.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Conjoint Analysis vs Direct Price Testing Finding the Best Method for SaaS Pricing Optimization
SaaS companies often approach pricing with a deceptively simple question: “What price will convert?” That question is useful, but it is rarely the question that determines whether a pricing change succeeds. A $99 plan can fail because the buyer wanted an annual commitment, because SSO belonged in the plan rather than an add-on, or because the company charged by seat when the buyer expected to pay per resolved support case.
The stakes are higher in 2026 because SaaS offers now combine subscriptions, usage allowances, credits, overages, AI features, and enterprise services. Cursor charges a recurring fee while also managing model usage. GitHub Copilot combines granted seats, included AI credits, and usage-based overage. Intercom combines helpdesk seats with Fin outcome charges. A direct price test can show whether one displayed price changes immediate conversion. It cannot reliably tell a leadership team whether the underlying offer is built correctly.
Monetizely’s position is clear: conjoint analysis is the better primary investment for SaaS pricing optimization when a company is changing packaging, a pricing metric, or segment coverage. Direct price testing belongs at the end of the process, as controlled validation of a chosen offer, not as the method that designs the offer.
Software pricing needs a design answer before it needs a number
Conjoint analysis presents buyers with competing offers that vary across several attributes, including price, and asks them to choose. A well-built study can estimate the trade-offs a prospect will make between a lower price, stronger security, more included usage, premium support, annual terms, or an outcome-based metric.
Direct price testing works differently. We use the term narrowly: a company randomly exposes comparable live prospects to different fully disclosed prices for the same offer, then observes paid conversion, revenue, activation, and later retention. Randomization can establish a causal relationship between a displayed price and a measured outcome. It does not reveal how demand would change if the company also changed the package, meter, or target buyer. A 2018 Management Science field experiment illustrates the strength of randomized prices for causal inference, while also underscoring how carefully such tests must be designed.
Exhibit 1: The methods answer different questions, but SaaS leaders should sequence them rather than treat them as substitutes
| Decision question | Conjoint analysis | Direct price testing | Monetizely’s call |
|---|---|---|---|
| What does the buyer evaluate? | A set of competing offers with different features, terms, meters, and prices | One live offer at two or more prices | Start with conjoint when the offer itself is under review |
| What can the company learn? | Relative value of features, packages, price points, and price metrics | Actual response to a specific price in a live buying flow | Use direct tests to confirm a final rate |
| How does it handle segments? | Can estimate different preferences for SMB, mid-market, and enterprise buyers | Requires enough traffic and conversions within each segment | Conjoint has the stronger read on segment design |
| What is its main evidence? | Stated choices under controlled trade-offs | Paid buyer behavior | Behavioral evidence is stronger only when the tested offer is already right |
| Where can it fail? | Hypothetical choices can overstate real purchase intent | Small samples, traffic mix, sales intervention, and short test windows can distort results | Build realism into conjoint, then protect randomization in live testing |
The table points to a simple distinction: conjoint helps leaders decide what to sell and how to charge, while direct testing helps them validate the selected price for that defined offer.
Monetizely’s 5-Step Pricing Framework begins with the idea that a price is the visible end of a chain of decisions, not the chain itself. As discussed in Monetizing Agentic AI, the framework moves from commercial intent to a workable price system through five connected decisions:
The sequence matters because each step constrains the next one. A company that wants penetration in startup accounts may need a low-friction self-serve plan. A company selling into regulated enterprises may need security, onboarding, invoicing, and annual commitments that justify a higher contract value. Pricing research should expose those choices before a team debates whether the number should be $79 or $99.
A direct price experiment begins at step four. Conjoint can inform steps two, three, and four together. That wider decision range is why it is the better buy for most SaaS pricing work.
Current AI SaaS pricing makes the problem concrete. Each offer below combines a price with a target buyer, package structure, and pricing metric. Changing only the headline price would leave most of the commercial design untouched.
Exhibit 2: Four public SaaS offers demonstrate four different pricing architectures
| Vendor and published pricing, accessed September 3, 2026 | Target buyer | Packaging structure | Pricing metric | What the offer requires research to answer |
|---|---|---|---|---|
| Cursor - Pro at $20 per month; Teams Standard at $40 per user per month; Teams Premium at $120 per user per month | Individual developers, collaborative software teams, and enterprises | Individual tiers, team tiers, and custom enterprise controls | Per user, with included usage pools and on-demand model usage | Which developers value premium agent capacity, and when does usage need to supplement the seat? |
| GitHub Copilot - Business at $19 per granted seat per month; Enterprise at $39 per granted seat per month | Organizations and GitHub Enterprise Cloud customers | Business and Enterprise plans with larger AI-credit pools at the higher tier | Granted seats, included AI credits, then $0.01 per extra AI credit | Which controls and AI capacity justify the Enterprise step-up? |
| Devin - Pro at $20 per month; Max at $200 per month; Teams from an $80 monthly minimum; enterprise usage billed in ACUs set by order form | Individual technical users, power users, teams, and enterprises | Self-serve tiers plus contracted enterprise usage | Subscription plus included quota, on-demand credits, and enterprise Agent Compute Units | Where should a buyer move from predictable access to variable compute spend? |
| Intercom with Fin - Essential starts at $29 per seat per month plus $0.99 per Fin outcome; Advanced starts at $85 per seat per month plus outcomes | Support teams from startups through larger customer-service organizations | Helpdesk tiers combined with a separately metered AI agent | Seats for human support staff and outcomes for Fin | Which buyer trusts an outcome charge, and what counts as a billable resolution? |
The common lesson is not that every SaaS company should copy usage pricing. It is that a buyer is choosing an offer, not merely accepting or rejecting a number.
