How to build a free trial conversion model

Before any instrumentation, write the five stages the model is going to track and the one-line definition for each. Signup is the trial signup event with the account identifier attached. Activation is the two or three in-product events that pattern-match to what paid customers do in their first session. Feature adoption is the deeper workflow events that confirm the user is using the product for the job it was bought for, not just the walkthrough. Trial-end is the time-boxed expiration event, with the conversion status flagged at the moment the window closes. Paid is the billing event, sourced from the billing system and not from a sales dashboard. OpenView PLG and ProductLed research both land on the same point: the teams that outperform on trial conversion are the ones that treat the funnel as a named sequence of events with clean definitions, because the model, the dashboards, and the lifecycle automations all learn against the same labels.

• Write a one-line definition for each of the five stages and store the definitions as code comments on the cohort view, so the next person to touch the model sees the same labels
• Pick two or three activation events, not ten, so the signal stays sharp and the lifecycle sequencing stays legible to the growth team
• Pick two or three feature-adoption events that are observably deeper than the activation set, so the model can tell a true power user from a trial tourist
• Source the paid stage from the billing system as a boolean on the trial record, never from a CRM sales stage, so the label the model learns against is clean

Once the stages are defined, instrument the events in the product telemetry tool with the trial account identifier attached, and build a cohort view in the warehouse that joins every trial from the last ninety days to its stage-by-stage outcome. One row per trial, with the signup timestamp, the acquisition source, firmographic fields from the enrichment provider, every stage event with its timestamp relative to signup, and the paid-conversion boolean. Ninety days is the minimum because shorter windows do not give the slow-converting trials time to resolve, and because the signal most teams miss is the gap between a seven-day converter and a twenty-eight-day converter. Appcues trial research and Reforge PLG research both call out the same discipline here: the cohort view is the artifact the whole program depends on, and teams that build it as a durable warehouse view ship a durable model, while teams that pull it ad hoc find themselves rebuilding it every month.

• Instrument the five stages with the trial account identifier on every event, so the join from telemetry to the CRM and the billing system is clean on the first pass
• Build a cohort view in the warehouse with one row per trial and columns for every stage event, firmographics, source, and the paid-conversion boolean
• Spot-check the join by sampling twenty converted trials and twenty expired trials against the raw billing and telemetry records, so the view is trusted before anyone reports on it
• Version the view with a dated name so the monthly reruns do not silently overwrite the training set that produced the deployed model

With the cohort view in place, compute the stage-to-stage conversion rate for the last ninety days and compare the end-to-end trial-to-paid rate to the public B2B SaaS benchmark of 10 to 25 percent. OpenView PLG benchmarks consistently land inside that band, with the higher end reserved for products that require a credit card at signup and the lower end for opt-in trials with no payment gate. The benchmark is a sanity check, not a target. The useful work is in the stage-level numbers: signup to activation, activation to feature adoption, feature adoption to trial-end, and trial-end to paid. A below-band global rate is almost always driven by a single bad stage, and a model that reports only the global number cannot tell you which. ProductLed research is explicit on this point: trial programs that move inside the 10 to 25 percent band do it by diagnosing and fixing the worst stage first, not by raising a global number through brute acquisition.

• Compute the stage-to-stage conversion for the last ninety days and lay it out as a five-stage bar chart, with the end-to-end rate at the bottom
• Overlay the 10 to 25 percent benchmark on the end-to-end rate, with a note that the higher end assumes a credit-card-up-front model
• Flag the worst stage in red, which is usually signup to activation or feature adoption to trial-end, so the team can agree on where to pull the first lever
• Share the chart with product, marketing, and sales in one review, so the three teams see the same numbers and agree on the diagnosis before any lever is pulled

A single global funnel hides the two segmentation layers that drive the biggest differences in conversion: acquisition source and firmographic fit. Compute the five-stage conversion for each major acquisition source, usually organic, paid, referral, and existing-list, and compute it again for high-fit and low-fit cohorts using the ICP score from the enrichment provider. The segmentation step is the one most teams skip and the one that pays off the most, because it tells you whether paid acquisition is converting below its channel baseline, whether a specific referral is doing the work the sales team thinks the product is doing, and whether the high-fit cohort is being left to self-serve when it should be routed to a named success manager. OpenView PLG benchmarks are consistent that trial-to-paid rates vary by source and by fit by multiples, not by percentage points, and that any program that pulls levers against the global number is tuning the channel mix as much as the user experience.

• Compute the five-stage conversion for each acquisition source with a minimum cohort size of fifty signups per source before any rate is reported
• Compute the five-stage conversion for the high-fit and low-fit cohorts using the ICP score from the enrichment provider, so the fit layer is visible alongside the source layer
• Flag any source whose end-to-end conversion is more than fifty percent below the organic baseline, because that source is a candidate for a volume cut or its own lifecycle treatment
• Flag the high-fit cohort separately as the pool a success manager will work from in the lever step, so the segmentation and the lever design land in the same artifact

The first lever against a weak signup-to-activation stage is a state-based onboarding email sequence. Pair every activation event with the specific email that drives at it, suppress the email the moment the event fires, and sequence against in-product state rather than against calendar days. A user who activates on day two should not receive the day-three activation nudge, and a user who has not activated by day five should receive a tighter nudge that references the specific event that is missing. ProductLed and Appcues both show the same pattern: nurture that is sequenced against user state converts materially better than nurture sequenced against the signup date, because state-based sequencing meets the user where the trial actually is rather than where the lifecycle tool thinks the trial should be. Keep each email short, action-oriented, and tied to one state transition, so the program is debuggable one message at a time.

