Improve Pricing with Conversion and Retention Data

Task

Optimize pricing and packaging using conversion and retention data.

Summary

Use cohort behavior and observed economics to refine packaging, price, and expansion paths.

Use Conversion and Retention Data to Improve Pricing and Packaging

Task ID: S5-14

Executive summary: A pricing change is successful only when it attracts suitable customers, preserves long-term revenue, and improves the economics of serving them. This article explains how to compare customer cohorts, connect conversion with retention, interpret average revenue per account and net revenue retention correctly, test package changes, and set an improvement target that reflects the company’s own market and data.

A company launches a cheaper package and new-customer conversion rises. Sales celebrates. Three months later, support demand is up, upgrades are rare, and many of those customers are leaving.

Another company raises prices. Average revenue per account rises immediately, but the smallest customers begin cancelling, partner-sourced deals slow down, and the customers who remain receive more discounts at renewal.

Both companies can claim that a headline pricing metric improved. Neither yet knows whether the new package created a better business.

The work is therefore not simply to “raise prices” or “improve conversion.” It is to determine which combination of price, included features, usage limits, contract terms, and buying process produces customers who convert, receive value, renew, expand, and remain economical to serve.

The operating principle

Evaluate pricing and packaging as a customer-cohort problem, not as a one-time price-setting exercise.

A cohort is a group of customers that shares a meaningful starting condition. The simplest example is all customers who started in the same month. For pricing work, more useful cohorts often combine several conditions:

• acquisition date;
• package purchased;
• acquisition channel;
• customer segment;
• initial discount;
• contract length;
• geography;
• first-value behaviour;
• pricing version in effect at purchase.

The company then follows each cohort from first exposure to the offer through purchase, activation, renewal, contraction, expansion, and cancellation.

This matters because conversion and retention answer different questions. Conversion shows whether customers will begin paying under a package. Retention shows whether the company continues delivering enough value to justify the price. Expansion shows whether customers can grow naturally within the package structure.

A higher conversion rate can be harmful when it brings in customers with poor product fit, high support requirements, low willingness to pay, or little reason to renew. A lower conversion rate can be acceptable when a package screens out unsuitable buyers and improves gross margin, retention, or sales efficiency. Research on free trials, for example, finds that customers acquired through a trial can behave differently from regular customers and that their subsequent usage affects retention differently; trial conversion should therefore not be treated as equivalent to durable customer acquisition.

The same principle applies to product-led growth, partner sales, marketplaces, referrals, integrations, outbound sales, and enterprise deals. Each channel can produce a different mix of customer fit, discounts, implementation effort, expansion potential, and revenue ownership. A package that works in self-serve acquisition may create unattractive marketplace fees or partner margins. An enterprise package may support high average revenue but require implementation and support costs that are invisible in subscription revenue.

This task appears after the company has established a product and a reasonably repeatable subscription operation because pricing analysis depends on observable behaviour. The company needs enough transactions, renewals, usage records, and customer outcomes to distinguish a promising package from a temporary sales effect. Adding more channels before this is understood can multiply the wrong customers, discounts, and service obligations.

“No dependencies specified” should not be read as “no prerequisites.” At minimum, the review requires stable customer identifiers, billing history, plan history, channel attribution, discount records, cancellation dates, and a consistent definition of recurring revenue. Product usage, onboarding, support, and cost-to-serve data substantially improve the analysis.

What the pricing review must reveal

A credible review should explain four connected results:

1. Who converts under each offer?
2. Which customers remain and expand?
3. What does each package cost to sell, onboard, support, and deliver?
4. Which package changes are likely to improve the whole system rather than one isolated metric?

The analysis should cover both price and package design. Price is the amount charged. Packaging determines what customers receive, which limits apply, how usage is measured, what requires an upgrade, and how plans differ.

A package review may therefore examine:

The review should not assume that a three-tier “good, better, best” layout is automatically appropriate. Customers may prefer a flat subscription, per-seat price, usage-based charge, bundled offer, or a hybrid structure. Research on multipart pricing explicitly treats fixed access charges, included usage, variable charges, and add-ons as interdependent choices: customers’ package choices depend partly on uncertainty about how much they will use.

