Welcome back to Good Better Best.

Each week, we break down real pricing, packaging, and product moves from SaaS and AI leaders and share the ideas worth stealing.

Earlier this week, we hosted a roundtable on all things AI Credits with three pricing leaders who have been navigating credit models for the last year:

It was an awesome session. Below, I’ve broken down my top 10 takeaways. If you want to watch the full recording, check it out here.

Let’s get to it.

This week we observed 310+ changes. The highlights:

Check out more updates on PricingSaaS →

Gong, Miro, and Databricks are very different businesses: a seat-based revenue platform, a horizontal collaboration tool, and a consumption-native data cloud. Yet they kept landing on the same lessons. Below are ten of my favorites.

Manu's framework is a clean mental model to start with. Good credit design answers three questions, in this order:

The trap is designing from the bottom up, starting with GPU time or tokens. Credits exist to close the gap between the value you deliver, what you charge, and what it costs you. To the extent you can, keep the complexity on the inside. Internally, all three layers can be messy. For customers, it should be simple enough that they don’t "need a PhD to explain the bill" to steal a phrase from Rita.

Gong defines a credit around the thing customers already associate with Gong: a call. When an autonomous agent reads or analyzes your conversations, a credit roughly equals a call. That maps cleanly to perceived value and willingness to pay, and it stays consistent no matter which feature consumes it.

The opposite approach that Rita warned against is the restaurant menu: a long list of agentic features, each with its own credit price, that has to be updated every week. One of her beta customers summed up the buyer's view: "I manage so many different budgets and so many different contracts. Please just keep it simple for me."

The takeaway: pick one familiar, durable unit and let new features map back to it.

Gong tested putting its AI role-play trainer on consumption, talked to beta customers, and decided against it. Usage pricing is "a magnifying glass on any flaws" in a product. With credits, you're betting that customers will see enough value to keep paying every time they use it.

Their rule of thumb: put autonomous agents and heavy AI processing on credits, and leave lighter or still-maturing features in the core package.

Before metering, ask two questions.

Jeff described a move from "phase one" to "phase two" AI credits.

In phase one, Miro priced on artifacts: the doc, slides or stickies AI created for you. That works when the output is predictable. With agents, you don't know how many steps it'll take or how much reasoning it will require to deliver an output.

Gong and Miro reached the same solution to this. They split agentic work into three stages and meter each one, so the bill reflects how much work the agent actually did:

Gong launched with the read stage alone (roughly a credit per call analyzed) and is adding work and output as it rolls out custom agents. Miro is moving from charging for the finished artifact to charging for the work done.

Don't lock each AI feature to a fixed price. Instead, decide how many credits each action costs, and treat that number as a dial you can turn. That's how Gong does it: analyzing a call costs a set number of credits, and Gong can change that number without touching anyone's contract.

The trick is to start conservative, so every change you make later is good news. If an action starts at 0.5 credits, you can drop it to 0.3, and customers get more for the same cost. Jeff put it simply: "You can change it later as long as you're giving people more. If you're trying to take some away, it's really painful."

Rita added two related points:

Model prices change by the week, so margin math done at the token or model level goes stale fast. Manu's answer is to manage profitability one level up, at the workflow: the job the customer is paying to get done. Here's how he breaks it down:

The payoff is a margin view that holds up every time a new model ships, because it tracks the work customers buy, not the tokens you pay for.

Credit mechanics are standardizing. Jeff's default if he were starting today: monthly allotments, no rollover. Begin there, and add complexity only when you can say why it's worth it.

Zendesk needed seasonality handling for support volume. Miro needed mechanics that fit PLG. Those are good reasons.

When you do add structure, Manu's three-bucket model is a useful template:

On rollover, Manu suggests partial rollover for enterprise customers as a trust-builder, but not 100%. Credits need to expire eventually.

When an audience member asked about fungibility across products, the answer was unanimous:

Separate pools turn one budget into several and leave credits stranded where customers don't need them. Unless you have a strong reason, give customers one balance they can spend wherever they get the most value.

Enterprise customers won't adopt credits they can't govern. Jeff and Rita both learned this after launch, and both landed in the same place: controls aren't a nice-to-have. They're a launch requirement.

At Miro, the controls enterprises wanted, by model, person and team, were "far greater than we knew at launch." Jeff's verdict: you can't launch credits to enterprises without them. Rita agreed: "You only have one chance to launch something really well." If customers won't use it because the controls aren't there, you've lost your momentum.

Gong launched with basic governance and added more later. Its credit profiles now cap credits per user, per feature, per month. They also answer the top enterprise question: "How many credits will I need?" Usage is too variable to forecast, so Gong now tells customers to budget against the limits they set, not an estimate.

Before you launch, make sure an admin can see and cap usage by user, team, feature and model. That's what gives buyers the confidence to commit.

Every panelist said some version of this: you can't design or launch credits from the pricing team alone, it’s critical to lean on other stakeholders who will inevitably be impacted by the new model.

Here's what that looks like in practice:

Credits touch pricing, product, finance, billing and sales all at once. Treat the launch as a company project, not a pricing project.

Credit models will keep changing as costs fall and agents get more capable. The teams getting it right aren't the ones with the perfect model at launch. They're the ones who built a model they can evolve without breaking trust.

Thanks to Rita , Jeff, and Manu for sharing what they've learned. Join the conversation in the PricingSaaS community, or watch the full recording on demand.

Thanks for reading! If you’re working on AI monetization and want to learn more about how we help, book time here.

Until next time,

Rob