Technology Contracts Toolkit: SaaS, Licensing, Service Levels, and Exit

By Casey Scott McKay · 2024-10-29

Technology agreements fail on eight provisions and the license grant is not one of them. This toolkit runs a deal from intake through renewal and exit, and routes each stage to the Marksy documents that do the work. It explains why a subscription conveys access rather than a copy and what follows from that, why the royalty of technology procurement is the renewal cap and the exit specification rather than the liability cap, and where the money actually goes - renewal pricing, overage and true-up, modules, professional services, and migration. It covers scope and user definitions, service levels and the chronic-failure termination right worth more than any credit, the data provisions including model training, indemnity and the super-cap, security and subprocessors, statements of work and ownership of custom development, the AI provisions now standard, and a twenty-minute triage for agreements that cannot justify full review.

IP and Technology > Information Technology | Toolkit | Published 29 October 2024 - Updated 22 January 2026 | Casey Scott McKay - marksy.us

Summary. Technology agreements fail on eight provisions and the license grant is not one of them. This toolkit runs a deal from intake through renewal and exit, and routes each stage to the Marksy documents that do the work. It explains why a subscription conveys access rather than a copy and what follows from that, why the royalty of technology procurement is the renewal cap and the exit specification rather than the liability cap, and where the money actually goes — renewal pricing, overage and true-up, modules, professional services, and migration. It covers scope and user definitions, service levels and the chronic-failure termination right worth more than any credit, the data provisions including model training, indemnity and the super-cap, security and subprocessors, statements of work and ownership of custom development, the AI provisions now standard, and a twenty-minute triage for agreements that cannot justify full review.

Keywords: deal structure · subscription versus license · access grant · authorized users · affiliate definition · usage metrics and overage · uptime and exclusions · chronic failure termination · customer data ownership · usage and aggregated data · model training terms · ip indemnity carve-out · security super-cap · incident notification hours · data processing agreement · subprocessors · renewal cap · exit and export specification · statement of work ownership · portfolio management

Start Here

Ellersley Manufacturing runs eighty-one technology agreements and has never read most of them since signature. Four arrive on a desk in the same week.

The customer relationship platform is renewing for the fourth time. The original agreement had a ninety-day notice period, no price cap, and exit language committing to "standard export functionality." The renewal quote is up a fifth.

A new field service platform is in negotiation, with a real competitive alternative and a mid-six-figure annual spend.

A statement of work for custom integration work is ready to sign, and nobody has looked at who will own the code.

And an existing vendor has announced that customer content may be used to train its models, effective at the next renewal, under a unilateral amendment clause nobody knew was in the agreement.

Four documents, one portfolio, and one recurring cause: the terms that matter are read once at signature by people focused on price.

This toolkit answers three questions.

What is actually being bought? A subscription conveys access, not a copy, and every consequence follows from that.
Which eight provisions decide the outcome? They are not the ones customers negotiate.
How is a portfolio managed rather than accumulated? A register, a position sheet, a triage, and a calendar.

If you read only one thing, read What You Are Actually Buying. It frames the distinction that organizes everything else.

Subscription Is Not a Licence

A traditional software licence conveys a right to reproduce and use a copy, which is an exercise of the copyright owner's exclusive rights under 17 U.S.C. § 106. The customer holds a copy, and 17 U.S.C. § 117 supplies background rights to the owner of a copy.

A subscription conveys nothing of the kind. The customer receives access to a service running on the vendor's infrastructure. No copy is delivered, the essential-step provision does not apply because the customer owns no copy, and first sale under 17 U.S.C. § 109 is irrelevant for the same reason.

What follows. The customer cannot continue using the software if the agreement ends, cannot transfer its rights, cannot inspect or modify anything, and depends entirely on the vendor's continued operation. Termination is not a loss of a licence; it is a loss of access to a running system holding the customer's data.

Which is why the exit provisions matter more than the grant, and why they are negotiated last by nearly everyone.

The adjacent structures. An on-premises perpetual licence delivers a copy, so the exit risk is obsolescence rather than loss of access. A term licence self-hosted combines the exit analysis of a subscription with the operational burden of on-premises. A managed service is a three-party arrangement where flow-down governs. A reseller purchase incorporates the vendor's terms by reference while the reseller disclaims everything. And API terms govern integrations the business depends on and are almost never negotiated.

