Usage-Based Pricing for SaaS: The CEO’s Guide to Getting It Right

Usage-based pric­ing for SaaS is being sold to you as the future, and it might be — but most of the founders rush­ing toward it are about to trade a fore­castable busi­ness for an unfore­castable one with­out under­stand­ing what they’re giv­ing up. The pitch is seduc­tive: charge cus­tomers for what they actu­al­ly con­sume, land small­er, expand auto­mat­i­cal­ly, ride the AI wave. All true, some­times. What nobody on the billing-ven­dor blogs will tell you is that the same mechan­ic that makes usage-based pric­ing expand rev­enue on the way up makes it con­tract rev­enue on the way down — and an acquir­er pay­ing you a mul­ti­ple of recur­ring rev­enue cares enor­mous­ly about which way you’re point­ed.

I’ve watched founders bolt a con­sump­tion meter onto a busi­ness that had no usage vari­ance worth meter­ing, and I’ve watched founders cling to per-seat pric­ing while their AI fea­ture qui­et­ly destroyed their gross mar­gin. Both are expen­sive mis­takes. This guide is about mak­ing the call delib­er­ate­ly.

I’m going to cov­er what usage-based pric­ing actu­al­ly is, the four mod­els you’ll choose between, the unit eco­nom­ics that decide whether it helps or hurts you, the spe­cif­ic way it changes your reten­tion math, the AI-dri­ven shift that’s push­ing every­one toward it right now, and — most impor­tant — what a buy­er thinks when they see “usage-based” on your rev­enue. If you want the broad­er menu of how SaaS com­pa­nies charge, start with the SaaS pric­ing mod­els guide; this arti­cle goes deep on one mod­el and the deci­sion to adopt it.

What Usage-Based Pricing Actually Is

Usage-based pric­ing (also called con­sump­tion-based or metered pric­ing) charges cus­tomers in pro­por­tion to how much they use your prod­uct, rather than a flat fee per user or per month. If a cus­tomer runs 2 mil­lion API calls, sends 500,000 emails, or process­es 80,000 trans­ac­tions, they pay for 2 mil­lion, 500,000, or 80,000 — not for a seat, and not for a tier they may or may not fill.

Con­trast that with the two pric­ing mod­els most SaaS com­pa­nies start with:

Model	What you charge for	Revenue behavior	Forecastability
Per-seat subscription	Number of user licenses	Fixed monthly until seats change	High — you know next month's revenue today
Flat / tiered subscription	Access to a feature tier	Fixed monthly until tier changes	High
Usage-based	Actual consumption (calls, GB, transactions, compute)	Floats up and down with customer activity	Low — depends on what customers do
Hybrid	A base subscription plus metered overage	Floor is fixed; upside floats	Medium

That last col­umn is the one founders under­weight. Per-seat pric­ing gives you the recur­ring rev­enue pre­mi­um — the rea­son SaaS gets high­er val­u­a­tion mul­ti­ples than almost any oth­er busi­ness mod­el is that the rev­enue is pre­dictable. You can finance growth against it, plan hir­ing around it, and an acquir­er can under­write it. Pure usage-based pric­ing trades some of that pre­dictabil­i­ty for the chance at faster expan­sion. Whether that trade is smart depends entire­ly on your busi­ness — which is the whole point of this arti­cle.

One def­i­n­i­tion­al note before we go fur­ther. Peo­ple use “usage-based,” “con­sump­tion-based,” and “metered” inter­change­ably, and for prac­ti­cal pur­pos­es they’re the same idea: you meter a unit and bill for it. “Out­come-based” pric­ing is a new­er, dis­tinct thing — charg­ing for a result (a resolved sup­port tick­et, a closed deal) rather than a unit of con­sump­tion. I’ll treat out­come-based as the fron­tier case at the end, because it’s where AI is push­ing the mar­ket and it car­ries its own risks.

The Four Usage-Based Models You’ll Choose Between

“Usage-based” isn’t one pric­ing scheme — it’s a fam­i­ly. The four struc­tures below dif­fer in how much risk they push onto the cus­tomer ver­sus keep on you, and that risk allo­ca­tion is the real deci­sion.

