.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
How Much Should You Pay for a Virtual Assistant Virtual Assistant Pricing Models Explained
The phrase “virtual assistant” now covers products that do radically different work. One may suggest code to a developer who remains fully responsible for the final output. Another may answer a customer, issue a refund, update a CRM record, and close a support case without a person stepping in. Treating both as the same purchase creates predictable trouble: either buyers overpay for assisted work or underfund an agent that carries real operating responsibility.
For this discussion, we mean software-based AI virtual assistants rather than outsourced human assistants. The central question is not whether AI is involved. It is what work the assistant completes, how often a human must review it, and whether the buyer can verify the result. Public price cards now span $19 per user per month for a coding copilot, $0.50 to $0.99 for a resolved customer conversation, and $2 for an AI-agent conversation.
Monetizely’s position is clear: buyers should pay primarily for verified completed work only when a virtual assistant can finish a bounded task without material human intervention. When people still do or closely review the work, the primary meter should remain a user seat or a capacity commitment.
The word “assistant” hides four different products
The market’s pricing confusion starts with a category error. An AI assistant can be a productivity tool, a workflow helper, an autonomous service agent, or an enterprise automation layer. Each removes a different amount of human work. Each also gives the supplier a different cost profile and gives the buyer a different budget risk.
Four public B2B SaaS price cards show how far apart those models have become as of September 7, 2026.
The table carries one important message: a published AI price is not a comparable unit until the buyer knows what event triggers the invoice. A $0.50 resolution, a $0.99 outcome, a $2 conversation, and a $19 seat answer four different commercial questions.
Monetizely’s 5-Step Pricing Framework puts the price point fourth for a reason. The sequence begins with goals and segmentation, then moves to packaging, pricing metric, price points, and operationalization. A company first decides which buyer it serves and what that buyer needs; it then builds the offer, chooses what will be counted, sets the rate, and makes billing and reporting work in practice. The sequence, developed more fully in Monetizing Agentic AI, prevents a common mistake: choosing a clever price before deciding what the customer is actually buying.
For virtual assistants, the pricing-metric decision rests heavily on autonomy. The Agentic Monetization Spectrum, or AMS, measures three things: zero-human ability, operational domain, and output-to-cost ratio. Zero-human ability asks whether a person still performs most of the work, mainly reviews it, or is largely absent. Operational domain asks whether the assistant handles one task, one workflow, or work across functions. The output-to-cost ratio asks whether the economic value rises roughly with cost or dramatically outpaces it. As all three rise, a seat becomes a weaker anchor and a verified output becomes a stronger one.
Applied to virtual assistants, the distinction is practical rather than theoretical.
| Virtual-assistant archetype | Zero-human ability | Operational domain | Output-to-cost ratio | AMS score* | Primary meter Monetizely recommends |
|---|---|---|---|---|---|
| Employee copilot, such as a coding assistant | 1 | 1 | 2 | 4 of 9 | Named user, with a defined capacity backstop |
| Workflow assistant that drafts, routes, and prepares work for approval | 2 | 2 | 2 | 6 of 9 | Platform or capacity commitment, then measured usage |
| Customer-service virtual assistant that closes routine cases | 3 | 2 | 2 | 7 of 9 | Verified resolution |
| Cross-functional virtual assistant that completes multi-system work | 3 | 3 | 3 | 9 of 9 | Verified business outcome, with a platform minimum |
*Scores use 1 for small, 2 for medium, and 3 for large on each AMS dimension.
The scoring points to a firm rule. GitHub Copilot belongs on a seat-led model because the developer still owns the work and judges its quality. A customer-service agent that authenticates a customer, finds an order, changes an address, and confirms completion has crossed a different threshold. Its price should follow the completed case, not the number of support managers who happen to supervise it. GitHub’s $19 per-user plan and HubSpot’s 50-credit resolved-conversation charge reflect that difference in role.
A pricing model is not just a way to collect revenue. It assigns risk between supplier and buyer.
Seat pricing gives buyers a stable bill. In exchange, they pay even when adoption is weak, and the supplier has limited direct exposure to whether the assistant produces useful work. That structure fits an employee copilot because availability and individual productivity are the product.
Credit and action pricing gives suppliers protection from variable model and system costs. The buyer, however, must forecast how many actions the assistant will take. Salesforce’s Flex Credit structure makes the unit explicit: a $500 pack contains 100,000 credits, and a listed Agentforce action consumes 20 credits.
Conversation pricing moves closer to business activity but remains a traffic meter. A surge in inbound contacts can raise spend even when the underlying problem lies in a product outage, late shipment, or billing error. A buyer should accept that meter only when a conversation is reliably close to completed work.
Outcome pricing puts more risk on the supplier. The vendor earns when the defined result occurs, while the buyer avoids paying for AI attempts that fail. Intercom states that Fin does not charge for unsuccessful attempts, but its outcome definition can include a procedure handoff. That is why buyers must read the definition beneath the headline rate.
A $2 conversation and a $0.50 resolution can produce a sixfold budget gap
Rate cards often look comparable in a sales presentation because every vendor promises automation. The financial picture changes once the buyer models the invoice trigger.
Consider a support organization with 10,000 inbound cases per month and a 60% verified autonomous resolution rate. The following figures apply each vendor’s public unit price to the event that vendor counts.
The difference between the first and third rows is $17,000 per month, or 6.7 times the spend, before comparing product capabilities, support-plan costs, implementation work, or contract discounts.
No serious operator should infer from that table that one vendor is universally superior. The products may handle different channels, integrations, and levels of complexity. The correct inference is narrower and more useful: the buyer must first decide whether it wants to pay for demand, activity, or completed work. Without that decision, a pricing comparison is largely theater.