Intake

Identify the deal structure, because the forms differ and so do the risks.

Identify what data the service will hold — personal data, regulated data, trade secrets, or nothing sensitive — which determines whether the security addendum is the centerpiece or an afterthought.

Assess the dependency. A system of record the business cannot leave requires exit terms; a peripheral tool does not.

Establish the spend and the term.

Assess the vendor. An established vendor with thousands of customers will not move on architecture. A growth-stage vendor with a large deal in front of it will move on nearly everything.

Establish whether a competitive alternative is real, which is the single largest source of leverage.

Write a half-page intake sheet and derive the negotiation plan from it. Redlining every clause of a small subscription with an established vendor produces a form response and consumes the review budget.

Scope, Users, and Metrics

Read the access grant's adjectives. Non-exclusive, non-transferable, for internal business purposes. Non-transferable defeats assignment in an acquisition without consent; internal business purposes forecloses service-bureau use.

Test the user definition against how the business actually works. Named users, concurrent users, seats, or metered consumption, and whether a departing employee's seat can be reassigned.

Fix the affiliate definition. Whether subsidiaries may use the service and whether their usage counts. The most common source of unexpected true-up invoices, and it is fixed in one sentence at signature.

Address contractors and outsourced functions.

Understand the usage metric and who measures it.

Negotiate the overage mechanic, not the rate. Automatic billing at list, throttling, and a good-faith negotiation are very different outcomes.

Read the restrictions — benchmarking, competitive use, and reverse engineering — noting the interoperability interaction at 17 U.S.C. § 1201 and the narrowed reach of 18 U.S.C. § 1030 over conduct exceeding contractual terms.

Service Levels

Read the uptime number as minutes. 99.9 percent is roughly forty-three minutes a month; 99.5 percent is over three and a half hours; 99 percent is more than seven.

Read the exclusions. Scheduled maintenance, emergency maintenance, force majeure, customer-caused issues, third-party networks, and beta features. Unlimited scheduled maintenance defeats the commitment.

Check the measurement method and whether degradation counts as downtime.

Do not negotiate larger credits. They are the exclusive remedy in nearly every form and they are a pricing adjustment rather than compensation — typically a percentage of one month's fee, capped, for an outage that stopped the business.

Negotiate a chronic-failure termination right instead. A defined number of breaches in a rolling period, with termination and a pro-rata refund. This converts the service level from a discount into an exit, and vendors trade it readily because they understand what is being given up.

Negotiate support separately. Response and resolution targets, severity definitions, and coverage hours. Severity definitions matter most because the vendor usually assigns severity.

Ask for a deprecation policy where integrations will be built, because interface stability is rarely committed and a deprecated interface breaks the customer's integration.

The Data Provisions

The most valuable section in the agreement and the one negotiated least.

Customer data ownership. The customer owns its data; the vendor holds a limited licence to process it to provide the service. Anything broader transfers an asset for no consideration.

Usage data and telemetry. Reasonable in principle. The questions are aggregation, de-identification, whether re-association is possible, and whether it may be shared or used to train models.

Aggregated and anonymized data. The standard clause permits any use. Scrutinize it where the customer's data is competitively sensitive or where aggregation across a small industry permits inference.

Model training. Now the most contested term. Whether customer content may be used to train, whether outputs may reflect it, and whether an opt-out exists. Treat a training permission as a disclosure and evaluate it against the confidentiality classification of the content — which is Ellersley's fourth problem, arriving through a unilateral amendment clause.

Feedback. Assigned or licensed broadly to the vendor. Standard, generally acceptable, and worth limiting where the customer's personnel will be deeply engaged in product design.

Data location and residency, stated specifically rather than aspirationally, because multi-region deployments, backup replication, and support access all move data.

Confidentiality. The agreement's confidentiality provisions create the duty that makes misappropriation actionable. 18 U.S.C. § 1839; 18 U.S.C. § 1836.

Indemnity and Liability

The intellectual property indemnity should cover patent, copyright, trademark, and trade secret claims arising from the service, with standard exclusions for modifications, combinations, use outside the agreement, and continued use after notice.

The remedy structure. Procure, modify, or terminate and refund. The vendor will choose refund, which is why a customer dependent on the service should negotiate a transition period before termination becomes the remedy.

