SaaS Pricing and Packaging Playbook

Charge for a unit that rises with the value your customer gets, usually a base fee plus usage or credits; ladder three or four tiers on top; write your discount rules before the first negotiation; and pair every price increase with new value.

Price is the lever that needs no new product. In McKinsey’s classic analysis of 2,463 companies, a 1% price rise with no loss of volume lifted operating profit by 11.1%, against 3.3% for a 1% gain in volume^[1]. Those are old averages, not a SaaS benchmark, but the market is moving: the top 500 SaaS and AI companies with public pricing made more than 1,800 pricing changes in 2025, about 3.6 each^[2], and three in four companies in Kyle Poyar’s 2026 survey changed pricing or packaging within a year^[3].

In my view, the more common early-stage error is underpricing. It quietly caps revenue, weakens positioning (“if it’s cheap, it must be basic”) and starves the budget you need to acquire customers. This playbook stays with the SaaS decisions; for general research methods such as willingness-to-pay interviews, see the Pricing, Packaging & Monetization Playbook.

Value-based pricing sets your price by the value customers receive, not your cost to deliver or a competitor’s price. Done well, it captures a share of the surplus you create, and it lets price rise as the outcome you deliver grows. Cost still sets the floor: every AI action has an inference cost (Step 3).

The key is your value metric: the unit you charge for, chosen so it scales with the value a customer gets. Get it right and pricing feels fair: customers pay more only as they get more. Get it wrong, by charging for a unit customers don’t connect to value, and every price conversation becomes a fight. Some common choices:

• Payments tool: volume processed.
• Email platform: contacts or sends.
• Support tool: resolved tickets.
• API or infrastructure: calls or compute units.

Test each candidate with four questions. Does the bill rise when the customer gets more value? Can the buyer predict next month’s bill? Can you measure it reliably? And is it hard to game, so customers don’t ration their use to save money?

What if you don’t have a price yet?

This playbook assumes a price exists and asks whether it is the right one. If you have no price yet, the decisions come in a fixed order. Get the order wrong and you end up redoing the earlier ones.

1. Pick the value metric with the four tests above. That decides what you charge for. The amount comes later.
2. Find out who signs, and their limit. The user, the buyer and the budget holder are often three people. Ask one question in discovery: “If you wanted this, whose budget would it come from, and what can they approve without a second signature?” A price just under that limit closes faster than a “fair” price just over it.
3. Start with one paid plan. At pre-seed, one plan on the metric you expect to keep teaches you more than three tiers. Add tiers once buyers clearly differ, as Step 4 describes. A free plan is a separate decision with its own conditions (Step 3).
4. Test willingness to pay before you print it. Interview 8 to 10 target buyers and ask the four price-sensitivity questions. The method is in the Pricing, Packaging & Monetization Playbook, and the free Price Sensitivity Analyzer scores the answers. Look at the range, not the average.
5. Launch the price as a test with an end date. Write down the price, what would make you change it, and when you will look again: after one full sales cycle, not after the first objection.

In my view, anchoring on competitor price pages is the most common shortcut and the least useful. A competitor page shows what they charge. It doesn’t show what your buyer values or how much they can sign for.

It depends on how customers get value, but the direction is clear (Figure 2): hybrid pricing is gaining fastest, usage pricing in some form is already the norm, and seats hold on mainly in companies above $150M ARR. A quick map:

• Per-seat. Charge per user. Simple and predictable; best when value scales with team size. Under pressure where value doesn’t track headcount.
• Usage-based. Charge per unit consumed (API calls, compute, events). Aligns cost to value, but risks “bill shock” and less predictable revenue, so it needs metering and spend alerts.
• Hybrid. A base subscription plus usage or credits: a predictable floor plus upside from power users. In Benchmarkit’s survey of 316 SaaS companies for Maxio, hybrid companies reported the highest median growth, 21%; that is self-reported and correlational^[4].
• Tiered. Not a model but packaging, which sits on top of any of the three (Step 4).

For an AI product, I would rarely use pure per-seat pricing, and I would charge per outcome only when the outcome is easy to see, clearly caused by your product and written into the contract. For most, a hybrid (a base fee plus included usage or credits) is the safer default.