1. Pay-as-you-go. The cus­tomer pays a per-unit rate for exact­ly what they con­sume, with no com­mit­ment. Cloud infra­struc­ture (AWS, GCP, Azure) is the canon­i­cal exam­ple — you’re billed per com­pute-hour or per GB. Low­est fric­tion to land, low­est rev­enue pre­dictabil­i­ty for you. Best when the cus­tomer gen­uine­ly can’t pre­dict their own usage and would balk at com­mit­ting.
2. Tiered (grad­u­at­ed) usage. The per-unit price changes as vol­ume cross­es thresh­olds — the first 10,000 units cost X each, the next 40,000 cost less each, and so on. This rewards growth and gives larg­er cus­tomers a vol­ume dis­count with­out a sep­a­rate nego­ti­a­tion. Most metered SaaS lands here.
3. Pre­paid cred­its / com­mit­ment. The cus­tomer com­mits to a pool of usage up front (often annu­al­ly) and draws it down. This is the mod­el that recov­ers most of the recur­ring-rev­enue pre­dictabil­i­ty you lose with pure pay-as-you-go: you book the com­mit­ment as recur­ring, the cus­tomer pays whether or not they ful­ly con­sume, and over­age bills on top. If you go usage-based and care about your fore­cast, this is usu­al­ly where you want to steer larg­er accounts.
4. Hybrid (base + over­age). A fixed sub­scrip­tion floor plus metered charges above an includ­ed allot­ment. This is the most com­mon struc­ture among grow­ing SaaS com­pa­nies for a rea­son — it pre­serves a pre­dictable base (the floor an acquir­er can under­write) while cap­tur­ing expan­sion as cus­tomers con­sume more. Rough­ly speak­ing, the base pro­tects your down­side and the meter cap­tures your upside.

If I had to give one default to a $5M–$15M ARR founder adopt­ing metered pric­ing for the first time, it’s hybrid for the base of the mar­ket, pre­paid com­mit­ments for the enter­prise top. That com­bi­na­tion keeps the pre­dictable recur­ring rev­enue that dri­ves your val­u­a­tion while let­ting the meter do the expan­sion work.

The Decision: Does Usage-Based Pricing Even Fit Your Business?

Here’s the ques­tion almost nobody asks before switch­ing: is there enough vari­ance in how cus­tomers use my prod­uct to make meter­ing worth it?

Usage-based pric­ing only cre­ates val­ue when cus­tomer con­sump­tion varies a lot AND that vari­ance cor­re­lates with the val­ue cus­tomers receive. If every cus­tomer uses rough­ly the same amount, meter­ing just adds billing com­plex­i­ty for no gain — you could have set a flat price at the aver­age and saved every­one the spread­sheet. If con­sump­tion varies wild­ly but does­n’t track val­ue (cus­tomer A burns 10x the com­pute of cus­tomer B but gets no more busi­ness val­ue from it), then meter­ing charges your heav­i­est users the most while deliv­er­ing them the least rel­a­tive val­ue, and they’ll churn or rene­go­ti­ate.

The way to think about this is the same way you should think about every­thing in pric­ing: pack­age around what cus­tomers actu­al­ly use and where they get val­ue. Cus­tomers resent pay­ing for things they don’t use — that’s the “why am I pay­ing for that?” con­ver­sa­tion that pre­cedes a dis­count request or a churn. Usage-based pric­ing is, at its best, the clean­est pos­si­ble answer to that objec­tion: you pay for what you use, full stop. But it’s only clean if your unit of mea­sure is a unit of val­ue. Pick the wrong meter — one that tracks your cost instead of their val­ue — and you’ve built a pric­ing mod­el your cus­tomers will fight.

Run this fil­ter before you com­mit:

Test	Usage-based fits	Usage-based doesn't fit
Consumption variance across customers	High (10x+ between light and heavy users)	Low (everyone uses about the same)
Does the metered unit track customer value?	Yes — more usage means more value received	No — usage tracks your cost, not their benefit
Can the customer predict their own usage?	Either way works (commitment for no, PAYG for yes)	—
Is your gross margin stable per unit of usage?	Yes	No — heavy usage destroys your margin (the AI trap, below)
Do you need a forecastable revenue base?	Use hybrid or prepaid, not pure PAYG	Pure PAYG will wreck your forecast

If you fail the first two tests, the answer is to fix your pric­ing strat­e­gy with­in a sub­scrip­tion mod­el, not to bolt on a meter.