Carve the indemnity out of the liability cap. An intellectual property indemnity capped at twelve months of fees does not answer a patent claim.

Obtain a security-incident super-cap. Breach exposure routinely exceeds twelve months of fees, and a separate higher cap is the standard compromise.

Carve out confidentiality breach and willful misconduct.

Accept mutual exclusion of consequential damages, because a vendor's exposure is otherwise unbounded and it will not agree.

Understand what drives vendor flexibility. Deal size and insurability. A cap the vendor's insurance covers is negotiable; unlimited liability is not.

Security, Privacy, and Subprocessors

The security addendum. Technical and organizational measures, encryption at rest and in transit, access controls, logging, vulnerability management, and testing — reviewed against the customer's own obligations rather than as boilerplate.

Set the incident notification deadline in hours, and confirm whether it runs from discovery or from confirmation. The vendor's delay consumes the customer's regulatory clock, and this is a genuine trade because it is operationally meaningful for the vendor.

Obtain certification reports annually, with the right to review exceptions.

Execute a data processing agreement where personal data is involved. Cal. Civ. Code § 1798.140; Cal. Civ. Code § 1798.100.

Confirm the vendor can delete a named individual's data on request, because many systems disable an account without removing the record. Cal. Civ. Code § 1798.105.

Subprocessors. A list, notice of additions, and a termination right where an objection cannot be accommodated. Objection rights reduce to notice plus termination because a vendor cannot maintain per-customer infrastructure.

Flow down sector rules. 15 U.S.C. § 6801 for financial institutions; 45 C.F.R. § 164.410 for the health business associate relationship.

Check for misleading security statements. 15 U.S.C. § 45.

Term, Renewal, and Exit

Auto-renewal is standard and the notice period is the trap. Sixty or ninety days, and a missed window commits the customer for another year. Calendar it on the day of signature, in a system that alerts the business owner and not only the lawyer.

Cap the price increase, tied to a published index or a fixed percentage. The largest long-term exposure in most portfolios and it costs the vendor nothing at signature — which is Ellersley's first problem and the reason its renewal quote is up a fifth.

Consider term length honestly. A multi-year commitment buys a discount and removes the annual exit, which deserves more thought for a system of record than it usually gets.

Address the unilateral amendment clause. Whether the vendor may change terms by posting, whether notice is required, and whether the customer's only remedy is termination. It quietly rewrites everything else between renewals, and it is how Ellersley's model training term arrived.

Specify the data export. Format, completeness including custom fields, attachments, and audit history, mechanism, window, and cost. "Standard export functionality" is not a commitment and it produces records without the fields that matter.

Obtain transition assistance. Continued access at the existing rate for a defined period after termination, and cooperation with a successor vendor.

Require deletion certification after the export window.

Do not rely on source code escrow for a hosted service. Code without infrastructure and operational knowledge is unusable. A hosted continuity arrangement or a right to run the software on the customer's infrastructure is worth more.

Test the exit before signature. Ask the vendor to describe in writing exactly what a departing customer receives. The answer is informative either way.

Statements of Work and Custom Development

Ownership does not follow payment. The developer owns what it writes absent a written assignment, because the work made for hire categories are narrow and a services contract is not automatically within them. 17 U.S.C. § 101; 17 U.S.C. § 201; 17 U.S.C. § 204.

Use present-tense assignment language, and take a licence back to any vendor background technology needed to use the deliverable.

Define deliverables and acceptance criteria specifically, with a process and a remedy for failure.

Read the statement of work's own terms, because it frequently incorporates weaker terms than the master agreement or purports to supersede it — which is Ellersley's third problem.

Address open source in deliverables, because a component's licence conditions travel with the code regardless of the vendor's warranties.

Define the developer's reuse carve-out rather than leaving it open. Most vendors will not assign generic components, and a defined carve-out is reasonable.

The AI Provisions

Training on customer content. The threshold question, and enterprise tiers increasingly prohibit it or make it opt-out while consumer tiers frequently permit it.

Outputs and ownership. Vendors typically assign outputs to the customer with a caveat that similar outputs may be generated for others — accurate, and unsettling to customers.