Customer behaviour can also be influenced by predictability and perceived risk, not merely by the mathematically cheapest tariff. Empirical research has found persistent flat-rate and pay-per-use biases; in the study, pay-per-use bias was associated with greater churn, while flat-rate bias was not. The practical lesson is not to exploit misjudgment. It is that a package must be judged by actual selection, usage, satisfaction, and retention rather than by assuming every customer calculates the lowest possible bill perfectly.

The cohort view

Begin with acquisition cohorts, but do not stop there. At minimum, compare cohorts by:

• pricing version;
• package;
• customer segment;
• acquisition channel;
• monthly or annual commitment;
• discounted or full-price purchase;
• self-serve or sales-assisted purchase;
• time to first useful result;
• implementation complexity.

For every cohort, track a common sequence:

flowchart LR
    A[Offer viewed] --> B[Trial, demo, or checkout]
    B --> C[Paid account]
    C --> D[First useful result]
    D --> E[Continued use]
    E --> F[Renewal]
    F --> G[Expansion, contraction, or churn]
    G --> H{Keep, change, or retire package}

Text description: The analysis follows customers from first exposure to the offer through purchase, first value, continued use, renewal, and subsequent revenue movement. Package decisions are made only after the company can see where each cohort succeeds or fails.

This sequence prevents a common mistake: optimizing the checkout while ignoring what happens after purchase.

For example, suppose Package A converts 8% of qualified visitors and retains 65% of its first-year recurring revenue. Package B converts 6% but retains 92%, needs less support, and expands more often. Package A may still be useful as an entry offer, but only if its role in the customer journey and its total economics justify it.

A practical method for the review

The review should be owned jointly by product, revenue operations, finance, sales, and customer success. No single function sees the whole result. Product sees usage; finance sees revenue and margin; sales sees objections and discounts; customer success sees adoption and renewal; revenue operations connects customer, channel, and billing records.

A focused review can usually proceed through the following sequence.

Establish definitions before analysing results

Create a metric dictionary that answers:

• What counts as an account?
• Are parent and subsidiary accounts combined?
• When does a trial become paid?
• How are refunds and credits handled?
• Does recurring revenue include usage charges?
• How are currency movements treated?
• How are customers who pause and reactivate classified?
• Are marketplace and partner fees recorded as revenue reductions or selling costs?
• Does expansion include price increases, added seats, added usage, and cross-sold products?
• How are migrated customers assigned to pricing cohorts?

This step is not administrative housekeeping. Public companies themselves calculate similarly named retention measures differently. Cloudflare, for example, compares annualized revenue from the same paying-customer set across four quarters, excludes new customers, and also excludes free customers that converted to paid during the interval. Snowflake warns that its key metrics may differ from similarly named measures used by other companies.

A company that changes definitions midway through the review can manufacture an improvement without improving the business.

Build an account-level cohort table

Use one row per customer account per measurement period. Include:

• account and parent-account identifiers;
• cohort start date;
• channel;
• segment;
• package and pricing version;
• list price and actual price;
• discount and promotion;
• contract term;
• starting and ending recurring revenue;
• expansion, contraction, and churn;
• active seats or usage;
• onboarding completion;
• first-value date;
• support effort;
• direct delivery cost;
• renewal status.

Do not calculate retention from aggregate company revenue. First compare each account’s revenue across periods, then total the results. This prevents new customers and expansion from hiding losses elsewhere in the base. SaaS Capital similarly recommends following a consistent customer cohort and comparing account-level revenue before aggregation.

Compare mature cohorts fairly

A January cohort has had more time to renew than a July cohort. A self-serve monthly customer may reveal churn within weeks, while an annual enterprise account may not produce a renewal decision for a year.

Use fixed maturity windows such as:

• conversion within 30 days of first qualified exposure;
• activation within 14 or 30 days of purchase;
• 90-day customer and revenue retention;
• six-month expansion and contraction;
• first annual renewal;
• twelve-month net revenue retention.

Do not compare a mature cohort with an immature one as though both have faced the same opportunities to churn or expand.

When the customer count is small, report the number of accounts and the revenue concentration beside every percentage. One lost enterprise account can dominate a cohort. Several tiny accounts can dominate logo churn while having little revenue effect.

Diagnose before prescribing

For every weak cohort, identify where the problem begins.

A conversion problem may come from unclear value, excessive price, unsuitable qualification, poor sales execution, or a difficult buying process.

A retention problem may come from poor fit, weak onboarding, missing features, excessive complexity, unreliable delivery, unexpected charges, or a package that allows purchase before customers can realistically succeed.

