Multi-Year Is a Switching Cost, Not a Discount

Harvey

L2

Harvey's multi-year contract structure is not a pricing mechanism. It is the formalization of an operational reality that takes 12–18 months to build. The switching cost is not contractual. It is operational. Harvey's Vault product (storing up to 10,000 documents per project) and its Workflow Builder allow firms to co-create proprietary workflows for specific practice areas: M&A due diligence checklists, regulatory compliance templates, litigation research pipelines, fund formation documents. By 2025, Harvey had 25,000+ custom agents operating on its platform, generating 400,000+ daily agentic queries. Each of those agents was built by or with a specific client to handle that client's specific work patterns. The math for switching is straightforward: rebuilding a library of practice-area-specific agents at a competitor requires months of Legal Engineer time, months of retraining, and months of productivity degradation while the new library reaches maturity. At $1,200/lawyer/ month for Harvey's baseline pricing, the cost of that degradation period typically exceeds the annual contract value. No rational law firm finance committee approves a switch once they run that calculation. Harvey structures multi-year offers with 3–8% annual price escalators — reasonable enough to not trigger resistance, but compounding over a 3-year contract to a meaningful revenue increase. The escalator is offered as part of the multi-year deal, not as a post-renewal renegotiation. This means Harvey locks price increases in before the client's team has done the internal math on switching costs — at which point they would not need to offer an escalator at all. The 98% gross retention Harvey achieved by end of 2024 is not a feature of Harvey's contract terms. It is a feature of the operational depth built into every account. The contract documents what is already economically true. Weinberg's architectural vision for the platform makes the multi-year dynamic even stickier over time: Harvey is explicitly building "document-based persistent memory" and "multi-tenant collaboration across organizations" — capabilities that would tie Harvey further into a firm's matter-management infrastructure and inter-firm workflows. Each new capability layer deepens the switching cost without requiring a new negotiation.

Decagon

L3

Decagon's switching cost is encoded in Agent Operating Procedures — the business logic layer that sits between the AI and the enterprise's operations. Each AOP is a documented procedure: when a customer asks about X, do Y, escalate to Z if condition A. AOPs encode institutional knowledge that typically exists only in the heads of experienced support managers — the edge cases, the exceptions, the exact routing rules that have been refined through thousands of real customer interactions. Rippling deployed 75+ custom AOPs across 12+ product lines. Notion ran a formal five-vendor RFP and chose Decagon; post-deployment, Notion's AOPs represent a operational knowledge base that no competing vendor has access to. The commercial insight is about timing. Decagon's 4-week pilot structure means pricing is committed before the full operational depth is visible. By the time a customer has 70+ AOPs configured and thousands of ticket interactions accumulated, they have already signed. The offer — "we'll structure this as a multi-year deal" — has its maximum commercial impact at the moment the pilot completes and the switching cost is beginning to accumulate, not 18 months later when it is already prohibitive. Jesse Zhang's pricing framing makes the multi-year dynamic explicit from a different angle: "Human labor is generally like an order of magnitude larger than software spend." (Zhang, 20VC, September 2025.) When Decagon replaces labor, it is not competing with a software budget that has annual review cycles. It is replacing a headcount allocation. Headcount decisions — and their reversals — are not annual contract terms. They are structural organizational changes. Multi-year contracts align Decagon's revenue model with the organizational reality of the decision it is replacing. The switching cost in Decagon's case is also competitive intelligence. Every conversation that flows through Decagon's platform — resolved or escalated — contributes to a customer-specific dataset of interaction patterns, successful resolution paths, and escalation triggers. A competitor starting from scratch would not have access to that accumulated intelligence. The customer who switches loses their own institutional memory of their customers' behavior, encoded in Decagon's AOP library.

Glean

L2

Glean's switching cost is the permissions layer — and it took three to four years to build. The technical moat is not search quality or LLM integration. It is the infrastructure that enforces each user's exact access rights across 100+ enterprise SaaS applications in real time. When Glean indexes Slack, Confluence, Jira, Salesforce, and GitHub, it does not just index the content — it maps the permission hierarchy so that each employee sees exactly what they are authorized to see and nothing else. This is the feature that enterprise CISOs demand and that competitors without years of production deployments cannot replicate quickly. Once a company-wide deployment is live, the switching cost has three components: First, re-indexing. A company with 100+ connected data sources, 5,000+ employees, and three years of indexed organizational knowledge would need months of engineering work to replicate Glean's knowledge graph at a competitor — assuming a competitor could even match the 100+ connectors. Second, behavioral dependency. The user behavior signal — "If I can't find it on Glean, it doesn't exist" — is documented in Glean's own customer evidence. At 40% DAU/MAU (2x the SaaS industry benchmark) and 5 queries per user per day (on par with Google consumer search), Glean has crossed from productivity tool into daily workflow dependency. That is not a switching cost in the legal sense; it is a switching cost in the behavioral sense that is arguably harder to overcome. Third, the Enterprise and Personal Graph accumulation. Arvind Jain's two-graph architecture — the Enterprise Graph (organization-wide knowledge with permission-aware access) and the Personal Graph (each employee's individual knowledge and activity patterns) — means that Glean's value compounds with time. A three-year-old Glean deployment is not equivalent to a one-year-old deployment. The graphs are richer, the personalization is deeper, and the switching cost is proportionally higher. Glean's pricing reflects this: contracts span 1–3 years with no sub-annual agreements. The $1M+ contract segment grew 3x in one fiscal year. Company-wide deployments doubled year-over-year. These are expansion economics driven by operational depth, not by new feature releases or marketing campaigns.

Gong

L2

Gong's switching cost story is instructive as both a positive case and a cautionary one — because the 2023 stall revealed a structural gap in how the switching cost was built. The positive case: Gong accumulated 10B+ sales interactions across its customer base. Each customer's deployment generates a unique institutional record — recorded calls, coaching scorecards, deal outcomes, behavioral patterns for each rep. Amit Bendov's founding vision was never transcription; it was structured data from conversations. "The vision was never about transcription... I wanted something that would take all the information and translate it into structured data." (Bendov, Sequoia Training Data podcast, 2025.) By 2025, that data was powering 18 AI agents, Ask Anything queries, AI Briefers that produced 19% win rate improvements, and Gong Orchestrate — a platform for defining and executing GTM plays. A Gong customer with 3 years of call recordings, trained coaching scorecards, and defined GTM plays encoded in Gong Orchestrate has a switching cost that is primarily informational: losing Gong means losing the institutional memory of how their sales team has operated. That record does not transfer to Salesforce, HubSpot, or any CRM that could theoretically replace Gong's functionality. The cautionary case: Gong's initial contract structure was seat-based annual. When sales hiring froze across SaaS in 2023, NDR degraded from ~140% to ~16% YoY growth — not because customers churned, but because their teams contracted. The seat model tied Gong's revenue to the customer's headcount decisions, not to the value Gong was delivering. Multi-year contracts with value-based components — not seat-only — would have partially insulated Gong from this structural exposure. Bendov's public commitment by 2025: "We're going to be moving to a value-based model. Today we're mostly on a seat-based model... if AI saves 70 percent of seller time, and you're charging per seat, you're basically giving away the majority of the value." The lesson: multi-year contracts that capture both time-lock and value-based pricing protect against the switching cost being undermined by external factors (headcount freezes, organizational restructuring) that are not within the customer's control.