The Unit Economics: Why Usage-Based Pricing Can Cut Both Ways

This is where the billing-ven­dor guides go qui­et, because it’s where the mod­el can hurt you.

Per-seat SaaS has a beau­ti­ful prop­er­ty: your cost to serve a cus­tomer is rough­ly fixed, so as you raise prices or the cus­tomer adds seats, almost all of the incre­men­tal rev­enue drops to gross mar­gin. Usage-based pric­ing breaks that prop­er­ty when your cost of goods sold scales with usage — which is exact­ly the sit­u­a­tion AI fea­tures cre­ate.

Walk through it with real­is­tic num­bers. Sup­pose you charge $0.10 per unit of usage and, for a tra­di­tion­al soft­ware fea­ture, your mar­gin­al cost per unit is near zero. A cus­tomer who con­sumes 100,000 units pays you $10,000 and costs you almost noth­ing to serve — gross mar­gin near 100% on that rev­enue. Won­der­ful.

Now add an AI fea­ture where each unit of usage trig­gers a mod­el infer­ence that costs you $0.04 in com­pute. Same $0.10 price, same 100,000 units, same $10,000 of rev­enue — but now your cost of goods sold is $4,000. Your gross mar­gin on that usage just fell to 60%. If a com­peti­tor forces you to price at $0.07 per unit to win the deal, you’re at $7,000 rev­enue against $4,000 cost — 43% gross mar­gin, which for a SaaS busi­ness is the dif­fer­ence between a pre­mi­um mul­ti­ple and a dis­count one. (These fig­ures are illus­tra­tive and meant to show the direc­tion and size of the mar­gin effect, not cur­rent mar­ket rates — ver­i­fy your own per-unit cost before pric­ing.)

Scenario	Price / unit	Cost / unit	Revenue (100K units)	COGS	Gross margin
Traditional feature	$0.10	~$0.00	$10,000	~$0	~100%
AI feature, full price	$0.10	$0.04	$10,000	$4,000	60%
AI feature, discounted to win	$0.07	$0.04	$7,000	$4,000	~43%

This is the trap. Usage-based pric­ing on top of a usage-based cost can com­press your mar­gin pre­cise­ly when a cus­tomer ramps up — the moment you thought was your win. The fix isn’t to avoid usage-based pric­ing; it’s to price the meter above your ful­ly-loaded mar­gin­al cost with enough head­room to sur­vive com­pet­i­tive pres­sure, and to watch gross mar­gin by fea­ture, not just com­pa­ny-wide. If your blend­ed mar­gin looks healthy because lega­cy high-mar­gin rev­enue is mask­ing a bleed­ing AI line, you have a prob­lem hid­ing in plain sight — the kind that sur­faces in dili­gence at the worst pos­si­ble moment. For the full treat­ment of how to allo­cate and read these costs, see cost of goods sold for SaaS with­in your pric­ing mod­el and the broad­er SaaS unit eco­nom­ics frame­work.

The dis­ci­pline here is the one I push on every CEO: every­thing con­nects back to unit eco­nom­ics. A pric­ing mod­el that looks like a growth sto­ry on the rev­enue line can be a mar­gin sto­ry going the wrong way under­neath. Decide it delib­er­ate­ly. If you choose to com­press mar­gin to take mar­ket share fast — a legit­i­mate move when cap­i­tal allows and costs will fall — make it an explic­it, eyes-open trade-off, not an acci­dent you dis­cov­er lat­er.

How Usage-Based Pricing Changes Your Retention Math

Per-seat sub­scrip­tion rev­enue is sticky in an obvi­ous way: until the cus­tomer can­cels or removes seats, the rev­enue recurs. Usage-based rev­enue is sticky in a sub­tler and more dan­ger­ous way — it recurs only as long as the cus­tomer keeps using.