When each model fits. Per seat breaks when one user can run up heavy AI usage. Usage or credits fit when value and cost both rise with use, which is why credit models more than doubled in 2025 (Figure 2). Per outcome is still rare: in Bain’s June 2026 analysis of public pricing at about 200 B2B SaaS companies, about four in five vendors introducing AI pricing chose capacity models such as prepaid credits, and only about 10% of AI meters charged for outcomes^[6]. Figure 4 shows the split and Bain’s three conditions for outcome pricing.

How to set up outcome pricing

1. Define the outcome and what does not count. Intercom bills a Fin resolution when the customer confirms the answer helped or leaves without asking for more help, but not when the customer asks for a person, and it deducts the charge if the customer comes back to the same conversation^[7].
2. Set a floor and a cap. A base fee covers your fixed costs; spend alerts or hard limits protect the buyer. As of October 2026, Fin’s base plan for use with an existing helpdesk costs $49 a month including 50 resolutions, then $0.99 each, with minimum commitments^[8].
3. Price outcomes by value. Intercom bills a qualified prospect, such as a booked call or a started trial, at $9.99, and a prospect routed to self-serve at $0.99^[7].

How inference cost shapes a hybrid

Work out your cost per unit of value, not per token: the model, tool and hosting cost of one resolved ticket (the free AI Inference Cost Modeler does the arithmetic). Price each unit to clear your target gross margin, make one credit equal one unit the buyer understands, include typical use in the base fee and sell top-ups. Costs move fast: Epoch AI found in March 2025 that the price of reaching a fixed level of performance fell 9 to 900 times a year, depending on the milestone, though the fastest drops may not last^[9]. Review credit rates and margins every quarter.

Should an AI product have a free tier?

Only if you can cap what each free user costs and free users turn into paying customers. In my view that also needs a product that spreads through use, an upgrade trigger tied to your value metric, and free users from companies you want as customers.

Work out what a free user costs (AI actions a month times your cost per action, plus hosting and support), then divide what free users cost by the number who upgrade. Illustrative only: at $0.40 a month for six months, with 3 in 100 upgrading, free costs $80 per paying customer. Compare that with what a customer costs to win elsewhere.

Cap free use by actions, credits, model or time; at launch in December 2024, GitHub Copilot Free gave 2,000 code completions and 50 chat messages a month^[10]. Stop free use at the cap and add sign-up checks against abuse.

If those conditions don’t hold, use a free trial, a reverse trial (paid features for a set time, then a limited free plan), a guided pilot, or bring your own key, where the user pays the model provider. Cursor, for example, lets users add their own API keys for its chat models; the provider bills them directly, and on Teams and Enterprise plans Cursor still charges $0.25 per million tokens^[11]. My article on trials, freemium and demos compares the options.

In my view, three to four tiers. Every extra option is one more comparison before a buyer can choose. The research on choice overload is mixed, but a meta-analysis of 99 observations found it does appear, depending on how complex the options are, how hard the decision is, how unsure buyers are of what they want and what they are trying to achieve^[12]. That describes most plan pages. Good, Better, Best works because it gives most buyers an obvious middle choice while anchoring the range.

That anchoring is the part founders underuse. A meaningfully higher top tier makes the middle look reasonable; in Simonson’s classic experiments, options gained share when they became the compromise between two others^[13]. If your top plan is only slightly above the rest, you are probably under-charging the customers who would pay more. The gaps between tiers, your price ladder, should make each step up feel like a different product, not a rounding error.

Adding a top tier above Free and Pro

A new top tier has two jobs. It sells to the buyers who need more, and it makes Pro look reasonable. It does the second only if it sits far enough above Pro, and the first only if it unlocks a bigger job, not just higher limits. Admin controls, security review support, SSO and priority support are the usual candidates because they matter to the larger buyer.

The table below uses illustrative tiers, not Example Co.’s, which has no free plan. Pro is $40, and 10 of every 100 paying customers move to Business. Free users are left out because they add no revenue.

Even at $100, Pro still carries most of the revenue. The point of the top tier isn’t to replace Pro. It catches the buyers who would have paid more anyway. Check your own mix in the free Pricing & ARPU Modeler (Step 5).

Does pricing psychology work?

Within reason. In three field experiments at a retailer, prices ending in 9 raised demand in all three^[14]; the usual explanation is that $29 reads closer to $20 than to $30. Those were retail products sold to consumers, not software, so test the effect on your own traffic. Psychology polishes a sound structure; it can’t rescue a price disconnected from value.

Should implementation be a separate fee?