The authorship gap. Purely machine-generated material may not be copyrightable, because 17 U.S.C. § 102 requires authorship. An assignment conveys whatever rights exist, which may be none.

Output indemnity, increasingly offered and always conditioned on using the service as directed and not disabling filters. The conditions carry the clause and they should be checked against the customer's actual workflow.

Accuracy disclaimers, universal, and they do not relieve the customer of its own obligations to its customers or regulators.

Input restrictions, reconciled with the customer's own compliance obligations.

Human review of inputs and outputs for abuse monitoring or quality, and whether it can be disabled — a confidentiality question dressed as an operations detail.

Model provenance and change rights. Which models, hosted where, whether third-party model providers are named subprocessors, and whether the vendor may change underlying models. A customer that validated behavior against one model has no commitment that it persists.

Where the Money Goes

Five places, and none of them is the licence grant.

Renewal pricing. Increases of twenty or thirty percent are ordinary where no cap was negotiated, and migration is expensive enough that the customer usually pays.

Overage and true-up, driven by the user and affiliate definitions, producing an unbudgeted invoice that is correct under the agreement as written.

Modules and add-ons, priced on the assumption that the customer is already committed.

Professional services, on time and materials, exceeding the estimate, under weaker terms than the master agreement.

Exit. Data in an unusable format, no transition period, and a migration under time pressure at the vendor's convenience.

The allocation rule. A customer with limited leverage should spend it on the renewal cap and the exit provisions rather than on the liability cap. The liability cap matters if something goes wrong; the renewal and exit terms matter every year regardless.

Multi-Tenant Realities to Accept

There is one version. Everyone upgrades together; a customer cannot stay on an old release or require testing before changes reach production. Advance notice of material changes and a sandbox are the realistic asks.

Configuration is not customization. What the customer can change is what the vendor built configurable.

Integrations depend on vendor interfaces, and stability is rarely committed.

Performance is shared, and degradation frequently does not count as downtime.

Data location may not be what the customer assumes.

Subprocessor changes are a fact of operation.

The conclusion. The customer cannot control the service. It can control whether it can leave, which is why the exit provisions carry the weight the operational provisions cannot — and why time spent asking for per-customer infrastructure is time not spent on the terms a vendor can actually give.

The Twenty-Minute Triage and Portfolio Management

For agreements that cannot justify a full review, this order finds the most exposure per minute.

One: the renewal clause. Notice period, auto-renewal, price cap. Two minutes. Two: the exit clause. Export format and completeness, transition assistance, deletion. Three minutes. Three: the data clauses. Ownership, usage and aggregated rights, model training. Four minutes. Four: the liability cap and carve-outs. Three minutes. Five: user and affiliate definitions against how the business works. Three minutes. Six: service level exclusions and whether credits are exclusive. Two minutes. Seven: incident notification deadline. One minute. Eight: the unilateral amendment clause. Two minutes.

Then manage the portfolio. A register with vendor, service, spend, term, renewal date, notice date, price cap, exit terms, and data sensitivity. A position sheet with preferred, acceptable, and walk-away language for the twelve terms that matter. A spend threshold for full review. Standardized security and data processing addenda so the customer proposes rather than reacts. A concession log, because the next negotiation with the same vendor starts from the last one. A review at the second renewal, because usage has changed. And an annual one-page report: total spend, agreements without price caps, agreements without exit terms, and agreements renewing in the next two quarters.

What the report produces. Most legal departments discover, the first time they run it, that they hold several agreements nobody has read since signature, at least one of which renews automatically at uncapped pricing on a service the business could not now leave. That discovery is the point of the exercise.

Ellersley's Four Documents

The fourth renewal. The company is in the position the exit provisions exist to prevent. Before the notice date, do three things: run an actual data export and see what comes out; document an evaluation of a competitor with an implementation estimate and a migration plan; and bring the utilization data — seats provisioned versus active, modules purchased versus used, consumption against entitlement. Vendors expect a price discussion and are unprepared for a utilization discussion. Then ask for the terms refused at signature: a price cap, an export specification, and transition assistance. A vendor facing renewal risk will move on all three, and this renewal is the only opportunity the agreement will ever get.

