A PLG signal model is not a dashboard chart and it is not a product-usage report. It is a predictive layer that watches what workspaces, users, and seats do inside the product, scores every account on buyer intent, and decides which accounts get an AE touch, which get a nurture track, and which get left alone to keep growing on their own. Teams that build this well raise sales-assisted conversion without raising headcount, because they stop guessing which workspace is a company about to buy and which is a side project that will never pay. This guide walks the full arc of building the model: inventorying the in-product signals, scoring the account rather than the user, weighting the signals against paid outcomes, computing a composite score, routing the hot accounts to a named AE, and iterating monthly.
Time: 2-3 weeks setup, ongoing
Start by writing down every in-product event that paid customers tend to do and that non-converting workspaces tend to skip. The usual list for a B2B SaaS product is short and durable: feature adoption on the paid-tier features, workspace creation, seat invites sent and accepted, data imported from a source system, integration connected, core-workflow runs, and any admin-settings action that only a buyer would perform. Write each one as a named event in the telemetry tool, with the workspace identifier and the user identifier both attached, so the signal can be rolled up two ways. OpenView PLG and Reforge research both land on the same discipline here: the signal inventory is the artifact the model depends on, and teams that write it down once and reuse it ship a durable program, while teams that improvise the list every quarter watch the model drift because the input layer keeps changing.
The unit of a PLG signal model is the account, not the user. A single power user who invites nobody is a different buying signal from a workspace with four invited seats and two of them active daily, and a model that scores at the user layer will miss the second case entirely. Build an account-rollup view in the warehouse that joins every event to its workspace, counts active users per week, sums seats invited and seats accepted, and flags whether an admin-settings action has fired in the last fourteen days. The rollup is where PLG scoring stops being a usage report and starts being a sales signal, because buyers are companies and the signal has to live where companies live. Appcues and ProductLed research are both consistent on this point: account-level rollups outperform user-level scores on sales-assisted conversion by a wide margin, because the sales motion operates on accounts and the score has to speak the same language.
Before you combine signals into a composite, measure how predictive each one is on its own against the paid outcome. Pull the ninety-day cohort of workspaces, join each to its final paid outcome, and compute the conversion rate for workspaces that fired each signal versus workspaces that did not. The signals that lift conversion by a factor of three or more are the ones that deserve the most weight, the signals that lift by less than fifty percent are candidates to drop, and the signals that correlate with conversion only because they correlate with time in the product get a smaller weight so they do not dominate. Reforge research is clear that the first version of a PLG signal weighting should be interpretable, because the sales team has to trust it to act on it, and a weighting that cannot be explained to an AE will be routed around no matter how precise it looks on paper.
Combine the weighted signals into a single composite score, scaled zero to one hundred, with the four signal families contributing explicit shares. A durable split is roughly feature adoption at thirty, workspace growth at thirty, data depth at twenty, and buyer behavior at twenty, tuned against the validation set rather than against intuition. Keep the composite interpretable by exposing the four family subscores alongside the headline, so an AE opening the CRM record sees the composite, the four family scores, and the top three contributing signals in one view. ProductLed and OpenView benchmarks both show that interpretable composite scores outperform black-box scores on sales adoption, because the sales team actually looks at them, and a score that an AE will not open is a score the model never gets to deploy. The point is not model accuracy in a notebook, the point is operational trust in the CRM.
A model that scores once a week is a report. A model that scores every day and writes to the CRM is an operating layer. Run the scoring job every twenty-four hours against every active account, write the composite to a stand-alone numeric field on the account object, write the four family scores to their own fields, and attach the top signals as a short readable string. The frequency matters because PLG buying windows are short: a workspace that crosses the hot threshold on a Tuesday and does not get a touch until the following Monday has usually either already decided or already moved on. Appcues and Reforge research both call out the same discipline here: the lift from a daily scoring cadence over a weekly one is larger than the lift from most model-quality improvements, because latency compounds inside a short buying window and the sales touch has to arrive while the signal is still fresh.
