How to Tell a Customer Their AI Usage Is Costing You (Without Losing Them)

Three or four customers pay you the exact same $49/month. One of them barely touches the AI feature. Another runs it on every page load, every day, and last month their usage alone cost you more than they paid. In your Stripe dashboard, they look identical. The only place the difference shows up is in a number your invoice never breaks out, and once you see how to tell a customer their AI usage is costing you more than they pay, you realize you can't just ignore it forever.

Here's the short version: once your per-customer cost data confirms one customer is genuinely unprofitable, not just for one weird month, but consistently, you have four options. You can repackage your plans so their usage fits a higher tier. You can add metered or overage pricing. You can renegotiate with them directly. Or you can let them go. Which one you choose depends on the customer, the gap, and what your product can support. But the conversation itself? That's what most founders get wrong, because they lead with the one thing they should never say out loud.

Why "Our AI Costs Went Up" Is the Worst Way to Start

Before I get into the playbook, I need to talk about the trap almost every founder walks into.

The instinct is natural. You open your AI provider bill, see the number, trace it back to one customer, and your first thought is: "I need to tell them our costs went up." It feels honest. It feels transparent.

It's also the fastest way to make your customer feel like they're being punished for using your product the way it's supposed to work.

Think about it from their side. They signed up for your product. They used the features you built and marketed. Now you're telling them it costs you too much to serve them? That doesn't sound like transparency; it sounds like you didn't think through your own pricing, and now that's their problem.

This is why the AI usage-based pricing conversation is harder than a normal price increase. A normal increase can be anchored to new value: new features shipped, better support, a stronger product. But when the real reason is "your usage pattern costs me money," there's no new value to point to. The customer is getting the same thing. You're just charging more for it.

So the framing matters enormously. I'll get to exactly what to say in Step 4. But first, make sure you're solving a real problem and not reacting to noise.

Step 1: Confirm It's Real, Not a One-Month Blip

One bad month doesn't mean you have a margin-negative customer. It means you had a bad month.

AI usage is spiky. A customer might run a heavy batch one week; maybe they imported a big dataset, or a new team member started using the feature aggressively, and then their usage drops right back down. If you react to a single spike, you risk having a stressful conversation over something that would have corrected itself.

Here's what I'd check before acting:

• Is this customer's AI cost consistently above what they pay? Not once, but for at least two or three billing cycles in a row.
• Is the pattern growing or stable? A customer who costs $55 against a $49 plan for three months is different from one who went from $30 to $55 to $90.
• Is there a reason? Sometimes the spike is legitimate product usage you should have anticipated. Sometimes it's an integration loop or a bug on your side.

If you don't have per-customer cost numbers yet, that's the first problem to solve. You can't have a conversation about a number you haven't actually calculated.

Step 2: Decide Which Move You're Actually Making

Once you've confirmed the problem is real, don't jump straight to "I need to raise this customer's price." That's one option, but it's not the only one, and it might not be the best one.

There are four paths. Each one fits a different situation.

Repackage into tiers. If the customer's usage pattern is actually common among your heavier users, the answer might not be a one-off conversation; it might be a new plan tier. You create a higher tier with more generous usage limits and a price that covers the cost, then migrate them (and future heavy users) into it. This works best when you have multiple customers who are heading toward the same usage level; it's a product decision, not a customer-by-customer negotiation. If your goal is to raise prices for AI features without losing customers at scale, this is usually where to start.

Add metered or overage pricing. Instead of a flat monthly fee, you charge for usage beyond a certain threshold. Below the threshold, everything works the same. Above it, the customer pays per unit (per request, per thousand tokens, whatever maps to your actual cost driver). This is the most precise fix, but it adds billing complexity, and it only works if your customer understands what "usage" means in your product.

Renegotiate directly. Sometimes the right move is a straightforward, honest conversation with one specific customer. This works best when it's a high-value relationship, maybe they're on a custom deal, or they're a design partner, or they send you referrals. You're not changing your pricing structure for everyone. You're adjusting the economics of one relationship that stopped working.

Let them go. I'll cover this separately below, because it deserves more than a bullet point.

The point is: figure out which category this customer falls into before you decide what to say. A pricing-structure change is a product decision. A direct renegotiation is a relationship decision. They need different conversations.

Step 3: Time It to a Moment, Not a Mood

Don't send the email the same afternoon you see the number. The worst version of this conversation happens when you're frustrated, and the customer has no context for why it's coming.