An expansion problem may mean that upgrade triggers are unclear, additional value is weak, limits occur too early, limits occur too late, or the company has not created a useful next package.

Pricing should not be used to disguise a product or onboarding problem. If customers who reach first value retain well but few customers reach it, package design may need to include better onboarding rather than a lower price. Product analytics tools commonly distinguish acquisition cohorts from behavioural cohorts for this reason: customers who complete important actions may retain differently from those who merely signed up.

Form a package hypothesis

Each proposed change should state:

• the customer segment affected;
• the observed problem;
• the proposed price or package change;
• the behaviour expected to change;
• the main success measure;
• the retention and margin guardrails;
• the period required to evaluate the result.

For example:

Mid-market customers who use advanced reporting retain well, but many prospects choose the lower tier and later discover that reporting is unavailable. Moving a limited reporting feature into the lower tier is expected to improve qualified conversion and first-value completion. The change will be accepted only if six-month revenue retention, support cost, and upgrades to the full reporting tier remain within agreed guardrails.

This is more testable than “make the lower tier more attractive.”

Test without contaminating the comparison

A randomized controlled test provides the strongest causal evidence when customers can be assigned fairly to alternative offers. Controlled experiments use randomization to separate the effect of a change from other differences between customers. Microsoft’s experimentation research also warns that bad instrumentation, incorrect metric interpretation, or an unexpected imbalance between test groups can produce confident but wrong conclusions.

Pricing experiments require particular care:

• Do not show materially different prices to customers who are likely to compare notes unless the test has a defensible basis.
• Keep taxes, currencies, channel fees, and contract terms consistent or model them separately.
• Avoid changing the package, onboarding, sales script, and target customer simultaneously.
• Predetermine the primary metric and guardrails.
• Run the test long enough to observe the relevant retention behaviour.
• Preserve the assigned offer in the data even if a salesperson overrides it.
• Record overrides, discounts, and exceptions.

Where randomization is impractical, use a staged rollout, geographic or channel holdout, matched cohorts, or interrupted time-series analysis. Treat the result as weaker evidence and document competing explanations.

A six-week operational cycle might look like this:

Six weeks is an operating example, not a universal deadline. Companies with annual contracts may need twelve months or more to validate renewal effects. A high-volume monthly product can learn more quickly, although early results still may not predict long-term retention.

Measure ARPA and NRR without fooling yourself

The primary measures for this task are average revenue per account (ARPA) and net revenue retention (NRR). Both are useful. Neither is sufficient alone.

Average revenue per account

A practical periodic formula is:

Use accounts rather than users when the paying customer is an organisation. If individual users are the economic customer, average revenue per user may be appropriate instead. State the denominator explicitly.

Calculate ARPA by:

• package;
• segment;
• channel;
• pricing version;
• contract term;
• acquisition cohort;
• geography where prices materially differ.

A rising overall ARPA can have several meanings:

• customers upgraded;
• the company raised prices;
• usage increased;
• more high-value customers were acquired;
• low-revenue customers churned;
• several small accounts were consolidated;
• the customer mix shifted toward enterprise accounts.

Only some of these represent better packaging. If low-value customers leave, ARPA may rise while total recurring revenue, customer count, and future expansion decline. ARPA should therefore be read beside conversion, customer retention, gross revenue retention, expansion, support cost, and contribution margin.

Net revenue retention

A common formula is:

New-customer revenue is excluded because NRR measures what happened to the starting customer base. An NRR above 100% means expansion and price increases exceeded contraction and churn for that cohort.

NRR combines several mechanisms:

• successful adoption and expansion;
• added seats;
• increased usage;
• cross-sold products;
• contractual price increases;
• reduced discounts;
• downgrades;
• lost usage;
• partial cancellations;
• full churn.

Because price increases count as expansion in many definitions, NRR can improve before the company knows whether retention has been harmed. Pair it with gross revenue retention (GRR), which excludes expansion:

Also inspect customer retention, product usage, support demand, renewal discounts, and cancellation reasons.

Set a contextual improvement target

“Improvement target set” should produce a specific commitment such as:

For new mid-market customers entering the revised package between October and December, increase twelve-month NRR from the current cohort baseline of 94% to at least 98%, while keeping qualified-opportunity conversion within one percentage point of baseline, GRR above 88%, and onboarding contribution margin non-negative.

