The enterprise CLM stack for legal teams running high-volume, high-complexity contract programs — where cycle time, redline accuracy, and post-signature obligation tracking all show up in the numbers. Four tools, four lifecycle phases, with named handoffs between them. All four moved in 2026: Ironclad shipped procurement agents and an MCP server, Litera relaunched around a single agent and renamed its metadata line, Docusign renamed the repository API this stack depends on, and Spellbook withdrew its published pricing — the last of which is the biggest change to how you budget the stack.
How the pieces fit
Ironclad is the CLM backbone and workflow engine. Every contract enters through Ironclad intake — a structured form capturing counterparty, contract type, risk tier, and required approvals before a clause is drafted. Ironclad routes the request to the right playbook, assigns the matter, and governs the approval chain. Post-signature it stores the executed agreement, tracks renewal dates and obligations, and fires expiry alerts. For programs processing 2,000–10,000+ contracts a year, this is where the operational discipline lives.
Two things changed here on 2026-08-05. Ironclad shipped AI Obligation Extraction, which turns renewal windows, discounts, credits, rebates, termination rights, and payment terms into tracked fields rather than prose someone has to re-read; a Contract Family Agent that assembles supplier-contract hierarchies without manual tagging; AI Redlining from Precedent; and an SAP integration sold as two packages, SAP Data Foundation and SAP Procurement Automation, connecting Ariba and S/4HANA. Separately, Ironclad now runs an official remote MCP server with OAuth, exposing roughly ten tools — create_obligation, get_obligation, list_obligations, list_obligation_types, update_obligation among them — to Claude, ChatGPT, Cursor, and other MCP clients. The practical effect is that the obligation layer is queryable by an agent without anyone building against the REST API first.
Spellbook handles AI-assisted drafting and first-pass review. The handoff from Ironclad is the intake trigger: when Ironclad routes a request to a drafting attorney, the attorney works with the relevant playbook positions as context. Spellbook flags deviations from playbook, offers pre-approved fallback language, and annotates risky provisions in plain language; the finished draft goes back into Ironclad for approval routing. The load-bearing property is that it reads the team’s own playbook rather than a generic legal corpus, so its suggestions reflect house positions instead of a model’s guess at market standard.
The framing to drop is “Word add-in”. The Word sidebar is still the daily surface, but Spellbook is sold as a platform of named pieces — Review, Draft, Compare, Ask, Playbooks, and Associate, an agent for multi-document drafting and review — and it reaches into email, Slack, and Salesforce. Connectors cover iManage, OneDrive, SharePoint, Google Drive, and Dropbox. Note what is absent: no public API and no MCP server, so Spellbook is the one layer in this stack you cannot automate against or address from an agent. Everything crossing that boundary crosses it as a human in a document.
Litera provides document quality control and metadata cleaning at the point of send. The handoff from Spellbook to Litera is the pre-send step. Litera Compare runs a final clause-level diff against the last-approved version to catch changes introduced during redline exchange. The metadata layer strips hidden track-changes, comment threads, and document history before a file reaches the counterparty — a real exposure on negotiated agreements, where prior positions sit inside the change history.
That metadata layer was renamed in 2026. The Metadact line is now Clean: Clean Desktop replaces Metadact Desktop, Clean Server replaces the on-premises server, Clean Cloud replaces Metadact inside Litera One, and Clean+ — a Litera-hosted service that removes the need to provision and patch your own cleaning server — reached general availability at the end of June 2026. Clean detects and removes 300+ metadata types across Word, PDF, Excel, and ZIP attachments in both classic and new Outlook. If your runbook still names Metadact, it names a product that no longer exists under that label.
Litera also relaunched as a company on 2026-07-15 around “one agent, one dataset”, selling the platform rather than the point products, with the Lito agent as the through-line across drafting, comparison, contract review, knowledge management, and business development. The next version of Lito is being shown across the Foundation and GrowthTech portfolio at ILTACON in Nashville, 23–27 August 2026. Treat that as positioning and a demo, not as capability you already own — what Lito does inside your tenant is a question for your Litera contract. The deterministic Compare and Clean engines, not the agent, are what this stack depends on at the send boundary.