The new platform. A mid-six-figure spend with a real alternative is the strongest position the company will hold in that relationship. Six asks, each with a stated reason: a renewal price cap; six months of transition assistance with a full export specification; a chronic-failure termination right traded for the credits; a security-incident super-cap; a model training prohibition; and the affiliate definition fixed against how the business actually works. A redline touching eleven clauses gets engaged; one touching a hundred gets a form response.

The statement of work. Do not sign it. Payment conveys nothing — the developer owns what it writes absent a written assignment, and the work made for hire categories are narrow. 17 U.S.C. § 201; 17 U.S.C. § 204. Add present-tense assignment language, a licence back to the vendor's generally applicable components, defined deliverables and acceptance criteria, an open source inventory, and a defined reuse carve-out. Then read the statement of work's own terms, because it probably incorporates weaker liability and confidentiality provisions than the master agreement.

The unilateral amendment. The clause was in the agreement at signature and nobody read it. The immediate question is what the amendment actually permits — training on customer content, and whether an opt-out exists. The second question is what the agreement's own remedy is, which is usually termination. Where the content is confidential, regulated, or competitively sensitive, treat the amendment as a proposed disclosure and respond accordingly: opt out where possible, negotiate an exception, or exercise the termination right. And add the unilateral amendment clause to the eight-point triage, because it quietly rewrites every other answer between renewals.

Talking to the Business

The negotiation succeeds or fails on whether the business owner understands three things.

A subscription is a dependency, not a purchase. No copy, no continuation on termination, no transferable rights. Everything about the exit terms follows from that, and an owner who understands it stops treating exit provisions as lawyerly caution.

Leverage exists only before signature. Afterward, the vendor's incentive to accommodate is a function of renewal risk, and renewal risk is a function of whether the customer can actually leave.

The negotiation has a budget. Six asks get engaged; sixty get a form response. The business should help choose which six, because at least two are commercial rather than legal.

Five questions to ask the business before drafting. How dependent will we become. What data goes in. How many people and which entities. What would migration cost if we had to. And what is the realistic alternative if this vendor becomes unacceptable. Those five answers determine the negotiation plan more than any clause analysis.

And what to report back after signature, in the register: what was obtained, what was conceded, when the notice date falls, and what the exit actually looks like. The people who negotiated it will not be the ones facing the renewal.

Twenty Failure Modes

Redlining everything, which gets a form response and consumes the review budget.

Negotiating the licence grant rather than the exit.

Ignoring the affiliate definition, producing an unbudgeted true-up that is correct under the agreement as written.

A named-user agreement for occasional broad usage.

Unlimited scheduled maintenance accepted as an uptime commitment.

Negotiating larger credits instead of a termination right.

Renewal pricing left uncapped, the largest recurring exposure in most portfolios.

The non-renewal notice window missed, which loses more value than any negotiated term.

A multi-year term committed to without noticing it removed two annual exits.

"Standard export functionality" accepted as an exit commitment.

No transition assistance, so migration happens under time pressure at the vendor's convenience.

Indemnity inside the liability cap.

No security-incident super-cap.

Incident notification measured in days, consuming the customer's own regulatory clock.

The unilateral amendment clause ignored, which rewrites everything else between renewals.

The statement of work assumed to be covered by the master agreement.

Payment assumed to convey ownership of custom development.

Source code escrow relied on for a hosted service.

Model training permitted without evaluating the content's confidentiality classification.

API and developer terms never read before building an integration the business depends on.

Building the Standing Playbook

Organizations that negotiate technology agreements regularly should stop analyzing each from scratch.

Write the position sheet. For each of the twelve terms that matter, the preferred language, the acceptable compromise, and the walk-away. It turns most negotiations into lookups and lets business people resolve the routine ones without counsel.

Set a spend threshold for full review, with the eight-point triage below it.

Standardize the security addendum and the data processing agreement, so the customer proposes rather than reacts.

Keep a concession log — what was given, to whom, and why. The next negotiation with the same vendor starts from the last one, and vendors remember even when customers do not.

Maintain the register, and calendar every notice date on the day of signature in a system that alerts the business owner.

Review at the second renewal, because usage has changed and the terms that fit at signature may not fit now.

Report the portfolio annually on one page: total spend, agreements without price caps, agreements without exit terms, and agreements renewing in the next two quarters. It is the document that gets exit terms taken seriously the next time.