This shows up direct­ly in your reten­tion met­rics, and the met­rics that dri­ve your val­u­a­tion behave dif­fer­ent­ly under metered pric­ing:

• Net rev­enue reten­tion (NRR) can be spec­tac­u­lar under usage-based pric­ing when cus­tomers ramp. Because expan­sion hap­pens auto­mat­i­cal­ly as con­sump­tion grows — no upsell motion, no new con­tract — healthy usage-based busi­ness­es often post NRR well above seat-based peers. That’s the upside the mar­ket is pay­ing for. (For the mechan­ics of how NRR com­pounds, see net rev­enue reten­tion.)
• Gross rev­enue reten­tion (GRR), though, is where the risk lives. GRR mea­sures how much of your exist­ing rev­enue you keep before any expan­sion. Under usage-based pric­ing, a cus­tomer who sim­ply uses less — a sea­son­al dip, a bud­get freeze, an inter­nal slow­down — shrinks your rev­enue with­out ever churn­ing. They’re still a cus­tomer. They did­n’t can­cel. But your rev­enue from them fell. Seat-based busi­ness­es don’t have this fail­ure mode at any­where near the same mag­ni­tude.

Here’s the part founders get wrong about NRR: net rev­enue reten­tion nets expan­sion against con­trac­tion and churn. A usage-based busi­ness can post 115% NRR while qui­et­ly car­ry­ing 85% gross reten­tion — the expan­sion from ramp­ing cus­tomers masks real rev­enue ero­sion from shrink­ing ones. A seat-based busi­ness with 115% NRR is almost always health­i­er than a usage-based busi­ness with the same num­ber, because the usage-based ver­sion has more rev­enue at risk to behav­ior rather than locked by con­tract.

The com­pound­ing math is unfor­giv­ing in both direc­tions. If your usage-based cohort grows rev­enue at 1.15x per year, after three years it’s worth 1.15³ ≈ 1.52x of where it start­ed — that’s the dream. But if a macro slow­down flips that cohort to 0.90x per year, after three years it’s at 0.90³ ≈ 0.73x — you’ve lost 27% of that rev­enue with­out los­ing a sin­gle logo. Per-seat rev­enue does­n’t swing like that, because behav­ior between renewals can’t touch it.

The prac­ti­cal impli­ca­tion: if you adopt usage-based pric­ing, track gross rev­enue reten­tion as care­ful­ly as net, and build the pre­dictable floor (hybrid base, pre­paid com­mit­ments) pre­cise­ly so your down­side in a slow quar­ter is bound­ed. This is the same instinct behind pre­fer­ring con­trac­tu­al­ly recur­ring rev­enue — it sta­bi­lizes oper­at­ing cash flow and makes a por­tion of your rev­enue pre­dictable, which is exact­ly what you want head­ing into a reces­sion or a sale.

The AI Shift: Why This Is Suddenly Everyone’s Question

You’re read­ing about usage-based pric­ing now for a spe­cif­ic rea­son: AI is break­ing the per-seat mod­el.

For two decades, SaaS was priced per seat because soft­ware was a tool a human oper­at­ed, and seats were a clean proxy for val­ue — more users, more val­ue, more rev­enue. AI agents break that proxy. When the soft­ware does the work instead of assist­ing a human doing the work, “num­ber of seats” stops cor­re­lat­ing with val­ue deliv­ered. A com­pa­ny might deploy one AI agent that does the work of ten peo­ple; charg­ing for one seat would mas­sive­ly under­price it, and charg­ing for ten seats is a fic­tion nobody will pay.

So the mar­ket is mov­ing. Usage-based pric­ing went main­stream over the last few years — well over half of SaaS com­pa­nies now use some metered com­po­nent — and hybrid mod­els (a sub­scrip­tion base plus usage) are pro­ject­ed to be the major­i­ty struc­ture across SaaS by the end of 2026. The fron­tier is out­come-based pric­ing: charg­ing for the result the AI pro­duces (a resolved tick­et, a qual­i­fied lead, a com­plet­ed task) rather than the units it con­sumed get­ting there. Ear­ly data sug­gests com­pa­nies lay­er­ing in out­come-based com­po­nents see mean­ing­ful­ly high­er reten­tion and sat­is­fac­tion — because the cus­tomer only pays when they got the thing they want­ed.