Choice-based conjoint research is especially valuable when a company must decide whether a feature belongs in a higher tier, an add-on, or the base product. Academic work has shown that conjoint can estimate reservation prices and simulate switching, cannibalization, and category expansion across alternative offers. Research on multipart pricing further shows how conjoint can model fixed fees, included use, variable charges, and optional features together.
A price-only test cannot answer a packaging question without accidentally turning the test into several overlapping tests. If a company changes its price, annual discount, included usage, and support level at once, it may lift conversion without knowing which change mattered. If it changes only price, it risks optimizing the wrong package.
Live tests establish causality only after the offer is held steady
Direct price testing earns its place when the company has a stable offer, meaningful self-serve traffic, short purchase cycles, and the ability to randomize prospects cleanly. A product-led SaaS company with 50,000 qualified monthly pricing-page visitors can learn a great deal from testing $49 versus $59 for an unchanged plan.
Enterprise SaaS usually lacks those conditions. Deals may involve security review, procurement, sales negotiation, legal terms, implementation services, and a buying cycle measured in months. A test that runs for two weeks may record more demo requests at a lower price but miss whether those buyers activate, expand, or renew. Short-term experiment effects can diverge from long-term effects as customers learn, compare, and adapt.
Direct price testing also fails when teams confuse exposure with evidence. A trustworthy test requires discipline:
- Randomize comparable prospects at the same time rather than compare this month’s price with last quarter’s price.
- Keep product claims, contract terms, trials, sales follow-up, and checkout flow constant across conditions.
- Measure paid conversion and revenue, not only clicks, demo requests, or free-trial starts.
- Follow accounts long enough to observe activation, refund behavior, downgrades, and early retention.
- Stop the test if visible price differences create unacceptable customer fairness, channel, or contractual risks.
Those requirements narrow the set of SaaS companies that can use direct testing as their first research method. They do not narrow the value of conjoint for offer design.
AI pricing raises the value of conjoint because the central choice often concerns the billing metric. A seat is familiar when a person remains the main actor. An outcome becomes more credible when the AI completes work with little human intervention. Usage may be necessary when compute cost rises sharply with demand.
The Agentic Monetization Spectrum, or AMS, gives leaders a disciplined way to make that call. It scores an agent on three dimensions: zero-human ability, meaning how much work the agent completes without a person; operational domain, meaning whether it handles a task, a workflow, or work across functions; and output/cost ratio, meaning whether value rises roughly with cost or far faster than cost. More autonomous agents with a broader domain move the pricing anchor away from seats and toward output or outcomes.
For the following exhibit, 1 means small, 2 means medium, and 3 means large. The scores are Monetizely’s judgment of the public offers, not claims made by the vendors.
Exhibit 3: The AMS points to different research priorities across AI SaaS offers
| AI SaaS offer | Zero-human ability | Operational domain | Output/cost ratio | Total | Primary pricing implication |
|---|---|---|---|---|---|
| GitHub Copilot Business | 1 | 1 | 2 | 4 | The developer remains the anchor, so seat pricing with usage limits is credible |
| Cursor Teams | 2 | 2 | 2 | 6 | Team seats remain useful, but agent capacity and model usage must be priced deliberately |
| Devin | 2 | 2 | 2 | 6 | A subscription can support adoption, while variable compute charges protect economics |
| Intercom Fin | 3 | 2 | 2 | 7 | An outcome metric fits better because the product is expected to resolve work, not merely assist an employee |
The scores strengthen the case for conjoint as the primary method. A company considering an outcome charge needs to test the buyer’s willingness to accept the definition of an outcome, the value of automation, the expected volume, and the fallback path when a human intervenes. A/B testing $0.99 versus $1.19 per outcome cannot decide whether an outcome is the best unit in the first place.
Intercom’s published Fin pricing makes the point sharply. Fin charges $0.99 for a resolution or defined procedure handoff, while its broader platform also charges per seat. The company has had to define when an outcome is counted and when a failed interaction is not billable. That is a pricing-system decision before it becomes a rate-testing decision.