Good timing triggers:

• Renewal. If the customer is approaching a renewal date, that's a natural moment to revisit pricing. They're already expecting a conversation about the next period.
• A usage milestone. If you can show the customer how much they've used, "You ran 12,000 AI requests last month, which is 4x what most customers on this plan use", that creates a factual anchor rather than an emotional one.
• A new feature or plan launch. If you're building new tiers anyway, the migration conversation is a natural fit.
• After a value moment. If the customer just had a win with your product, closed a deal using your AI feature, hit a milestone, that's when they're most receptive to hearing "you're getting a lot of value from this, and here's how we can make sure the economics work for both of us."

The principle is simple: the customer should already be thinking about their relationship with your product before you introduce a change. If the conversation comes out of nowhere, it feels punitive. If it comes at a natural moment, it feels like an evolution.

How to Tell a Customer Their AI Usage Is Costing You, What to Actually Say

This is the part everyone skips to, so I'll be direct.

Don't open with your costs. Open with their usage.

Here's the difference:

Cost-framed (don't do this): "Hey, our AI infrastructure costs have increased significantly, and we need to adjust pricing to reflect that."

Usage-framed (do this instead): "I noticed you're one of our heaviest users of [feature name]; you ran [X] requests last month, which is about [Y]x what most customers on this plan use. That's a good sign; it means the feature is clearly useful to you. I want to make sure the plan you're on actually fits how you're using the product."

The second version does three things the first one doesn't:

1. It starts with a fact about them, not about you.
2. It frames their heavy usage as a positive signal, not a problem.
3. It opens a conversation about fit rather than delivering a verdict about price.

From there, explain the change you're proposing, whether it's a new tier, a usage-based add-on, or a custom price, and give them options. Nobody likes being told, "Your price is going up, period." People are much more receptive to "here are two options that fit your usage pattern better."

Give at least 30 days' notice before the change takes effect. That's standard, and anything less feels rushed. If the customer is on an annual plan, the change should take effect at renewal, not mid-contract.

When They Push Back

They might. And that's fine.

The two most common responses:

"Can you just cap my usage instead of raising my price?" This is a reasonable ask, and you should have a position on it. A soft limit, where the customer gets a notification when they hit a threshold but isn't immediately cut off, is usually better than a hard limit. A soft limit means the customer sees a warning ("you've used 90% of your included AI requests this month") but can keep going. A hard limit means the feature stops working. Hard limits protect your margin, but they also create the worst possible user experience at the exact moment the customer is getting the most value. If you're going to cap, cap softly and charge for overages. That's the difference between a soft limit and a hard limit, and for most SaaS products, soft wins.

"I'll just leave." That's their right. And honestly? Depending on the numbers, it might be the right outcome for you too.

It's OK If They Leave

This is the part nobody says out loud: sometimes the right answer is to let an unprofitable customer churn.

I know that feels wrong when you're a small SaaS with a small customer base. Every customer feels precious. But if a customer consistently costs you more in AI spend than they pay you, keeping them doesn't grow your business; it shrinks your margin. Every month they stay, you're effectively paying them to use your product.

That doesn't mean you push them out. It means you offer a fair path forward, a plan that covers the cost, or a usage structure that works for both sides, and if they don't take it, you let them make that choice. You didn't lose a customer. You stopped subsidizing one.

An unprofitable customer AI SaaS founders hold onto out of fear is more dangerous than the churn itself. The customer you should worry about losing is the one who pays you well and uses your product moderately. The one who costs you $80 to serve on a $49 plan? That's not a customer. That's a cost center.

Build This Into a Routine, Not a One-Off

If you're having this conversation for the first time, it probably means you haven't been checking your AI cost per customer regularly enough.

Here's what I'd build into my monthly routine:

• Pull up the per-customer cost view. Look at who's above the cost threshold you set in Step 1.
• Flag anyone who's been margin-negative for two or more cycles.
• For flagged customers, decide which of the four paths applies.
• If it's a structural issue (many customers trending the same way), it's a plan change. If it's one customer, it's a conversation.

That's it. Ten minutes a month to make sure no one is quietly eating your margin.

If you want to size the gap before you have the conversation, the free margin calculator can show you exactly how far off the numbers are.

If you've read this far, you probably already know which customer I'm talking about. You've seen the number. You've done the math. The hard part isn't the math; it's the email.

AI Observly is built to show you exactly which customers are profitable and which ones aren't, by customer, by feature, by plan, so you're walking into that conversation with real numbers, not guesses.