Out­come-based pric­ing is the purest pos­si­ble answer to “pack­age around where the cus­tomer gets val­ue.” It’s also the hard­est to oper­ate, for two rea­sons this arti­cle has already armed you to see:

1. The mar­gin trap is worse. When you charge for an out­come but your cost to pro­duce it is vari­able AI com­pute, a hard out­come (one that takes the mod­el many expen­sive attempts) can cost you more than you charge. You’re now under­writ­ing the vari­ance in your own pro­duc­tion cost, not just the cus­tomer’s usage. Price the floor care­ful­ly or you’ll lose mon­ey on your best cus­tomers.
2. The mea­sure­ment prob­lem is real. You and the cus­tomer have to agree on what counts as the out­come, and that agree­ment has to sur­vive an audi­tor in dili­gence. Only a small frac­tion of SaaS com­pa­nies had imple­ment­ed true out­come-based pric­ing as of a cou­ple of years ago, pre­cise­ly because con­sis­tent­ly mea­sur­ing out­comes is hard.

My read for a $5M–$15M ARR CEO: the AI shift is real and you should not be defend­ing per-seat pric­ing for an AI fea­ture that deliv­ers val­ue dis­con­nect­ed from seat count. But move toward usage-based — and only exper­i­ment with out­come-based on a con­tained slice — with the mar­gin dis­ci­pline and reten­tion track­ing from the sec­tions above. The com­pa­nies get­ting hurt right now are the ones that changed the pric­ing mod­el with­out chang­ing how they watch the eco­nom­ics under­neath it.

What an Acquirer Thinks When They See “Usage-Based”

If you’re build­ing toward an exit — and most of the CEOs I work with are — then the most impor­tant audi­ence for your pric­ing mod­el isn’t your cus­tomers. It’s the acquir­er who’ll one day under­write your rev­enue and pay you a mul­ti­ple of it. Get this part wrong and you can grow rev­enue while shrink­ing your mul­ti­ple.

Here’s how a sophis­ti­cat­ed buy­er reads usage-based rev­enue, and it maps direct­ly onto the six things that actu­al­ly dri­ve a rev­enue mul­ti­ple:

What the buyer evaluates	How usage-based pricing affects it
Recurring nature of revenue	Pure pay-as-you-go reads as less contractually recurring than a subscription — a multiple risk. Prepaid commitments and hybrid bases recover this.
Predictability	Floating revenue is harder to forecast; buyers discount uncertainty. A predictable base mitigates the discount.
Growth rate	High NRR from automatic expansion is genuinely attractive and can lift the multiple — this is the real upside.
Gross margin	The AI margin trap shows up here. A buyer will model margin by line; a bleeding AI meter caps your multiple.
Revenue at risk to behavior	Buyers stress-test GRR. Revenue that can shrink without a churn event is scored as riskier than contracted revenue.
Concentration	If one or two heavy-usage accounts drive the meter, that's concentration risk — one of the surest multiple killers.

The head­line: usage-based pric­ing is not inher­ent­ly good or bad for your val­u­a­tion. It’s a high­er-vari­ance bet. Done well — a metered mod­el where the unit tracks val­ue, mar­gin holds, a pre­dictable base anchors the rev­enue, and NRR runs hot — it can earn you a pre­mi­um mul­ti­ple on a faster-grow­ing top line. Done care­less­ly — pure pay-as-you-go, a mar­gin-bleed­ing AI meter, rev­enue con­cen­trat­ed in a few accounts that can qui­et­ly ramp down — it can hand a buy­er every excuse to dis­count you.

The recur­ring-rev­enue pre­mi­um is real and it’s still the foun­da­tion of SaaS val­u­a­tion. Usage-based pric­ing asks you to give up a slice of that pre­dictabil­i­ty in exchange for expan­sion upside. Whether that’s a good trade is, like most things in this busi­ness, a unit-eco­nom­ics ques­tion — and the answer should be delib­er­ate, defen­si­ble, and vis­i­ble in your num­bers long before a buy­er ever asks. For how all of this rolls up into a num­ber, see SaaS rev­enue mul­ti­ples and SaaS val­u­a­tion mul­ti­ples.

