Software Ate the World. Services Are Eating Software.

Enterprise software used to come with an invisible second bill. For every dollar paid to the vendor, roughly six more were spent on the people, systems, and work required to make the vendor’s product actually do something useful. That six to one ratio was the tax nobody questioned, because until this year there was no alternative. The software did not cost what the invoice said. It cost what the invoice plus the implementation partner plus the configuration team plus the ongoing internal headcount said, and that total was never reconciled on a single page.

AI just forced the reconciliation. This piece is about what enterprises are doing with the number now that they can see it, which public moves confirm the shift is structural, and how to tell which SaaS contracts on your own stack are worth renewing in 2026 and which ones are not.

Why $285 billion in SaaS value disappeared in a week

In early February, public software companies lost roughly $285 billion in value inside seven trading days. The S&P North American software index fell below the broader market’s forward earnings multiple for the first time in its history. Atlassian reported the first enterprise seat count decline in its company history. Workday cut 8.5 percent of its own workforce. Intuit fell 46 percent from its peak.

The trigger was not a recession or an earnings miss. Investors realized the per seat licensing model was built on an assumption that no longer holds: that every new hire needs a login. When one engineer plus a foundation model does the work of three, enterprises do not buy three seats. Revenue per customer falls. Growth slows. The entire sector was priced for a curve that has now bent.

The six dollar ratio and what it actually covers

In March, a Sequoia partner published a number that reframes how to read any enterprise software contract. For every dollar spent on software, roughly six are spent on services around the software.

That six dollars is not consulting in the abstract. It is specific line items most finance teams track separately:

• Implementation paid to the vendor or a certified partner to stand the product up.
• Configuration paid on an ongoing basis to extend the product to the business’s actual workflows.
• Integration paid to connect the product to every other system it touches.
• Internal headcount whose calendars are dominated by the product even if their titles do not mention it.
• Managed services retainers covering everything the vendor explicitly does not.

Aggregate the five, divide by the software line, and most enterprise stacks land between five to one and seven to one. AI did not attack the software. It attacked the five items above, because those five items are where AI is actually good, and they are where the margin was the whole time.

Three moves in thirty days that confirm where the value is going

Three public events inside the last month tell you which layer of the stack captures value from here.

Salesforce is dismantling its own interface. On April 15, Salesforce announced what it called the most ambitious architectural transformation in its twenty seven year history. Every capability on the platform is now exposed as an API or an agent tool. The co founder framed it directly: why should you ever log into Salesforce again. The CRM leader is voluntarily declaring its user interface optional, because the company can read its own stock chart and knows the margin is no longer going to sit at the interface layer.

A model provider just shipped the application layer its partner was building. On April 14, Anthropic’s chief product officer resigned from Figma’s board. Three days later, Anthropic launched a design product that competes directly with Figma’s core offering. Figma’s stock fell nine percent intraday. The company’s valuation, above $60 billion at its IPO peak, is now around $10 billion. Figma spent the last year integrating Anthropic into its product and ended the quarter discovering that its model partner had become its product competitor.

Revenue is compounding at the bottom of the stack. On April 7, Anthropic disclosed $30 billion in annualized revenue, up from $9 billion at the end of 2025. Enterprise customers paying over $1 million per year doubled from 500 to 1,000 in two months. Eight of the Fortune 10 are Claude customers. The money is moving to the foundation layer and being squeezed out of the one above it.

A system of record dismantling its UI, a foundation model shipping its customer’s product, revenue moving to the bottom and evaporating in the middle. That is not a cycle. That is a structural reallocation of where enterprise software margin is allowed to sit.

What enterprises are actually doing, not saying

The calls we get do not start with we want to replace our SaaS. They start with something narrower, and the pattern repeats across industries.

A financial services firm asks us to audit a workflow tool its operations team has been extending for four years. When we map what has actually been built on top of the vendor’s platform, most of it is business logic the client wrote. The renewal is no longer paying for software. It is paying for permission to keep using logic the client already owns.

A fintech asks for a review of a reporting stack that costs six figures a year and dominates the analytics team’s calendar. The recommendation is rarely to switch vendors. It is to absorb three subscriptions and a managed services retainer into one custom interface against the data warehouse, and retire the coordination cost at the same time.

A media company asks about a document and contract stack spread across four vendors. The integration layer is the legal team itself, which means partner level hours have been absorbed into the stack as unpriced middleware. A unified custom build against a foundation model replaces the tools and recovers the hours simultaneously.

The replacement decision is almost never about the software. It is about everything around the software that the software never charged for explicitly and never could have.

The workflows worth replacing versus the ones worth keeping

Nothing here argues for replacing systems of record. The ERP stays. The core HRIS stays. Anything with a compliance moat, audit history, and regulatory certification is not a weekend project and should not be treated like one.

What moves is the layer above. In broad terms:

• Keep: ERP, core HRIS, regulated financial systems, anything with embedded audit trails and multi year certification.
• Audit hard: CRM extensions, marketing automation, internal BI, workflow automation, any tool where a separate services partner is on retainer.
• Strong replacement candidates: internal admin tools, middle office dashboards, reporting glue, point solutions bought to give a small team access to data that already lives in the warehouse.

The test is economic, not technical. When the license plus the services around it exceeds what it would cost to build and maintain the same capability directly, the contract is paying rent on something the company could own.

Three questions to put on the next renewal

These should be answered in writing before any auto renewal clears.

1. What percent of this contract value is the software, and what percent is implementation, configuration, and internal headcount to run it. If finance cannot produce the number, the ratio is worse than anyone thinks.
2. If this product were replaced with a custom build, what would actually be lost. The data usually lives in the warehouse. The workflow logic was written in house. The compliance surface is often the only genuinely moated piece. Name the moated piece explicitly.
3. Who would build the replacement. If the answer is nobody on staff, the follow up is who has shipped production systems against this category in the last year. That question separates enterprises that save money in 2026 from enterprises that sign a larger version of the same contract.

Where this leaves us

The argument is not that SaaS is over. The argument is that the economics of the stack just inverted. Software margin is compressing toward the foundation model cost. Services margin is inheriting what used to sit above it. The firms that capture value from here are the ones that can actually build at that layer, against a specific business, and hand back a system the client’s team can run.

Most firms pitching this work are not engineering firms. They will advise on the decision and bill you for the advising. The ones worth the meeting are the ones that build the system and train the team to maintain it.

That is the conversation we have with clients every week. If it is the conversation you are having this quarter, we should talk.