Charge a one-time fee when setup is real work that varies by customer: custom integrations, data migration, single sign-on. Bundle it into the subscription or the higher tiers when setup is light and standard, or when a fee would slow the self-serve and small deals you need. Either way, your price has to cover it.

Size the fee from hours by role at each person’s loaded hourly cost, plus third-party costs and a margin for scope risk. In my view, expect procurement to ask you to waive it; keep the scope written down and trade a waiver for a longer term or annual prepayment. A one-time fee is not recurring revenue, so it does not add to ARR.

Not from your cheapest tier. Your blended ARPU, the average revenue per account across your whole tier mix, tells you whether the packaging works, and it has to clear your CAC and payback. Then look at the revenue mix, the share of revenue each tier drives. If most comes from the entry tier, your upgrade path isn’t pulling; in my view, healthy packaging concentrates revenue in the middle and top tiers, where the value and the willingness to pay are highest. Two companies with identical list prices can have very different economics depending on where their customers land.

The free Pricing & ARPU Modeler computes blended ARPU and revenue mix from your tiers, and the Unit Economics Calculator checks that ARPU against CAC and payback.

Give procurement four things before they ask, defend the price with value, and trade any concession for something in return.

What procurement needs

1. A one-page value case. The problem, today’s baseline and the improvement in money, from the customer’s own numbers, plus what doing nothing costs each month.
2. How the price is built. Value metric, tier, inclusions and list price.
3. The alternatives. Building in-house, a competitor or no change, with an honest note on each.
4. Risk answers. Security answers, any certification you actually hold, a data processing agreement, service levels and references.

Discount rules to write down first

• Who can approve which discount level, set before any deal is live.
• Every discount has a logged reason and an end date.
• No discount without a “get” in return.

In my view, the best gets are annual prepayment, a longer term, a larger commitment, a faster signature date, a wider rollout, or permission to use the customer as a reference.

• Can concede: payment terms, a ramped start, extra onboarding, a one-time fee or a capped renewal increase.
• Hold: your value metric, the list price on paper (show the discount as a separate line with an end date) and, for an AI product, a usage cap, because inference costs rise with use.

How big an annual discount?

I found no study with its own sample that shows which annual-prepay discount works best at Seed, and “two months free” is arithmetic, not a benchmark: about 17% off (Figure 7). Set yours this way:

1. List what it buys and costs. Cash on day one and one renewal decision instead of twelve, but you also give it to customers who would have stayed anyway.
2. Find your break-even from churn. The break-even discount is 1 minus (months a monthly customer pays in its first year ÷ 12). Annual buyers may already be your most committed, so treat it as an upper bound.
3. Price the cash. A larger discount pays only if prepayment replaces money you would otherwise raise or borrow.
4. Don’t count on retention. SaaS Capital’s 2025 survey of more than 1,000 private B2B SaaS companies found “Annual contracts show 101% median NRR and 90% median GRR”, against 100% and 89% for month-to-month^[15]. That compares companies; it is not the effect of moving one customer to annual.
5. Trade before you discount. Offer a price lock at renewal or extra onboarding first.

Carefully. What follows is my view, not a published finding. I’d protect existing customers when you raise prices: give loyal customers a grace period sized to your contract and renewal cycle, move in steps rather than one jump, and tie the increase to new value the customer can feel. In my view, “here’s more, and here’s the new price” is far easier to accept than the same product at a higher price. Notion paired its increase with new value: it folded its AI add-on into its Business plan and raised that plan from $15 to $20 per user a month^[2].

Then keep reviewing. Your competitors are changing (see the introduction): of the top 500’s changes in 2025, 43% were to feature packaging, 40% to the price or its structure and 17% to usage limits^[2]. Revisit packaging and price once or twice a year, test willingness to pay on real customers (the Price Sensitivity Analyzer runs the Van Westendorp questions), and track ARPU, net revenue retention and expansion.

What changes by stage?

What mistakes should I avoid?

• Charging for a unit customers don’t connect to value. Run the four tests in Step 1.
• Billing for an outcome the buyer can dispute. Define what counts and who records it (Step 3).
• Unlimited free use of your most expensive feature. Cap it (Step 3).
• A top tier only slightly above the rest. It anchors nothing (Step 4).
• Waiving the implementation fee, then absorbing custom work nobody priced. Trade the waiver for something (Step 4).
• Discounting before procurement asks. Your first offer becomes the new starting point.
• A deep annual discount that hides weak retention until the first renewal. Set it from your churn break-even (Step 6).