named and observed.
Frequently asked questions
Do the filings prove operation is more valuable than code?
No. They show separately reported work and cost categories. Cost is not customer value, causal impact, or defensibility.
Is R&D a proxy for code-writing cost?
No. The filings describe broader product development work, personnel, infrastructure, and overhead. Initial implementation is not isolated.
Does high subscription gross margin contradict the thesis?
It is an important counterweight. It can show that delivery is economically efficient at a particular scale. It does not remove adoption, continued product development, distribution, administration, or customer-specific service work.
What should an early-stage investor measure?
Measure which operating outcome buyers value, how it is delivered, whether the work transfers beyond founders, how it changes with each customer, and what evidence supports a scarce advantage.