The period that closes

HubSpot charges $0.50 when its Customer Agent resolves a conversation, but it can't know a conversation is resolved until 72 hours after the customer last wrote, and then the answer is locked. A billable event that is decided after the fact has to land in some period, and the period it lands in has to be closed at some point.

On April 14, 2026 HubSpot changed what its Customer Agent charges for. The announcement, headed "now you pay when the task is complete", moved it "from $1.00 per conversation to $0.50 per resolved conversation," paid "via HubSpot Credits at a cost of 50 credits per resolution." A conversation is an event the system sees start; a resolution is a judgement about how it ended, and the interesting part of HubSpot's documentation is how long it waits before making that judgement and what it does once it has.

The previous post was about usage that reaches the billing system late because someone upstream measured it first. HubSpot's case is different in kind, because when the period ends the event may not exist yet. That moves the problem from the pipeline to the calendar, and specifically to the moment a billing period stops accepting anything new.

Seventy-two hours, then locked

HubSpot's help article on the agent defines a resolution as a conversation that "includes at least one reply from the customer agent that shares a content source" or "performed an action (e.g., a password reset)," with "no qualifying handoff to a human agent within 72 hours of the last visitor response." Only handoffs the visitor starts count; "manual agent assignment, including assignment through workflows, is not considered a qualifying handoff and does not, by itself, prevent the conversation from being counted as resolved."

The timing follows from the definition. "The resolution status is evaluated and set 72 hours after the last visitor response," and after that, "subsequent actions, such as messages from a live agent, a transfer request, or negative feedback, will not change the resolution status." HubSpot is candid about the side effect on its own dashboards: "resolution rate data for the current week will show a lag due to this 72-hour evaluation window."

Two details stop the window being a fixed three days. For email, "if a customer replies after more than 72 hours in an email thread, the conversation is reopened and the 72-hour evaluation window resets," and "if the customer agent delivers another resolution, additional credits are consumed." So one thread can produce several billable events spread over weeks. Lead qualification runs on the opposite rule: the status "is updated immediately once the qualification is set," so the same product has one kind of resolution decided at once and another decided three days or more after the fact.

Those resolutions draw on a pool with hard edges. HubSpot's credits and billing article says credits "reset every month, aligned with the start date of your usage period," that "unused credits expire at the end of each usage period and do not roll over to the next month," and that "overages are billed on a monthly basis, in arrears, in increments of 10 HubSpot Credits, rounding down." A conversation whose last visitor message lands in the final three days of a usage period is resolved, if it is resolved at all, in the next one. The help articles I read don't say which period's credits it draws from, and either answer is defensible: the month the work happened in, or the month the judgement was made in.

Modelling it, on paper

As in the earlier posts, this maps HubSpot's published rules onto Revenue Cloud Usage Management using the public usage-consumption skill in the SalesforceLabs repository. It's a thought experiment about the problem, not a claim about how HubSpot bills, and the platform behaviours below are the repo's account of them, not something I've run.

The skill's rules about the close are blunt. Usage has to be booked "into a PAST period," because "drawdown and final rating settle only when a period completes; the current period stays open indefinitely." Once a period has been orchestrated, a UsageSummary that has reached a complete status "never reopens, and the journals stay Pending forever with no error." Clearing the usage data doesn't help either: the clear "drains the summaries but the period state survives, so re-recording into the same period strands the journals exactly as above," and re-testing one period "needs a full account reset and an asset rebuild." The skill's orchestration script has a dedicated exit code for the result, "stalled with journals pending — usually stranded behind a period that closed before the usage was recorded."

Put a HubSpot-style resolution through that and the first decision is which date the TransactionJournal row carries. Dated to the conversation, an event from the 30th belongs to a period that has to stay open for at least three days after it ends, and the reopened email thread makes "at least" open-ended: a resolution delivered in week three of a reopened thread would be dated into a period that may already have closed, and by the skill's account it would sit in Pending without an error. Dated to the evaluation, the event always lands in an open period, and the close can run on the calendar, but the period then records when judgements were made, not when the work was done.

The second decision is when to close, and it belongs to operations more than to configuration. With conversation dating, the close can't run at month end; it has to wait out the longest evaluation window the product has, and then accept that anything decided later is either stranded or deliberately re-dated. That makes the adjudication window a parameter of the finance calendar, which is an odd place for a support product's rule to end up, and it means whoever changes the window in the product is also changing when the books can close.

The third is the lock. HubSpot's rule that later evidence "will not change the resolution status" is a commercial version of the platform's close: both refuse new information after a point, so that an amount, once computed, stays computed. The difference is in who hears about it. HubSpot's lock is written down for customers in its help centre, while the stranded journal, by the skill's own account, produces no error, which is why it opens by warning that "almost every mistake is silent."

Where the money becomes final

The first post said a usage business has to decide what it is counting; this one is about deciding when it stops counting for a given month. A meter can run forever, but revenue recognition and invoicing need a line after which the number for a period is fixed, and an outcome priced on a judgement moves that line to wherever the judgement is made.

Once the period is closed, the next question is what the closed total is drawn down against. Datadog publishes its rule for that in one line, "allotments + committed usage = included usage," and adds that "allotments do not carry over between billing periods." If a commitment, an allowance and an overage rate all apply to the same closed month, which one does a unit of usage use up first?

No comments yet