The choice should be based on the decision at hand, not on a generic preference for surveys or experiments. Monetizely scores each method from 1 to 5 against the common work of a SaaS pricing team.
Exhibit 4: Conjoint is the stronger first method for pricing and packaging decisions
| Evaluation criterion | Conjoint analysis | Direct price testing |
|---|---|---|
| Diagnose whether packages fit distinct buyer segments | 5 | 1 |
| Evaluate feature, service, term, and price trade-offs together | 5 | 1 |
| Compare seat, usage, and outcome pricing metrics | 5 | 1 |
| Estimate demand across several candidate price points | 4 | 3 |
| Measure actual paid response to one defined offer | 2 | 5 |
| Work when enterprise deal volume is low and sales cycles are long | 5 | 1 |
| Protect learning when a public price test would create fairness concerns | 5 | 2 |
| Total | 31 | 14 |
Direct testing wins decisively on one criterion: it observes real payment behavior. Yet pricing optimization is wider than a rate check. A SaaS company that has not settled its segmentation, package structure, and meter should not use live buyers to discover basic commercial design.
A 2021 study in the International Journal of Research in Marketing found that direct willingness-to-pay questions can suffer from hypothetical bias, but also showed that carefully designed de-biasing procedures can improve their usefulness. The implication is not that surveys are weak and experiments are strong. It is that every stated-preference method needs a realistic buying context, credible price levels, and a no-purchase option.
Buyer role and decision scope should determine the first method purchased. The table below applies the thesis without treating direct testing as a rival strategy.
Exhibit 5: Buyer profiles that should lead with conjoint or direct testing
| Buyer profile | Immediate decision | Recommended first method | Why |
|---|---|---|---|
| Chief product officer redesigning three SaaS tiers | Which features, services, and limits belong in each tier | Conjoint analysis | The team needs trade-offs across the full offer |
| Pricing leader moving from seats to AI usage or outcomes | What to bill for and how to package the transition | Conjoint analysis | The pricing metric must earn buyer acceptance before the rate is tested |
| Enterprise SaaS executive facing long, negotiated sales cycles | Whether the offer fits different industries and account sizes | Conjoint analysis | Live testing will be too slow, too sparse, and too exposed to sales-process variation |
| Product-led growth leader with high traffic and an unchanged plan | Whether $49 or $59 improves paid conversion and early revenue | Direct price testing | The company has already made the package and metric decision |
| Finance leader validating a finalized self-serve price change | Whether a selected list price causes immediate conversion loss | Direct price testing | The test can confirm the final rate before broad rollout |
Conjoint is therefore the better buy for the senior SaaS operator responsible for pricing optimization. Direct price testing is the better tool for the narrower job of validating a stable, self-serve price after the offer has been designed.
The practical error is to treat research as a choice between a survey and an experiment. The stronger architecture assigns each method a distinct job. Conjoint develops a small number of offers that match buyer segments and company economics. Direct testing then validates the best candidate where traffic, purchasing flow, and fairness conditions support it.
That sequence also makes executive decisions easier. Rather than asking a growth team to “test pricing,” the leadership team can ask a precise series of questions: Which segment are we trying to win? Which package serves it? Which meter does it understand? Which rate should we validate? Each question has an owner and a form of evidence.
The operator’s next moves should be concrete:
Write the pricing decision in one sentence before commissioning research. State whether the decision concerns segments, packages, metrics, rates, or billing operations. Do not fund a price test to answer a packaging question.
Treat the pricing metric as a board-level choice for AI products. Require product, finance, sales, and customer success leaders to agree on what the buyer will be billed for before debating price levels.
Build a conjoint study around real alternatives, including a no-purchase option. Use current competitor prices, credible feature descriptions, annual terms, and realistic usage levels so respondents face choices that resemble an actual buying decision.
Reserve direct testing for a limited number of final rate hypotheses. Run simultaneous randomized tests only after the offer, meter, and core message are fixed.
Create a post-launch evidence loop that connects research to billing data. Compare predicted segment uptake with actual conversion, usage, discounts, expansion, and retention, then update the next pricing decision with that evidence.
Footnotes
- https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
- https://www.sciencedirect.com/science/article/pii/S0167811620300422
- https://pubsonline.informs.org/doi/10.1287/mnsc.48.10.1350.272
- https://pubsonline.informs.org/doi/10.1287/mnsc.2017.2753
- https://prod.cursor.com/docs/models-and-pricing
- https://docs.github.com/en/copilot/concepts/billing/organizations-and-enterprises
- https://docs.devin.ai/admin/billing
- https://www.intercom.com/pricing
- https://www.intercom.com/help/en/articles/8205718-fin-ai-agent-outcomes