Add-on, bundle, or new tier? How to package AI without blowing up your pricing

If you run pricing for a software product, the past year has offered a rare gift: the two largest software companies on earth running public, high-stakes packaging experiments on AI, changing their answers in real time, and letting everyone else learn from it for free.

Google bundled Gemini into Workspace Business and Enterprise plans and took a price increase of roughly 17-22% to pay for it. Then, effective March 1 of this year, it reintroduced a paid add-on for heavy users of its most expensive AI capabilities, such as advanced video generation and deep reasoning. Bundle first, then carve the expensive edge back out. Microsoft went the other direction: after selling Copilot as a bolt-on for two years, it announced commercial price increases in the low-to-mid teens taking effect July 1, restructuring licensing around Copilot-included bundles.

Two giants, two opposite journeys, arriving at the same place: a hybrid where baseline AI is bundled and premium AI consumption is metered. That convergence is the real lesson, and it is worth unpacking for anyone packaging AI features in their own product.

The three structures and what they signal

The add-on SKU

The add-on says: this is optional and it is for power users. It keeps your base price stable, gives sales an expansion motion, and lets you observe true willingness to pay. The risk is adoption. An unbundled AI feature has to win its budget line every renewal, and low attach rates become an argument for your buyer to cut it. Microsoft's early Copilot attach data was the industry's most watched number for exactly this reason, and the shift toward bundling tells you what they concluded.

The bundle with a price increase

The bundle says: this is now what the product is. It maximizes adoption, removes the purchase decision, and resets the value narrative for the whole product. The risks are two. You anger customers who do not want the feature and now pay for it anyway. And you absorb the costs of your heaviest users with no meter to protect you. Google's add-on reversal in March is what it looks like when the second risk arrives: the top few percent of users were consuming AI compute far beyond what the bundled increase priced in.

The new tier

The new tier says: AI is the reason to upgrade. It preserves choice, creates a clean upsell story, and segments customers by appetite. The risk is that it strands your AI investment in a tier most customers never see, while competitors give a version away below you. Tiers work best when the AI capability maps to a genuinely different job, not just a faster version of the same one.

A working decision framework

When we work through this with software teams, four questions settle most of the debate:

• Does the feature deepen an existing job or do a new one? AI that makes your core workflow faster belongs in the core packaging. Charging separately for it invites customers to ask why the base product is slow. AI that does a new job, such as an agent that completes work the user previously did elsewhere, can carry its own price because it displaces a different budget.
• What is the marginal cost curve? Near-zero marginal cost features can be bundled freely. Features with real per-use compute costs need a meter somewhere, visible or not. Inference prices keep falling for equivalent capability, but customers migrate to frontier capability just as fast, so do not build packaging on the assumption that your costs disappear.
• How concentrated is usage? Pull the data. If your top 5% of users would drive 60% of AI costs, a flat bundle transfers margin from you to them. This is the pattern that forced Google's reversal, and it is the single most common modeling error we see.
• Who is the buyer for the AI value? If the AI feature's value lands on a different persona than your current buyer, an add-on or tier gives that persona something to own. If it lands on the same buyer, separate packaging just adds friction to renewal.

The pricing metric is the strategy

Underneath the packaging question sits a deeper one that most teams avoid: what unit are you actually selling? Seats price access. Usage prices activity. Outcomes price results. AI is dragging the whole industry down that list, because agents do work rather than assist it, and work has a market price that access never had.

Salesforce now sells agent conversations and consumption credits. Intercom prices its support agent per resolution. These are not gimmicks; they are attempts to align the price metric with the value event. You do not need to jump straight to outcome pricing, and for most products in 2026 we would advise against it until attribution is solid. But you should know which value event your AI feature creates, because that event is where your pricing metric is eventually heading, and your packaging today should not make that migration harder.

Practical guardrails, whatever you choose

Put a fair-use meter behind every bundle. Even if you never enforce it, you want the contractual right and the telemetry. Retro-fitting a meter after customers have unlimited expectations is a renewal fight you do not want.

Grandfather loudly, migrate quietly. Price changes on existing customers are where AI packaging goes to die in public. Give existing customers a dated bridge, communicate the value received before the price asked, and never let the increase land before the feature does.

Instrument value from day one. Whatever you charge, capture the numerator: hours saved, items processed, resolutions completed. Every future packaging decision, and every renewal conversation, gets easier when you can show the value ledger. This is the same discipline we push for measuring AI ROI internally, applied to your own product.

Re-decide annually. Google and Microsoft have each materially changed their AI packaging more than once in 24 months. Your first structure will not be your last, and packaging agility, the operational ability to change SKUs without breaking billing, sales comp, and renewals, is itself a competitive advantage right now.

Packaging is your AI strategy, made visible

The instinct in most companies is to treat AI packaging as a pricing detail to settle after the feature ships. The market is punishing that instinct. Packaging is where your AI strategy becomes legible to customers: it tells them whether AI is the product, a perk, or a premium. The companies converging on hybrid structures, bundled baseline plus metered premium, are not copying each other. They are all discovering the same physics: adoption needs the bundle, and economics need the meter.

Get both into your structure before your usage data forces the issue.