Verified resolution is the strongest meter for autonomous service work
For routine customer service, Monetizely’s position favors verified resolution as the primary meter. It aligns spend with the work a support leader would otherwise need a person to perform. HubSpot’s current customer-agent rate equates 50 credits with a conversation resolved without human help. Intercom similarly bills Fin outcomes once per conversation, with a published $0.99 rate for standard support outcomes.
A buyer should demand three conditions before accepting that model:
Those conditions are especially important because a virtual assistant can improve its invoice count without improving the customer experience. An assistant that closes tickets too aggressively may produce a high “resolution” number and more repeat contacts. A proper definition must include a reasonable recontact window, quality controls, and a clear path for the buyer to challenge disputed events.
For routine, well-bounded cases, our recommended price target is $0.50 to $1.00 per verified resolution, excluding the cost of any separate helpdesk seats, implementation work, or unusually complex multi-system actions. HubSpot’s public rate establishes the lower end of that range; Intercom’s published $0.99 outcome rate establishes the upper end for a broadly deployed support agent.
Credits belong beneath the rate card, not at its center
Credits have a legitimate role in AI pricing. They give suppliers a way to recover variable inference and orchestration costs. They also help buyers test an emerging assistant without committing to a large number of seats.
The problem starts when credits become the customer’s primary value story. A support leader does not budget for “50 credits.” The leader budgets for fewer contacts that require a human, lower backlog, faster response times, and more retained customers. HubSpot converts credit use into a customer-facing business event by charging 50 credits for a resolved conversation. Salesforce offers both a conversation model and an action-based credit option, giving buyers a more direct choice between paying for interaction volume and paying for system work.
Our view is not that buyers should reject credits. Buyers should insist that credits sit behind a plain-language commercial promise. For an internal workflow assistant, a capacity pool may be sensible because the work is still reviewed and hard to value one task at a time. For an autonomous service assistant, credits should not replace a result-based primary meter simply because they are easier for the vendor to calculate.
A short pilot should prove prevented work before it earns scale funding
A 90-day deployment can answer the buyer’s central question: did the virtual assistant prevent paid work, or did it merely move work around?
The pilot should begin with a narrow workflow, such as order-status questions, password resets, appointment changes, or basic account updates. Broad launches produce too many confounding factors. A single workflow provides a clean baseline for contact volume, human handling time, escalation rate, repeat-contact rate, and customer satisfaction.
The following scorecard keeps the evaluation tied to operating results rather than AI usage.
| Test measure | What to compare | Threshold for expanding spend |
|---|---|---|
| Verified autonomous resolution rate | AI-resolved cases divided by eligible cases | A sustained rate high enough to remove measurable human workload |
| Repeat-contact rate | Customers who reopen or recontact within the agreed window | At or below the human-handled baseline |
| Escalation quality | Escalations with the right context and next action | Better than the prior routing process |
| Cost per prevented human case | Total AI spend divided by verified cases that did not need human handling | Below the fully loaded cost of the displaced human work |
| Billing accuracy | Vendor-billed outcomes compared with the buyer’s own event log | Near-perfect reconciliation before expanding the commitment |
The scorecard means a virtual assistant should earn broader deployment through evidence, not enthusiasm. A low unit price is expensive when it creates rework. A higher price can be justified when the assistant removes a real, auditable block of operating cost.
Procurement teams often focus on rate discounts. That is necessary but insufficient. Definitions, exclusions, and reporting rights will determine what the company actually pays over the contract term.
Before signing an outcome-based agreement, buyers should require:
These terms do not make a weak product strong. They do ensure that a promising product is measured against the work it claims to complete.
Virtual assistants will keep moving from copilots toward autonomous workers. The pricing model should move with them, but not ahead of their actual reliability. Paying an outcome rate for a heavily supervised assistant gives the vendor credit for work the customer still performs. Paying seats for an agent that independently resolves thousands of cases lets the vendor keep the upside while the buyer absorbs adoption risk.
The durable architecture has one named primary meter: verified completed work for autonomous, bounded workflows. A modest platform commitment may support security, integrations, reporting, and predictable vendor capacity, but it must not obscure the primary promise. The assistant earns scale when it completes more of the work the buyer values.
Operators should act on that position in five concrete ways:
Classify each virtual assistant before requesting proposals. Separate employee copilots, reviewed workflow assistants, and autonomous agents rather than placing them under one “AI budget” line.
Set a price ceiling from the human work removed. Use the fully loaded cost of the avoided case or task, then leave room for quality assurance, exceptions, and the vendor’s margin.
Launch with one high-volume, low-ambiguity workflow. Order tracking is a better first test than a complex dispute process because both completion and failure are easier to observe.
Make finance own the event definition with operations. The people approving invoices and the people handling exceptions should agree on what counts as completed work before a rate is negotiated.
Reprice the relationship when autonomy changes. An assistant that begins as a drafting tool may later complete work on its own. Do not let an old seat model or an old outcome rate survive after the underlying job has changed.
Assumptions. Public U.S. list prices were checked on September 7, 2026. The budget model assumes 10,000 inbound cases each month and 60% verified autonomous resolution; it excludes base software subscriptions, implementation, taxes, negotiated discounts, and differences in product capability.
Public sources make the price checks repeatable
Monetizing Agentic AI. https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
Monetizely, pricing framework and Agentic Monetization Spectrum source pages: Step 1: Goals and Segmentation; Step 2: Packaging; Step 3: Choosing the Right Pricing Metric; The Agentic Monetization Spectrum; Step 4: Finding the Right Price Points; Step 5: Operationalizing Agentic AI Pricing.
Intercom, Fin AI Agent outcomes and pricing.
Salesforce, Agentforce Pricing.
GitHub, About billing for GitHub Copilot.
HubSpot, HubSpot Credits and Customer Platform pricing.