Docusign IAM is the signing layer and agreement intelligence platform. The handoff from Ironclad is the execution trigger: once approval completes, IAM sends the execution package, tracks signing status, and returns the executed agreement to Ironclad’s repository by webhook.
The repository API changed name and grew up. What was the Navigator API is now the Agreement Manager API, generally available to eligible Agreement Manager customers since May 2026. Its Iris extraction returns provisions as structured fields — expiration_date, renewal_type (for example AUTO_RENEW), renewal_notice_date, total_agreement_value, effective_date, jurisdiction, termination_period_for_cause — rather than as text to be parsed. Bulk ingestion runs through POST /v1/accounts/{accountId}/upload/jobs at 10,000 documents per job and 100MB per document, with pre-signed URLs valid eight hours and a three-day job TTL; single retrieval is GET /v1/accounts/{accountId}/agreements/{agreementId}. Auth is OAuth 2.0. This is the piece that makes a back-file migration into the stack a scriptable job rather than a services engagement.
Where the tools now overlap
The old case for this combination was that each tool is a phase specialist with no duplication. That is less true than it was, and pretending otherwise leaves you paying twice.
Ironclad’s AI Redlining from Precedent lands on Spellbook’s territory, and Ironclad’s AI Obligation Extraction overlaps what Docusign’s Iris returns from the Agreement Manager API. Neither overlap is a reason to drop a tool by itself, but both are a reason to decide in writing which system is authoritative for each function before rollout. Two systems extracting renewal dates into two repositories is how a team ends up with two answers to “when does this auto-renew”.
The split that still holds: attorneys draft where they read, and they resist being pulled into a browser workflow to accept a redline, so Spellbook earns its seat on adoption rather than capability alone. Litera earns its place because the deterministic Compare-and-Clean step has no equivalent in the other three and is the step most often skipped in an all-in-one deployment. Use Ironclad’s redlining for supplier paper coming through procurement intake, where nobody negotiates in Word anyway, and keep Spellbook for negotiated commercial agreements.
Cost reality
Split this into what is published and what is not, because the two halves behave very differently in a budget cycle.
Published or observed:
- Ironclad: Vendr’s tracked purchases put the median annual contract value at $40,000, with an observed range of $15,000 to $104,272 and average negotiated savings of 20.6%. The six-figure-plus bands that circulate for Ironclad sit above almost everything Vendr has actually recorded. For a 50–150 user program, plan against the top of that observed range rather than the median, and treat a quote materially above ~$105,000 as something to push back on with comparables.
- Docusign IAM: now a published ladder. Starter is $45/user/month annually with 100 envelope sends per user per year and one workflow; Standard is $50/user/month with a three-user minimum, unlimited envelopes through the web app, and three workflows; Professional is $80/user/month with a three-user minimum and ten workflows. Every IAM tier caps at 50 users. That cap, not the seat price, is what puts this stack into a custom quote — a 50-user program at the top published tier is roughly $48,000/year, and anything above 50 users leaves the ladder entirely.
Quote-gated, and one of them unquotable:
- Spellbook: no published price. Spellbook used to list Solo at $99 and Team at $179 per user per month; those tiers are gone, and the pricing page now shows only Law Firms and In-House Teams with no dollar figure and a note that price is set by licensed headcount. Third-party trackers put the entry seat anywhere from about $20 to $99 and enterprise from about $199 to $350, with a reported ten-seat and six-month minimum. A 5x spread on the entry tier is not a price band — it is evidence that nobody outside Spellbook’s sales team knows. Do not put a per-seat Spellbook number in a budget you will be held to; get the quote first, and get it early, because it is the line item with the widest variance in the stack.
- Litera: quote-only, sold as bundled enterprise contracts with the effective per-user rate depending on which modules are in the bundle. Compare plus Clean is the minimum this stack needs; Foundation and Kira are separate decisions.
