Your SaaS Bill Is a Fiction

By ExtraStrength

·5 May 2026·15 min read

You're building something.

You need search.

You find Algolia.

The slick sales team slimes you up and gets you ready.

Sign your soul, or at least, your companies, away.

Step one: "How many search requests do you expect per month?"

You have no users, or if you do, you have only a rough pen and paper idea of the next year.

You’re product in production? i mean, maybe its good, but your KPI says EXPLOSIVE GROWTH.

You have a landing page and a Figma file of whats happening over the next 5 years.

Now please, forecast your API volume.

Pick wrong and you'll either overpay for searches that never happened, or get slapped with overage charges at 1.5 to 2x the base rate when your product takes off.

Algolia's Grow tier charges $0.50 per 1,000 search requests over the included 10,000.

Go over? That's your problem.

Go way over? That's a very expensive problem.

This is the state of SaaS pricing in 2026. You're either paying per head, paying per guess, or paying per vibe (that's the enterprise "call us" tier).

None of it reflects what it actually costs to serve you - And I challenge that this will die.

The Seat Tax

Let's start with the obvious one. Per-seat pricing.

Salesforce Enterprise: $165 per user per month.

Slack Business+: $12.50 per user per month.

Jira: roughly $8 per user per month.

Microsoft 365: $57 per user per month.

Zoom Business: $18.32 per user per month.

HubSpot Professional: starts at $100 per user per month.

Now let's do the stack math for a 200 person company. Just those six tools.

Salesforce: $396,000 per year.

Microsoft 365: $136,800.

HubSpot: $240,000.

Slack: $30,000.

Jira: $19,200.

Zoom: $43,968.

That's $866,000 a year.

Call it $4,330 per employee just on six tools.

And the average enterprise company uses 275 SaaS applications.

Zylo's 2025 report, analysing 40 million licences across $40 billion in SaaS spend, pegged the average at $4,830 per employee per year. Other estimates run as high as $9,600.

Now the uncomfortable question.

What does it actually cost these companies to serve one more user?

SaaS gross margins run between 70% and 85%. Best in class companies hit 80% or higher. That means for every dollar you pay, roughly 15 to 20 cents covers the actual cost of goods sold.

Infrastructure.

Support.

Hosting.

Everything it takes to keep the lights on.

For a product like Salesforce, the marginal cost of user number 501 is close to zero. The servers are already running. The code is already written. The data centre is already humming.

Your $165 per month is not paying for compute.

It's paying to log in.

One SaaS infrastructure case study showed a company spending $45,000 per month on AWS for 500 customers. That's $90 per customer per month. After optimization, they got it down to $56. And that's the entire infrastructure bill, not the marginal cost of one more user.

Seat pricing isn't cost based. It's a tax on headcount. You hire 10 people, your SaaS bill jumps $50,000 before those people generate a single dollar of revenue. The tools don't cost more to run.

You just have more logins.

Lets Play A Guessing Game

Usage based pricing was supposed to fix this.

Pay for what you use.

Fair, transparent, aligned with value.

In practice, it means: predict the future or get punished.

Algolia is the perfect example.

You pick a tier based on estimated search requests.

Grow includes 10,000 requests per month. After that, you're paying $0.50 per 1,000.

The Grow Plus tier with AI features bumps that to $1.75 per 1,000.

The problem is that search volume isn't something you control.

Your users control it.

Your marketing controls it.

A viral TikTok controls it.

Every keystroke in a wrong implementation fires a new request.

Botnets hitting your site on a Sunday night control it.

One busy day and your bill looks nothing like your forecast.

Go over your committed volume?

Overage charges.

Pick a tier that's too high?

You're paying for ghost searches.

Here's the fun part, the free tier lets you have 1 million records in development, but drops to 100,000 on the paid Grow plan.

So you build your whole product against a generous sandbox, go to production, and immediately hit limits.

This isn't an Algolia problem.

This is a structural problem with the model.

Twilio charges per message and per minute.

AWS charges per compute hour.

Snowflake charges per credit.

Every one of these requires you to forecast usage before you know what usage looks like. And the vendor, who has vastly more data about usage patterns across their entire customer base, transfers all the forecasting risk to you.

That's backwards.

Stripe is the one usage-based model that actually works.

2.9% plus 30 cents per transaction.

Simple.

And crucially, your Stripe bill scales with your revenue.

More transactions means more money coming in. The cost correlates directly with the thing you care about.

Algolia's doesn't.

Your search volume has almost no correlation with your revenue.

A user searching 50 times and buying nothing costs you the same as a user who searches once and buys $10,000 worth of product.

You're paying for activity, not value.

Now ‘Add AI’ Boss

If seat pricing is a tax on humans, what happens when every human has two to five AI agents working on their behalf?

Microsoft answered first.

Copilot launched at $30 per user per month as an add on to existing M365 licences.

Then in March 2026, they announced M365 E7 at $99 per user per month, bundling Copilot, autonomous agents, and security tools.

Autonomous agent activity burns "Copilot Credits." Lol.

Need more?

Credit packs cost $200 for 25,000 credits per month.

So the seat price is $99.

But the actual cost depends on how much your AI agents do.

It's a seat-based model with usage-based pricing hiding inside it.

The worst of both worlds.

Salesforce had an even wilder ride.

Agentforce launched in October 2024 at $2 per conversation. It flopped.

Only about 8,000 of their 150,000+ customers signed up.

