AI Licensing Is the New Wild West: A Buyer's Framework for AI Features You'll Actually Use

AI is becoming part of almost every enterprise software renewal.

Microsoft 365 Copilot, Salesforce Agentforce, ServiceNow Now Assist, Atlassian Rovo, Adobe Firefly, Oracle Cloud AI and SAP Business AI all demonstrate the same shift: vendors are adding AI capabilities to existing suites, standalone products and cloud platforms. The commercial models are not consistent.

Some AI is included. Some requires an add-on seat. Some consumes credits or tokens. Some is priced by transaction, resolution or another business outcome.

That creates a new procurement problem. You may not know:

• Which AI features you already own.
• Whether a proposed add-on duplicates another tool.
• How quickly usage will grow.
• Whether unused credits expire.
• Whether a small pilot can become a large renewal liability.
• Whether a quoted USD price will create additional UK budget exposure.

Treat AI as a controlled procurement category, not an automatic upgrade. Use the framework below to evaluate the capability, the commercial model and the operational conditions before signing.

Source note: Vendor descriptions and pricing structures change quickly. The examples below reflect published vendor information and current market documentation available in September 2026. Confirm current terms, entitlements, regional prices and contract language before purchase. SO-LIC advisory recommendations are identified separately.

Map the AI packaging before comparing prices

Start by separating embedded AI from standalone AI.

Embedded AI in existing enterprise suites

Embedded AI appears inside software you may already operate:

• Microsoft 365 Copilot adds AI across Word, Excel, PowerPoint, Outlook and Teams. Microsoft publishes separate business and enterprise add-on pricing, while custom agent usage can introduce metered Copilot Credits through Copilot Studio. Microsoft also states that Copilot Chat is available at no additional cost with eligible plans, subject to feature and usage conditions. Review Microsoft’s current pricing.
• Salesforce Agentforce is one example of a suite vendor using consumption-based pricing. Depending on the offer, usage may involve Flex Credits, conversations or resolutions. Do not assess it in isolation from your existing Salesforce editions, Data Cloud configuration and automation estate.
• ServiceNow Now Assist is sold through product-specific AI tiers or add-ons with an “assists” usage concept. ServiceNow does not publish a universal public price list for all Now Assist configurations, so the order form must specify the entitlement, allowance and overage terms.
• Atlassian Rovo is included with eligible paid Cloud subscriptions. Atlassian allocates Rovo credits into a shared organisational pool. Additional usage, where enabled, can be billed separately. Read Atlassian’s Rovo licensing guidance.
• Adobe Firefly can be included within enterprise Creative Cloud or Express arrangements, with standard and premium generative credit allocations. Enterprise API and production use may be separately negotiated. Check Adobe’s generative credits guidance.
• Oracle Cloud AI generally follows cloud consumption principles. Costs may be based on tokens, transactions, search units, AI unit-hours or related OCI services. Review Oracle’s cloud price list.
• SAP Business AI distinguishes between included Base AI and Premium AI that uses SAP AI Units. SAP-delivered agents and other features may consume units by action, request or assigned user. Review SAP’s published AI pricing model.

Standalone AI tools

Standalone tools may look simpler because they are purchased directly for a defined purpose. They can still introduce:

• Per-user subscriptions.
• Tiered feature bundles.
• API or token charges.
• Minimum annual commitments.
• Separate data hosting and security costs.
• Currency exposure where the supplier invoices in US dollars.
• Overlap with capabilities already available in Microsoft, Salesforce, Adobe, Atlassian or other platforms.

Compare the total operating model, not just the advertised seat price.

Apply a seven-step buyer’s framework

1. Inventory where AI already exists

Create a consolidated view of your current estate before reviewing a new AI SKU.

Record:

• Vendor and product.
• Existing licence edition and tier.
• AI features already included.
• AI features enabled but unused.
• Administrators and business owners.
• Data sources connected to the feature.
• Usage records and available credit balances.
• Renewal date and termination rights.

Check productivity, CRM, ITSM, ERP, creative, analytics, collaboration and cloud platforms. A software asset management review should identify whether two products perform the same task, such as summarisation, knowledge search, document extraction, service triage or content generation.

SO-LIC advisory insight: Maintain this inventory in a controlled system such as LogicHarbour, linked to contracts, entitlements, owners, renewal dates and commercial obligations. Do not rely on vendor portals alone.

2. Classify the pricing model

Put every AI capability into one of four categories:

Ask whether credits are pooled or assigned by user. Confirm whether they reset monthly, expire annually, roll over, or are replenished automatically. Atlassian, for example, documents monthly Rovo allowance resets, while other vendors may use annual pools or contract-specific terms.

Do not treat the word “included” as meaning free. Establish the subscription tier required, the upgrade cost and the functionality that remains unavailable without additional usage.

3. Define the use case and measurable outcome

Choose one operational problem before choosing a product.

Examples include:

• Reduce service-desk triage time.
• Improve first-draft quality for regulated documents.
• Classify incoming cases.
• Extract fields from supplier invoices.
• Reduce time spent searching internal knowledge.
• Improve developer code-review throughput.

