Small AI Models, Not External LLMs, Are Keeping DocuSign's Costs Down
On its earnings call on September 3, 2026, DocuSign claimed that its Intelligent Agreement Management (IAM) platform processes workloads at "significantly lower marginal costs" than services that route to external language models. At the same time, IAM is growing fast: 15.1 percent of the company's total ARR, up from 12.6 percent the previous quarter. But the cost claim so far rests only on the company's own statements — no independent benchmarking or concrete cost figures have been provided.
What the company says
On DocuSign's earnings call for the second quarter of fiscal year 2027, held on September 3, 2026, CEO Allan Thygesen introduced the cost argument as part of a broader platform narrative. According to a transcript of the call published by The Motley Fool on Yahoo Finance (see the transcript), Thygesen said: "As we've outlined in a new series of blog posts, IAM's AI-native architecture processes workloads at significantly lower marginal costs than offerings that route to external LLMs."
Thygesen thus tied the claim to blog posts the company itself has published, and described the architecture as "AI-native." Beyond this, there are no independent measurements of the cost difference, no information about which models IAM uses, and no specific savings figures. The claim is, for now, a company statement — not a verified fact.
The numbers behind the growth
According to the transcript, Thygesen explained the growth of the IAM platform by pointing to the success of the company's platform strategy: "IAM now accounts for 15.1 percent of total ARR, up from 12.6 percent in Q1" (see the transcript). The increase of 2.5 percentage points in a single quarter is the clearest documented evidence that the platform push is gaining commercial traction.
The company also stated that customers have uploaded more than 300 million documents through IAM's Agreement Manager. These are company-reported key figures without independent verification, but they give an indication of the platform's scale — and of why marginal cost per document is a relevant topic at all: at such volumes, small differences in cost per processed document become measurable on the bottom line.
The other financial figures from the quarter, also company-reported, were according to the transcript:
- Revenue: $876 million, up 9 percent from the same quarter a year earlier
- Operating margin: 32 percent
- Free cash flow: roughly $300 million
It is worth noting that Thygesen argues the cost architecture is a competitive advantage, even as the company already delivers solid profitability with a 32 percent operating margin. Whether the claimed lower marginal costs are a driver behind today's margins, or a promise for the future, does not come through clearly in the transcript.
"One million contracts a day" — a headline, not a documented figure
On September 21, 2026, PYMNTS picked up the story with the headline "Docusign Processes 1 Million Contracts a Day on Small AI Models" — that is, that DocuSign processes one million contracts per day on small AI models (see PYMNTS). The small-models framing is consistent with Thygesen's cost argument from the earnings call.
The figure itself, however, is documented only at the headline level at PYMNTS. The available article text deals with ISO 42001 governance in the payments industry and does not substantiate the contract volume, and "one million per day" does not appear in the earnings call transcript itself. The volume figure should therefore be read as a secondary, unverified headline claim — not as a fact — until DocuSign confirms it in its own blog posts or investor materials.
What "AI-native architecture" actually means here
The debate DocuSign is positioning itself in concerns a well-known industry choice: either build task-specific, smaller models in-house, or route work to large, external language models (LLMs). The external models are flexible, but you pay per token, and costs can become significant at large workloads. In-house, narrower models can in theory deliver "significantly lower marginal costs" at high scale — but they require investment in development, operations and maintenance, and may offer less flexibility.
DocuSign thus claims it has chosen the first path for IAM, and that it pays off. This is analysis, not a documented fact: for structured, repeatable tasks in contract processing — classification, field extraction, validation against known templates — a large general-purpose model is rarely needed, and a smaller specialized model can therefore be delivered more cheaply per unit. But this is general industry logic, not a description of what DocuSign has concretely built. What actually lies behind the company's "AI-native" architecture — model choices, size, training data and the size of the savings — is not specified in the available sources.
What the evidence actually supports — and what it doesn't
It is useful to distinguish between three categories:
Quarterly figures from the call (company-reported): Revenue, operating margin, free cash flow and IAM's share of ARR. These come directly from the earnings call and are documented in the transcript, but they remain self-reported figures without independent confirmation in the available material.
Document volume (company claim): That customers have uploaded more than 300 million documents through Agreement Manager is a claim from the company itself without external verification in the available material.
The cost claim (pure company statement): "Significantly lower marginal costs" is not supported by figures, benchmarks or technical details in any of the sources. It is a direction the claim points in, not a measurement. The same applies to PYMNTS's headline about one million contracts per day.
Questions that remain
Several key questions are unanswered in the available material:
- How many contracts are processed daily? PYMNTS's headline says one million per day, but the figure is not confirmed by DocuSign's own materials in the sources.
- Which models does IAM use? No names, sizes or architecture have been disclosed.
- How large are the savings? "Significantly lower" is not a number. Without quantification, it is impossible to assess the economic effect.
- How does the cost advantage affect the business model? The transcript does not provide a basis for saying how the claimed cost advantage is further leveraged, for example in the pricing of the services.
- How fixed is "AI-native"? Whether the term means in-house models, fine-tuned open models, or a hybrid with external LLMs for certain tasks is unclear.
DocuSign itself highlights a series of blog posts that are meant to explain the architecture. For a full assessment of the cost claim, these posts, together with independent analyses, will be essential reading. For now, the safest conclusion is that the company is reporting strong platform growth — and claims a cost advantage it has not yet publicly documented with measured figures.