• Write a message map that pairs each activation event with one email and one in-app message, so every stage transition has a single owning message rather than a chorus of nudges
• Sequence the sends against in-product state and suppress on event completion, so no user receives a nudge for an event they already fired
• Measure activation lift per message weekly, segmented by acquisition source and fit, so the message map can be tuned one row at a time
• Document every send with the triggering state, the suppression rule, and the measured lift, so the next person to tune the sequence has a clean trail

The second lever is an in-app prompt program tied to the feature-adoption stage. Pick the two or three deeper workflow events the funnel flagged as weak, write a short in-app prompt for each that surfaces the next action and the reason it matters, and gate the prompt on the user not having completed the event. The in-app surface is the right place for the feature-adoption lever, because the user is already in the product and the friction cost of acting on the prompt is a single click rather than a context switch. Appcues research is consistent that in-app prompts tied to specific workflow events outperform email on the feature-adoption stage by a wide margin, because the attention is already loaded on the task. Keep the prompt copy short, keep the surface non-blocking, and never stack two prompts on the same screen, because the second one erodes trust in both.

• Pick the two or three feature-adoption events the funnel flagged as weak, so the in-app program stays sharp rather than becoming a tour of every feature
• Write one short prompt per event, with a one-line value statement and a single primary action, and skip any secondary action that distracts from the next step
• Gate every prompt on the user not having completed the event, so repeat exposure is impossible and prompt fatigue stays low
• Measure feature-adoption lift per prompt weekly, segmented by acquisition source and fit, so the prompt set can be tuned one row at a time

The third lever is a named success manager outreach on the high-fit cohort, timed to the activation stage. Define the high-fit cohort against the ICP score and against the activation signal, route those trials to a named success manager rather than a round-robin queue, and agree on an SLA for first touch, typically inside one business day. The purpose of the touch is not to sell at the user, because the user is already inside the product. The purpose is to show up with context, because the funnel has already told the success manager which activation events fired, which feature-adoption events are missing, and which firmographic signals matter. OpenView PLG and Reforge both call out the same discipline: a product qualified lead program dies when the human touch treats a PQL like a cold inbound, and it compounds when the human touch treats a PQL as a warm conversation that the product has already opened. Keep the outreach short, specific, and tied to the next workflow the user is likely to need.

• Define the high-fit cohort against the ICP score and the activation signal, and size the cohort to what the success team can realistically touch at the agreed SLA
• Route the cohort to a named success manager based on territory or account ownership, so the first touch carries context rather than resetting the relationship
• Equip the success manager with a short talk track that references the activation events the funnel flagged and the specific feature-adoption events that are missing
• Measure accepted rate and converted rate on the high-fit cohort weekly, and tune the threshold on accepted rate rather than on volume, so the human lever stays trusted

The trial-end stage is the one most funnels under-serve, because the lifecycle tool is usually set up to send a generic expiration email on the last day. Replace that single send with a trial-end sequence that triggers at seventy-two hours before expiration, references the specific workflows the user actually ran, offers an extension path for a user who has activated but not yet decided, and offers a clean upgrade path for a user who has already hit the ceiling. The seventy-two-hour window is the moment the user is most likely to act, because the time-box is salient and the alternative is losing access to work already in flight. Appcues trial research and ProductLed benchmarks both call out the same lift: a trial-end sequence that references the user's own workflow state converts materially better than a date-based reminder, because the specificity reduces the perceived cost of the decision.

• Trigger the trial-end sequence at seventy-two hours before expiration, not on the expiration day, so the user has time to act before the clock runs out
• Reference the specific activation and feature-adoption events the user hit, so the sequence feels like a continuation of the trial rather than a generic reminder
• Offer an extension path for the activated-but-undecided user and an upgrade path for the ceiling-hit user, and route the hesitation signals to the success manager for the high-fit cohort
• Measure trial-end-to-paid conversion weekly, segmented by acquisition source and fit, so the sequence can be tuned one row at a time

Once the funnel and the three levers are in production, run a formal review every month for the first two quarters and quarterly thereafter. Lead with five numbers: stage-to-stage conversion, end-to-end trial-to-paid versus the 10 to 25 percent band, activation lift from the email program, feature-adoption lift from the in-app program, and accepted rate from the success manager cohort. Pick one lever per month to tune, change one thing at a time, and document every change with the triggering state, the expected lift, and the measured outcome. Reforge research is consistent that durable trial programs are tuned on a cadence and one layer at a time, and that teams that leave the funnel static for a quarter watch the lift decay without seeing it until the trial-to-paid number falls out of the band. Days-to-activation is the leading indicator that moves weeks before trial-to-paid does, so obsess over it as the monthly health signal and retrain the model when it moves.

• Report the five headline numbers to the weekly revenue review, so the trial conversion program stays funded on measured lift rather than on vibes
• Pick one lever per month to tune: the email map, the in-app prompt set, the success manager threshold, or the trial-end sequence, and change one thing at a time
• Retrain the stage model monthly on a rolling ninety-day window and compare calibration against the previous deployed version, so stale features get caught early
• Document every retrain and every lever change with the training window, the feature list, the production decision, and the measured outcome, so the next person to touch the funnel has a clean trail