Once the composite lives in the CRM, wire the routing. Define the hot threshold against the validation holdout, usually the top ten to fifteen percent of active accounts by composite, and route those accounts to a named AE based on account ownership or territory rather than to a round-robin queue. Agree on an SLA for first touch, typically under four business hours during business days for a hot account, and track SLA adherence weekly for the first ninety days. The point of the AE touch is not to pitch at the workspace, it is to show up with context, because the model has already told you which signals fired, which family is driving the composite, and which seats are active. OpenView and ProductLed are both clear that a product qualified account program dies when sales treats a PQA like a cold inbound, and it compounds when sales treats a PQA as a warm conversation the product has already opened.
The majority of active accounts will sit below the hot threshold, and that is correct. Build a product-led nurture track for the warm band that targets the specific signals the model is watching, so an account that is adopting features but has not invited seats gets a seat-invite nudge, and an account that has invited seats but not imported data gets a data-import nudge. Sequence the nurture against in-product state rather than against calendar days, so a workspace that imports data on day two does not receive the day-three import nudge. Keep the content short, action-oriented, and tied to the exact signal the account is missing. ProductLed and Appcues both show the same pattern here: nurture sequenced against user and workspace state converts materially better than nurture sequenced against the signup date, because state-based sequencing meets the account where it actually is rather than where the lifecycle tool thinks it should be.
Once the model has thirty days of production data, run a formal review every month for the first two quarters and quarterly thereafter. Lead with four numbers: conversion on the hot cohort versus the base rate, accepted rate on the hot cohort inside the AE team, signal lift from the nurture program, and composite calibration on the last thirty days of accounts. The calibration number is the one most teams miss and the one that most tells you when to retune: if accounts scored at seventy are converting at forty, the model is drifting and the weights are stale. Retune on a rolling ninety-day window every month, keep the signal set stable unless a review decision changes it, and version the deployed weights with a dated name so you can roll back cleanly. Reforge research is consistent that durable PLG signal programs are retuned on a cadence and that teams that leave the model static for a quarter watch its lift decay without seeing it until the sales team quietly stops acting on the score.
Start with four families: feature adoption on the paid-tier features, workspace growth from seat invites and accepted invites, data depth from imports and integrations, and buyer behavior from admin-settings actions. Keep the first version under twenty signals so every one is explainable in a sentence and the whole team can trust the composite.
Score at the account level. Buyers are companies, the sales motion operates on accounts, and a workspace with four active invited seats is a different signal from a single power user. User-level signals are inputs, but the composite that the AE acts on has to live on the account record.
Pull the ninety-day cohort of workspaces, join each to its paid outcome, and compute the lift of each signal: conversion rate for workspaces that fired the signal divided by conversion rate for workspaces that did not. Keep the top ten to fifteen signals by lift and tune the family weights on validation calibration.
Score every active account daily. PLG buying windows are short, and a weekly batch loses half the window to lag. The lift from a daily cadence over a weekly one is larger than the lift from most weight-tuning improvements, because the AE touch has to arrive while the signal is still fresh.
Set the hot threshold against the validation holdout and against the number of accounts the AE team can realistically touch at the agreed SLA, usually the top ten to fifteen percent of active accounts by composite. Tune the threshold on accepted rate rather than on volume, and tighten it before you widen it.
Retune monthly on a rolling ninety-day window, hold out the most recent two weeks for validation, and compare calibration against the previous deployed weights. Document the training window, the weight list, and the holdout calibration every time, so the next person to touch the model has a clean trail.
Strkr gives product, marketing, and sales a shared view of workspaces, seats, feature adoption, data imports, and daily PLG composite scores, so sales-ready accounts route themselves to the right AE and the warm band stays on a product-led nurture track that compounds.
No sales call, no migration consultant, no four-month implementation. Enter your card, get 14 days of the full Pro tier, cancel any time before day 14 with zero charge. Spin up a workspace, import your CSV, and have something useful before lunch.