I led an implementation trial, it was the biggest dumpster fire mess of a product.

My napkin math after implementing the trial (all credited by salesforce, of course).

Five agents handling 70 conversations per day came to roughly $20,000 per month, once we spent the initial 60k credit.

We walked away.

They called up a year later, letting me know that ‘we at Salesforce listen, we pivoted’.

Then pivoted again.

By May 2025 they introduced Flex Credits at $0.10 per action.

Then per user licensing at $125 per user per month.

Three pricing models in eight months.

They're now running all three simultaneously.

And the sales team cant explain what is which.

Just that ‘Salesforce Next’ is whats next.

Anyway, its a paradox. If your AI agent does its job well, it replaces human work. Fewer humans means fewer seats.

Fewer seats means less revenue for the vendor.

Salesforce internally handled 380,000+ customer support interactions with its own Agentforce agents, 84% fully resolved without human intervention.

That's fantastic for Salesforce, proud of u. happy 4 u.

It's terrible for their seat based revenue model.

The better the AI gets, the less the vendor earns.

Nobody has solved this yet.

The Hot Seat Conundrum

A SaaS company running at 80% gross margin is keeping 80 cents of every dollar you pay.

Their cost of goods sold, the infrastructure, hosting, support, and maintenance required to serve you, is 20 cents.

For a mature platform at scale, the infrastructure cost per user is typically 8% to 15% of revenue.

On a $165 per month Salesforce Enterprise seat, that's roughly $13 to $25 in actual infrastructure cost.

You're paying $165 for something that costs them maybe $20 to deliver.

That's an 8x markup. On the marginal seat, it's closer to 50x.

Compare this to physical goods.

A protein bar costs $0.50 to $0.70 to make and retails for $3 to $5.

That's a 5 to 8x markup and I wrote an entire article about how wild those margins are.

SaaS makes protein bars look like a charity.

The industry standard SaaS COGS benchmark is 10% to 20% of revenue. Which means the other 80% to 90% goes to sales, marketing, R&D, and profit.

Not to serving you.

To acquiring and retaining you.

You're not paying for the product.

You're paying for the sales team that took you to dinner to buy their product.

In fact, you paid for dinner, just via their credit card.

Is It All Smoke And Mirrors?

A few models are emerging that make more sense.

Or at least make the math more honest.

Outcome-based pricing. Sierra AI only gets paid when their agent successfully resolves an issue without human intervention.

Pay for a job well done. They hit $100 million in ARR in just 21 months and crossed $150 million by early 2026. $10 billion valuation.

Bret Taylor, their co-founder and former Salesforce co-CEO, said "The whole market is going to go towards outcomes-based pricing. It's just so obviously the correct way to build and sell software."

Zendesk moved to outcome-based pricing for AI agents at $1.50 per automated resolution. You pay when the problem gets solved. Not when someone logs in. Not when a search query fires. When value gets delivered.

Flat-tier, use-it-however-you-want pricing. Some tools are moving toward capability-based tiers where you pay for a level of access and use it as much as you need. No per-seat multiplier. No usage metering. It's closer to how you buy electricity: pick a plan, use it, pay a predictable bill - Googles Vertex AI for Retail is heading this way, and coming for Algolias lunch, and id argue is now a stronger offering.

When you're evaluating SaaS, stop comparing sticker prices. Calculate your effective cost per unit of value.

What does each CRM record actually cost you on Salesforce versus a $50 per month alternative?

What does each search actually cost on Algolia versus building it yourself on Vertex, Typesense or Elasticsearch?

But you need to play both sides, some companies will self-manage Elasticsearch infrastructure running $50,000 to $300,000 annually for infrastructure alone, plus one to three full-time engineers at $150,000 to $250,000 in salary.

Algolia at scale might actually be cheaper for some companies.

You can't know that without running the numbers.

That's the whole point.

Nobody wants you to run the numbers.

The pricing models are designed to obscure the comparison, not enable it.

So What.

SaaS pricing is designed to extract maximum revenue while appearing fair. Seats, usage tiers, enterprise contracts.

They all obscure the same reality: the marginal cost to serve you is a fraction of what you pay.

That's fine when the value justifies it.

Salesforce at $165 per month might be worth every cent if it's generating $1,000+ per user in pipeline.

Algolia at $0.50 per 1,000 searches might be a bargain if those searches drive conversions.

The problem isn't the price.

The problem is that the pricing model has no relationship to the cost or the value.

It's just a number someone in a growth team picked because the market would bear it, and number will, unfortunately, go up. Not now, but, number will.

AI is about to stress test every one of these models.

When agents multiply the number of "users" and "requests" by 10x overnight, the seat tax and the guessing game both collapse.

The vendors that figure out how to charge for outcomes, not access, will win. The ones that keep taxing logins will watch their customers replace humans with AI and downgrade their seat counts.

If you're building SaaS: price on value delivered, not access granted. If you're buying SaaS: calculate your actual cost per unit of output. The number will either validate the spend or make you furious.

Either way, you should know what it actually costs.

Shout out to all the legends I've ripped info from for this piece:

Key takeaways

The SaaS industry runs on pricing models that bear no relationship to actual delivery costs. Seat-based pricing taxes headcount, usage-based pricing transfers forecasting risk to the buyer, and AI agents are about to break both. With SaaS gross margins at 80%+ and marginal per-user costs measured in cents, the gap between what you pay and what it costs to serve you has never been wider. Outcome-based pricing is emerging as the honest alternative.