The target needs:

• a named cohort;
• a baseline period;
• a measurement window;
• a target value or range;
• an account and revenue sample-size requirement;
• conversion, retention, and margin guardrails;
• an owner;
• a review date;
• a rule for keeping, changing, or reversing the package.

Do not copy a public-company NRR target. Benchmarks differ materially by average account value, customer type, contract term, business maturity, and pricing model. ChartMogul’s 2024 data, for example, found that only the top quartile of higher-ARPA businesses consistently reached at least 100% NRR, while monthly and annual plans also showed different retention patterns. Its methodology excluded smaller businesses from some ARPA comparisons because they could distort the results.

A separate 2026 survey reported 103% median NRR for bootstrapped business-to-business software companies with between US$3 million and US$20 million in annual recurring revenue. That figure describes a particular survey population, not a minimum standard for every subscription company.

Even benchmarking platforms apply eligibility rules. Stripe’s subscription benchmarks, for example, require businesses used in peer groups to have at least 100 active subscriptions, a minimum operating history, and positive annual recurring revenue.

External benchmarks should help test whether a target is plausible. The actual target should come from the company’s baseline, economics, customer promise, and strategic intent.

What public examples teach

HubSpot: broader entry can reduce average revenue before later expansion appears

HubSpot announced a new seat-based model in January 2024 and implemented it for new customers in March. The change removed seat minimums for certain products, introduced paid core seats, and allowed unlimited free view-only seats in paid accounts. The stated design was to make it easier for customers to begin and then pay as their organisations grew.

The company’s 2024 filing reported that customer count rose from 205,091 at the end of 2023 to 247,939 at the end of 2024. Average subscription revenue per customer declined from US$11,384 to US$11,343, with the company attributing the decrease partly to lower-priced Starter purchases and the new seat model.

By the first quarter of 2026, customer count had reached 299,458 and average subscription revenue per customer had increased from US$11,038 in the comparable 2025 quarter to US$11,722. The company attributed that increase principally to demand for Professional and Enterprise products and currency effects, partially offset by continued purchases of lower-priced Starter products.

These public figures do not prove that the seat change caused the later increase. Customer mix, product demand, acquisitions, exchange rates, and other changes were also involved. That limitation is the lesson.

A package intended to lower the entry barrier may depress aggregate ARPA while new cohorts enter. The company must follow those cohorts long enough to learn whether they adopt more products and seats, remain customers, and expand. Looking only at the first year’s overall ARPA could lead management either to reject a useful growth structure too early or to accept it without evidence that customers expand.

Netflix: a lower-price package can broaden conversion, but adoption is not the whole result

Netflix introduced an advertising-supported package as a lower-price alternative to its advertising-free subscriptions. By November 2024, the company reported 70 million monthly active users globally on the advertising-supported option and said that more than half of new sign-ups in countries where it offered advertising selected that package.

The example shows how packaging can create a different value exchange rather than simply discounting the existing product. Customers receive a lower subscription price; the company gains an additional advertising revenue stream.

The public adoption numbers establish that the package attracted substantial demand. They do not, by themselves, establish its retention, customer lifetime value, advertising margin, or effect on higher-priced plans. A company applying the lesson would compare advertising-plan cohorts with comparable advertising-free cohorts on engagement, cancellation, upgrade and downgrade movement, total revenue, and cost to serve.

The useful principle is to measure the complete economics of the package. High sign-up share is evidence of appeal, not final proof of value.

Failure modes, tradeoffs, and open questions

Optimizing conversion alone. A pricing page can convert more visitors by lowering prices or weakening qualification. Unless those customers activate, renew, and cover their support and delivery costs, the apparent improvement merely moves the problem downstream.

Optimizing NRR alone. NRR can rise because of price increases or expansion among a small group while many customers leave. Review GRR, logo retention, concentration, usage, and customer outcomes beside it.

Using only company-wide averages. Aggregates combine old and new pricing, monthly and annual accounts, small and large customers, and multiple channels. The result may describe none of them accurately.

Comparing cohorts at different ages. A three-month-old cohort has not had the same opportunity to renew or churn as a twelve-month-old cohort.

Ignoring package migration. Existing customers may be grandfathered, migrated immediately, moved at renewal, or given temporary credits. Each group needs a separate treatment flag. Otherwise, the company cannot distinguish package performance from migration policy.

