Power Virtual Agents is the conversational bot platform that Microsoft has folded into Copilot Studio, and the rebrand carried a meaningful change in how it bills. The legacy model priced capacity in monthly session packs, where a session was a defined window of conversation regardless of how many turns it contained. The Copilot Studio model meters consumption in messages drawn from a capacity pool, where richer actions such as generative answers and connected automations consume more messages per interaction. The two models size very differently, and a bot that looked affordable under session pricing can consume far more under message metering once it uses generative responses at scale. Buyers who provisioned under the old model and never re estimated the message draw are the ones who get surprised. Power Virtual Agents is where a conversational pilot meets a metering change that rewards no one who stopped paying attention.
Power Virtual Agents now licenses as Copilot Studio on a message based capacity model, with a legacy session model still in play for some estates and a seeded entitlement inside certain Copilot plans. The unit of consumption and the actions that drive it are what govern the bill.
Copilot Studio meters consumption in messages drawn from a purchased capacity pool. A simple scripted reply consumes few messages, while a generative answer, a connected action, or an automated workflow consumes several. The bill scales with the richness of the conversation, not just the count of conversations, which makes the action mix the decisive cost driver.
The legacy session model priced a defined conversation window regardless of turns and still governs some existing agreements. Separately, certain Copilot and M365 plans seed a baseline of Copilot Studio capacity, which can cover light usage before any paid pack is required. Knowing which model an estate is on and what is seeded is the starting point.
Power Virtual Agents produces three recurring waste patterns. The first is sizing message capacity on the old session assumptions. The second is letting generative actions inflate the message draw without governance. The third is buying paid capacity that the seeded Copilot entitlement already covers.
An estate provisioned under session pricing carries forward the capacity assumptions from a model that no longer applies. Under message metering the same bot volume can consume far more or far less depending on the action mix. Buyers who never re estimated the message draw after the transition size to the wrong number and either overpay or hit overage.
Turning on generative answers and connected actions enriches the bot and multiplies the messages each interaction consumes. Without governance over which topics use the heavy actions, the message pool drains faster than the conversation count would suggest. The cost of the bot detaches from its volume and tracks the action design instead.
Certain Copilot and M365 plans seed a baseline of Copilot Studio capacity. Organizations that already hold those plans sometimes buy paid message packs on top without first consuming the seeded allocation. The paid capacity duplicates an entitlement the tenant already owns, a recurring charge for capacity that was included.
The Power Virtual Agents bill responds to three levers. The message estimation re sizes capacity to the real action weighted draw. The action governance contains the generative consumption to where it earns its keep. The seeded reconciliation consumes the included capacity before any paid pack.
We model the message consumption per bot using the actual topic mix and the action weighting, not the legacy session count. The output is a capacity pool sized to the real draw, with headroom for the known peaks. Re estimating after the transition is what separates an estate that pays for its real conversational load from one carrying forward an assumption that no longer holds.
The sized capacity then anchors the EA renewal so the Copilot Studio line reflects the message draw the bots actually generate.
Generative answers and connected actions are reserved for the topics where they earn their cost, with lighter scripted paths handling routine queries so the message pool is not drained by interactions that did not need the heavy actions.
The seeded Copilot Studio capacity inside existing Copilot and M365 plans is reconciled and consumed first, so paid packs are bought only for the draw that exceeds what the tenant already owns.
Copilot Studio negotiates inside the Power Platform and Copilot envelope of the broader Microsoft agreement. The message capacity packs and the overage terms give a buyer with a validated draw estimate real leverage on the committed rate.
The message packs price lower at higher committed tiers. A buyer who has estimated the action weighted draw can commit the capacity that lowers the unit rate while protecting against overage. Committing blind to an unvalidated estimate risks locking the wrong tier under a metering model the buyer has not yet measured. Committing on a real draw estimate turns the volume into a negotiated discount.
Microsoft is driving Copilot adoption hard, and Copilot Studio is part of that push. A buyer scaling conversational agents holds leverage to negotiate the message rate, the overage terms, and the relationship between the seeded Copilot capacity and the paid packs. Treating the conversational platform as part of the broader Copilot negotiation, rather than a standalone buy, is where the recoverable concession sits.
The engagement is a message draw and seeded capacity diagnostic, an action governance and sizing model, and the integration of the result into the Power Platform and Copilot negotiation. The output is a conversational estate priced at the message draw the bots actually generate.
We model the message consumption across every bot using the real topic and action mix, reconcile the seeded Copilot Studio capacity inside existing plans, and quantify the paid capacity the estate truly requires. We identify the bots whose generative actions inflate the draw and the capacity already covered by seeded rights. The output is a defensible picture of the real message requirement.
We govern the generative actions to the topics that justify them, size the message pool to the real draw, consume the seeded capacity first, and fold the result into the Power Platform and Copilot negotiation. We commit the tier for the discount, negotiate the overage protection, and lock multi year rates. The output is a Copilot Studio position priced at real message draw and defensible through the term.
The Power Virtual Agents diagnostic models the action weighted message draw, reconciles the seeded Copilot Studio capacity, governs the generative actions, sizes the pool to real usage, and brings the result into the Power Platform and Copilot negotiation. The result is a conversational line priced at the messages the bots actually generate, not at the legacy session assumptions.