Train the people who sign. Most technology agreements are executed by business owners under a delegated authority, and the single highest-return intervention available is teaching them the eight triage questions.

What the Vendor Is Protecting

Reading the form from the vendor's side explains most of the terms customers find objectionable, and it tells a negotiator where to spend.

Uniformity is non-negotiable. A vendor operating one multi-tenant system cannot maintain per-customer infrastructure, per-customer subprocessors, or per-customer security configurations, and it will not agree to terms that would require any of them. This is refused regardless of leverage or deal size, and a customer that understands it early negotiates faster and spends its asks better.

Liability is bounded by design. Subscription economics assume exposure limited by fees. Movement is a function of deal size and insurability, which is why a security super-cap is obtainable and unlimited liability is not.

Competitive restrictions exist for a reason. Benchmarking, competitive-use, and reverse engineering prohibitions are there because the vendor's most dangerous customer is a future competitor.

The improvement loop is the business. Usage data, aggregated analytics, and feedback feed product development, and a vendor that cannot learn from its customers is operating blind. The negotiation is about scope and identifiability, not about whether the vendor gets anything.

Operational commitments have staffing consequences. Support targets, uptime, and notification deadlines are promises someone must be rostered to keep — which is exactly what makes a two-hour notification obligation a genuine trade rather than a free ask.

The sorting rule. Distinguish terms that cost the vendor money from terms that cost it nothing at signature. Price caps, exit assistance, indemnity carve-outs, notice periods, and data-use limits are frequently obtainable because none of them affects the vendor's operating model. Per-customer infrastructure, unlimited liability, deferred upgrades, and unilateral audit rights over a multi-tenant environment generally are not, and asking for them consumes the goodwill the obtainable terms need.

A Suggested Reading Path

Before signing:

For the code and components inside:

For the data obligations that travel with it:

Primary Authorities

| Authority | Proposition | |---|---| | 17 U.S.C. § 101 | Work made for hire definitions | | 17 U.S.C. § 102 | Subject matter; authorship | | 17 U.S.C. § 106 | Exclusive rights | | 17 U.S.C. § 109 | First sale; inapplicable without a copy | | 17 U.S.C. § 117 | Essential step and archival copies | | 17 U.S.C. § 201 | Ownership; written assignment | | 17 U.S.C. § 204 | Transfers require a signed writing | | 17 U.S.C. § 1201 | Circumvention; interoperability | | 18 U.S.C. § 1030 | Computer access; usage restrictions | | 18 U.S.C. § 1836 | Trade secret civil action | | 18 U.S.C. § 1839 | Trade secret definition | | 15 U.S.C. § 45 | Unfair or deceptive practices | | 15 U.S.C. § 6801 | Financial safeguards | | Cal. Civ. Code § 1798.100 | Notice and purpose limitation | | Cal. Civ. Code § 1798.105 | Deletion capability | | Cal. Civ. Code § 1798.140 | Service provider definitions | | 45 C.F.R. § 164.410 | Business associate breach notice | | Fed. R. Civ. P. 65 | Injunctive relief |

Forms and Templates

The License Agreement Template is the starting point for the licensing structures this toolkit compares against subscriptions, and reading it beside a vendor's subscription form makes the difference visible: one grants rights in a copy, the other grants access to a service, and the exit consequences diverge completely. The Assignment Agreement Template supplies the present-tense assignment language a statement of work needs, because payment conveys nothing and the work made for hire categories are narrow. The Portfolio Inventory Template doubles as the technology contract register — vendor, service, spend, term, renewal date, notice date, price cap, exit terms, and data sensitivity — which is the document that converts a pile of agreements into a managed portfolio and that surfaces the uncapped renewal before it arrives as an invoice. The Cease and Desist Template is the instrument for the rarer case where a vendor or a former customer misuses data or code outside the agreement's permissions.

Related Toolkits and Checklists

For the components inside the software and the obligations that travel with them, the Software, Data, and Open Source Toolkit. For the breach obligations that vendor terms determine, the Incident Response and Breach Notification Toolkit. For sensitive data held by a vendor, the Biometric and Sensitive Data Toolkit. And where the relationship involves joint creation rather than pure procurement, the Patent Licensing and Technology Transfer Toolkit covers the licensing terms and the know-how transfer.