Treating annual commitment as proof of retention. Annual plans can reduce short-term visible churn because customers cannot cancel monthly. The more important question is whether they renew when the commitment ends. ChartMogul’s billing research found stronger retention among annual plans in its 2024 dataset, but that relationship does not establish that contract length caused customers to receive more value.

Confusing correlation with a package effect. Retained customers may use a feature more often because they are already better-fit customers. Moving that feature into another tier may not create retention. Use experiments or staged tests when possible.

Changing too many variables at once. A simultaneous change to price, feature limits, sales compensation, onboarding, discounts, and target market may improve results, but the company will not know why.

Hiding discounts. List-price cohorts are meaningless when sellers, partners, and marketplaces routinely apply different discounts. Analyse realised price and renewal concessions.

Ignoring cost to serve. A package can improve ARPA and NRR while reducing profit if it adds onboarding, cloud-computing, implementation, partner, support, or compliance costs.

Creating arbitrary upgrade barriers. Customers should upgrade because the next package creates more value, not because the current package prevents them from completing the basic job they purchased it to do.

Using a free plan without defining its role. Free access may support discovery, product learning, referrals, or eventual conversion. It may also attract customers who never reach value or create substantial support and infrastructure cost. Randomized field research on free-to-paid app strategies found that paywall and product-design decisions interacted; seemingly customer-friendly components did not each improve subscription uptake when evaluated alone.

Overstating certainty with small samples. A cohort containing ten customers cannot support the same confidence as one containing ten thousand. Report account count, revenue concentration, variability, and confidence intervals. Where results are uncertain, make a reversible decision.

Important open questions should remain visible in the final review:

• Are customers leaving because of price, poor fit, missing value, or weak delivery?
• Does the package create an understandable path from entry to expansion?
• Are usage limits connected to customer value and company cost?
• Does a lower-priced package widen the market or cannibalize customers who would have paid more?
• Are partner and marketplace customers economically attractive after fees, commissions, support, and ownership disputes?
• How will existing customers be treated?
• How long must the company wait before renewal evidence is mature?
• What result would cause the company to reverse the change?

These are judgment questions. Data narrows the range of reasonable choices; it does not remove responsibility for the decision.

The decision and evidence at completion

The task is complete when the company can make a documented package decision using customer-level evidence—not when it has produced a new pricing page.

The required evidence should include:

Before the company depends on the revised package to scale partner, marketplace, product-led, expansion, or enterprise channels, the following should be true:

• customer and revenue records reconcile with billing;
• packages can be identified historically;
• conversion and retention can be segmented by channel;
• the company knows which customers each package is meant to serve;
• customers can reach a useful result within the package;
• support and delivery obligations are understood;
• upgrade and downgrade paths are operational;
• sales and partners know which offers they own;
• billing systems can implement the price and limits consistently;
• the improvement target has a baseline and measurement date;
• management has agreed on what would trigger continuation, revision, or reversal.

The key takeaway is straightforward: the best package is not the one with the highest price or the highest initial conversion. It is the one that brings in suitable customers, helps them reach value, retains and expands revenue, and produces workable economics across the channels the company intends to scale.

Sources

Primary sources. HubSpot, “Upcoming Changes to HubSpot’s Pricing,” January 2024. HubSpot annual and quarterly filings for 2024–2026. Netflix, “Netflix Celebrates Two Years of Advertising,” November 2024. Cloudflare and Snowflake regulatory filings describing net-retention calculations and limitations. Stripe Billing benchmarking documentation.

Open research. Datta, Foubert, and Van Heerde, “The Challenge of Retaining Customers Acquired with Free Trials,” Journal of Marketing Research. Cao, Chintagunta, and Li, “From Free to Paid: Monetizing a Non-Advertising-Based App,” Journal of Marketing. Iyengar, Jedidi, and Kohli, “A Conjoint Approach to Multipart Pricing,” Journal of Marketing Research. Lambrecht and Skiera, “Paying Too Much and Being Happy About It,” Journal of Marketing Research. Microsoft Research publications on controlled experiments, metric interpretation, and sample-ratio mismatch.

Public benchmark research. ChartMogul, retention and billing analyses based on anonymized subscription data. SaaS Capital, private business-to-business subscription retention definitions and 2026 benchmarks.