Define a baseline and target. Use measures such as:

• Minutes per transaction.
• Cost per case.
• Percentage of cases resolved without escalation.
• Error or rework rate.
• Adoption by eligible users.
• AI cost per completed workflow.
• Percentage of outputs requiring human correction.

Avoid vague objectives such as “increase productivity”. A measurable outcome gives procurement, IT and finance a common basis for approval.

4. Run a 90-day pilot with exit criteria

Use a controlled pilot, not a broad rollout.

Set:

• A defined user group.
• A fixed data set.
• Approved use cases.
• Named business and technical owners.
• Usage and cost monitoring.
• Security and privacy review.
• A baseline for comparison.

Set exit criteria before the pilot begins. For example, proceed only if the feature:

• Meets the agreed accuracy threshold.
• Reduces task time by the target amount.
• Stays within the approved cost per transaction.
• Passes data protection and access-control checks.
• Achieves sufficient active-user adoption.
• Has a clear operating owner.

Stop, redesign or renegotiate if the feature meets none of these conditions.

5. Model conservative, expected and aggressive consumption

Build three scenarios before approving a credit or usage commitment.

Use a simple model:

Annual AI cost = fixed licence cost + expected usage × unit price + integration and governance cost

Model:

• Number of eligible users.
• Percentage of active users.
• Requests per user.
• Credits or tokens per request.
• Seasonal peaks.
• Agent retries and failed workflows.
• Growth in connected data sources.
• Currency conversion and supplier price changes.

The aggressive scenario should include increased adoption, more complex prompts, automation at scale and higher data volumes. Do not use the vendor’s demonstration workload as your forecast.

For UK organisations, show both the supplier currency and the budget currency. A USD-denominated credit pack can change the effective sterling cost even when the supplier’s list price remains unchanged. Record the exchange-rate assumption and decide whether finance requires a contingency.

6. Negotiate guardrails before enabling usage

Request contract terms that make variable spend controllable.

Prioritise:

• Budget alerts at agreed thresholds.
• Hard usage caps where technically available.
• Admin approval for overage.
• No automatic replenishment without written authorisation.
• Rollover of unused credits.
• Clear treatment of expired credits.
• Rate protection during the initial term.
• Advance notice of pricing or meter changes.
• Opt-out rights for optional AI features.
• Downgrade rights after the pilot.
• Termination or suspension rights for unused add-ons.
• Exportable usage data.
• Defined service and support responsibilities.

Ask the vendor to show the meter in a worked example. Require separate lines for seats, credits, infrastructure, API calls, storage, implementation and support.

7. Assign ownership and review at renewal

AI licensing requires joint ownership.

Assign responsibilities across:

• IT: configuration, security and integration.
• Procurement: price, terms and competitive assessment.
• Finance: budget, currency and forecast accuracy.
• Legal and privacy: data processing, intellectual property and regulatory obligations.
• Business owners: adoption and measurable outcomes.
• Software asset management: entitlement, usage and renewal control.

Review monthly during the pilot and quarterly after production deployment. Start formal renewal governance 90–180 days before the renewal date, depending on contract complexity. This gives you time to compare actual consumption with the original forecast, remove unused seats, challenge overage and renegotiate credit terms.

Use this decision checklist before purchase

Approve an AI feature only when you can answer “yes” to the relevant questions:

• Have we checked whether the capability already exists in our estate?
• Is the use case linked to a measurable business outcome?
• Do we understand the billing unit?
• Do we know the included allowance and overage rate?
• Do unused credits expire or roll over?
• Can administrators set usage limits?
• Have we modelled conservative, expected and aggressive demand?
• Have we included GBP/USD currency exposure?
• Is data quality sufficient for the proposed use case?
• Are access, retention and audit controls defined?
• Is there a 90-day pilot and a documented exit route?
• Does the contract allow downgrade, opt-out or termination of the AI add-on?

Ask these questions before signing

1. Which AI capabilities are already included in our current licence?
2. Which features require an upgrade rather than a standalone add-on?
3. What exactly triggers a billable event?
4. Can one workflow consume multiple credits, actions or tokens?
5. Are failed, repeated or retried requests chargeable?
6. What happens when the allowance is exhausted?
7. Can the vendor change the meter or unit price during the term?
8. Are credits pooled across users, products or tenants?
9. Do credits expire, and is rollover available?
10. Can we disable AI usage without cancelling the base platform?
11. What usage reports can we export for software asset management?
12. Which data is used for grounding, training, logging and audit?
13. What is the effective cost in GBP under our approved exchange-rate assumption?
14. Can we reduce the commitment at renewal if adoption or outcomes are below target?

Make AI procurement an evidence-led decision

AI features can create value. They can also create overlapping entitlements, opaque consumption and renewal cost that is difficult to forecast.

Use your existing software licensing solutions, IT procurement controls and software asset management discipline. Browse the SO-LIC vendor directory to understand the wider catalogue, and use SO-LIC comparison guides when evaluating alternatives.

The correct question is not whether a vendor has added AI. Ask whether the capability solves a defined problem, uses data you can govern, produces a measurable outcome and operates within commercial guardrails you control.