How to Roll Out Usage-Based Pricing Without Breaking Your Business

If you’ve decid­ed usage-based pric­ing fits, the roll­out is where founders cre­ate avoid­able dam­age. A few rules:

1. Don’t flip your whole base at once. Intro­duce the metered mod­el to new cus­tomers and new fea­tures first. Forc­ing a repric­ing on your installed base is a churn event wait­ing to hap­pen — the same cau­tion that applies to any price increase, and meter­ing changes the con­tract more than a sim­ple per­cent­age bump.
2. Pick a meter your cus­tomer can see and trust. The unit has to be some­thing the cus­tomer can mon­i­tor, pre­dict, and con­nect to val­ue. A meter the cus­tomer can’t see is a billing dis­pute machine.
3. Build the pre­dictable floor in from day one. Hybrid base or pre­paid com­mit­ment — not because pure pay-as-you-go can’t work, but because the floor is what pro­tects your fore­cast, your cash flow, and your mul­ti­ple. You can always let cus­tomers con­sume above the floor; you can’t eas­i­ly un-float rev­enue after the fact.
4. Instru­ment mar­gin by fea­ture before you scale the meter. Espe­cial­ly for any­thing AI-backed. Know your ful­ly-loaded cost per unit and price above it with head­room. Watch gross mar­gin per fea­ture, not just blend­ed.
5. Track GRR along­side NRR. Set up the report­ing so a quar­ter of qui­et usage decline can’t hide behind expan­sion. The num­ber that warns you of trou­ble is gross reten­tion, not net.

Usage-based pric­ing is a pow­er­ful lever — and pric­ing is one of the most pow­er­ful levers you have, because changes flow almost straight to the bot­tom line and, from there, to your val­u­a­tion. But pow­er cuts both ways. The founders who win with it are the ones who treat­ed the switch as a unit-eco­nom­ics deci­sion with a delib­er­ate risk trade-off, not as a trend to fol­low. Decide it the same way you’d decide any major lever: with the math in front of you and the exit in mind.

Frequently Asked Questions

What is usage-based pricing for SaaS?

Usage-based pric­ing for SaaS charges cus­tomers in pro­por­tion to how much they con­sume — API calls, trans­ac­tions, giga­bytes, com­pute, or anoth­er metered unit — rather than a fixed fee per user or per month. It’s also called con­sump­tion-based or metered pric­ing, and it comes in pay-as-you-go, tiered, pre­paid-cred­it, and hybrid forms.

Is usage-based pricing better than per-seat subscription?

Nei­ther is uni­ver­sal­ly bet­ter. Usage-based pric­ing fits when cus­tomer con­sump­tion varies a lot and that vari­ance tracks the val­ue cus­tomers receive; per-seat pric­ing fits when usage is rough­ly uni­form and you val­ue rev­enue pre­dictabil­i­ty. Most grow­ing SaaS com­pa­nies land on a hybrid — a sub­scrip­tion base plus metered usage — to keep a fore­castable floor while cap­tur­ing expan­sion.

How does usage-based pricing affect net revenue retention?

It can raise net rev­enue reten­tion (NRR) because cus­tomers expand auto­mat­i­cal­ly as they con­sume more, with no upsell motion required. The catch is gross rev­enue reten­tion (GRR): under usage-based pric­ing, a cus­tomer can shrink your rev­enue sim­ply by using less, with­out ever churn­ing. Track GRR as care­ful­ly as NRR so qui­et usage declines don’t hide behind expan­sion.

Why is AI pushing SaaS toward usage-based pricing?

AI agents break the per-seat mod­el because the soft­ware does the work instead of assist­ing a human, so “num­ber of seats” no longer tracks val­ue deliv­ered. That’s push­ing the mar­ket toward usage-based and, at the fron­tier, out­come-based pric­ing — charg­ing for the result rather than the seat. The risk is mar­gin: AI com­pute makes your cost of goods sold scale with usage, so the meter has to be priced above your ful­ly-loaded mar­gin­al cost.

Does usage-based pricing hurt my valuation at exit?

Not inher­ent­ly — but it’s a high­er-vari­ance bet. Acquir­ers val­ue con­trac­tu­al­ly recur­ring, pre­dictable rev­enue, so pure pay-as-you-go can read as less recur­ring and get dis­count­ed. A pre­dictable base (hybrid or pre­paid com­mit­ments), sta­ble gross mar­gin, low cus­tomer con­cen­tra­tion, and strong gross reten­tion let usage-based pric­ing earn a pre­mi­um on a faster-grow­ing top line instead of a dis­count.