Implementation and carry. Ironclad configuration, Spellbook playbook build, and Litera deployment together are a year-one project, not a purchase. Compare deploys in days; Ironclad workflow configuration and playbook development run months. The recurring cost people forget is playbook maintenance — budget roughly one attorney-quarter per year to keep positions current, because both Spellbook and Ironclad’s precedent redlining amplify whatever is in the playbook, including the parts that went stale.
Match rules
This stack is the right pick when:
- Contract volume exceeds 2,000 agreements per year with genuine legal complexity — negotiated commercial agreements, not click-wraps
- The in-house team has 20+ attorneys, so per-attorney productivity gain covers per-seat cost
- You have IT capacity to own CLM workflow configuration, SSO/SCIM provisioning, and the Ironclad-to-Docusign webhook integration
- Outside-counsel spend is high enough that shortening drafting cycles moves legal spend measurably
- The GC or CLO will fund 6–12 months of playbook development before the stack runs at full efficiency
This stack is the wrong pick when:
- Contract volume is under 1,000/year — Ironclad is hard to justify below that; Juro or Concord covers the use case for less
- The team is under 10 attorneys, where per-seat AI cost outruns the benefit
- Playbooks do not exist yet — the AI layers amplify a playbook, and there is nothing to amplify
- Contracts are overwhelmingly standard-form with minimal negotiation, where sign-and-store is enough
- You are deliberately avoiding a CLM platform after a failed or deferred rollout — see the no-platform path instead, which is a different design, not a cheaper version of this one
Common variations
Swap Docusign IAM for native Ironclad signing. Ironclad includes an eSignature layer. Where the Agreement Manager API’s structured provision extraction is not needed — because Ironclad’s own obligation fields already cover it — running signing inside Ironclad removes a vendor and an integration. Rule for the swap: take it when annual envelope volume is under 5,000 and Ironclad’s repository fields are your single source of truth for renewals. Do not take it if you need a scriptable back-file migration; that is the Agreement Manager API’s strongest argument.
Substitute Spellbook Compare for Litera Compare — but not for Clean. Spellbook ships a Compare capability, so the comparison half of the pre-send step can move. The metadata half cannot: Spellbook has no equivalent to Clean’s 300+ metadata types across Word, PDF, Excel, and ZIP in Outlook. Rule for the swap: acceptable for the diff only, and only when the contract mix is Word-native and the team has no existing Litera licenses — Litera is frequently already paid for at teams staffed from AmLaw firms. Dropping the metadata step to save the license is the wrong trade; it is the cheapest line item and the one guarding the largest single disclosure risk.
Add Harvey for M&A diligence. The four tools handle commercial and operational contracts. Teams also running M&A diligence — asset purchases, equity deals, rep-and-warranty review — get more from adding Harvey alongside Spellbook for multi-document analysis, and Harvey now reaches primary law through its LexisNexis alliance, which narrows the gap to a dedicated research seat. Rule for the swap: justified above 3–4 internal transactions a year where diligence is not delegated wholesale to outside counsel.
Point the stack buy-side. Ironclad’s procurement agents and SAP packages make a supplier-contract configuration of this stack viable in a way it was not a year ago. If the driver is procurement savings rather than sell-side cycle time, start from the buy-side contract ops stack instead.
What this stack does NOT replace
- Outside counsel for bet-the-company litigation, securities work, or specialist regulatory matters — the stack improves commercial contract throughput, not legal judgment
- E-billing and matter management for controlling outside-counsel spend (Brightflag, now part of Wolters Kluwer, or Onit) — Ironclad manages contracts, not invoices
- A privacy compliance function for GDPR, CCPA, and AI Act obligations — contract data in Ironclad feeds a privacy program but is not one
- Legal research (Lexis+ with Protégé, Westlaw, CoCounsel) — Spellbook drafts against your playbook, not against case law, and novel questions still need a research layer
- Court-filing systems for active litigation