.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.
Pricing AI Copilots: Monetizing Human-AI Collaboration
AI copilots create a pricing problem that looks simpler than it is. A human opens an application, asks an AI for help, reviews the answer and remains accountable for the work. That pattern resembles traditional SaaS closely enough that per-seat pricing feels natural. Yet every time the copilot searches more data, invokes a larger model or turns a five-second prompt into a 20-minute delegated task, the vendor incurs costs that no longer resemble ordinary SaaS.
Microsoft 365 Copilot makes the tension unusually visible. Microsoft announced the product at $30 per user per month in July 2023, removed its 300-seat commercial minimum in January 2024, and by July 2026 had passed 30 million paid seats. At that point, Satya Nadella told investors Microsoft was moving “beyond per-seat to per-seat-plus-consumption” as Copilot expanded from chat into Cowork and autonomous work.
Monetizely’s position is that an AI copilot should still be priced primarily per seat because the human remains part of the production system. Consumption should sit on top only when the AI takes on long-running, compute-heavy work independently. Outcome pricing belongs further along the autonomy curve, where the software owns enough of the result for the vendor and customer to measure it fairly.
Human-AI collaboration makes the person the most durable unit of value
The strongest reason to keep the seat is not historical habit. It is what the product actually does.
Peer-reviewed workplace evidence shows that copilots change the productivity of individual workers rather than producing a clean, independent business outcome. A 2025 Quarterly Journal of Economics study covering 5,172 customer-support agents found that access to generative-AI assistance increased issues resolved per hour by 15% on average, with much larger gains among less experienced and lower-skilled workers. The same AI therefore created different value for different people doing the same job.
A 2026 Organization Science field experiment involving 791 professionals at Procter & Gamble found that individuals working with AI could match the performance of human teams without AI. Yet the researchers also found that human judgement continued to matter in evaluating and selecting the best ideas. AI was a teammate, not an independent producer whose output could cleanly be billed as an outcome.
Another Organization Science experiment found the same boundary from a different angle. Knowledge workers using GPT-4 completed 12.2% more tasks and worked 25.1% faster on tasks within the model’s capabilities, but were 19% less likely to reach the correct answer on a difficult task outside that capability frontier.
Those findings matter directly for pricing. When the buyer is paying to make an employee faster, broader or better informed, the employee is still the clearest recurring unit around which access and value organise.
The ranking follows from that observation.
| Rank | Pricing metric | Monetizely verdict | What to do in practice |
|---|---|---|---|
| 1 | Per seat | Primary meter | Charge for the employee-AI working relationship: access, context, governance and everyday assistance. |
| 2 | Usage or credits | Secondary meter | Apply to delegated work where model choice, runtime, tools and retrieval cause meaningful variable cost. |
| 3 | Per outcome | Wrong primary meter for a copilot | Reserve for agents that independently own a measurable result such as a resolved case, completed review or booked appointment. |
The table is not an argument for old-style unlimited SaaS pricing. It says something narrower: the seat should buy collaboration; usage should buy delegation.
Microsoft’s pricing path shows exactly where the flat seat stops working
Monetizely’s 5-Step Pricing Framework starts with Goals and Segmentation, where the company decides which customers and economic goals the offer must serve. Packaging turns those needs into distinct combinations of capabilities. Pricing Metric chooses the unit that makes what the customer pays rise with the value received. Rate Setting determines how much to charge against willingness to pay, alternatives and unit economics. Operationalization then makes the design executable through entitlements, metering, billing, sales compensation, reporting and renewal processes. Developed in Monetizing Agentic AI, the framework matters for copilots because a clever rate cannot rescue the wrong meter: a product that shifts from helping a human to doing work independently may need a change in the unit being charged, not merely a higher price.
For Microsoft 365 Copilot, the most consequential steps are packaging, pricing metric and operationalization. The company has moved much further on the last two than the familiar “$30 AI add-on” description suggests.
| 5-Step Framework step | Grade | Monetizely assessment |
|---|---|---|
| Packaging | B+ | Strong separation between free/included chat, paid work-grounded Copilot and more autonomous Cowork, but the buyer must still understand several licences and metered experiences. |
| Pricing Metric | A- | The seat remains the anchor while Cowork introduces consumption tied to model use, retrieval, tool calls and runtime. The missing piece is a simple included allowance that makes the two meters feel like one offer. |
| Operationalization | A | Microsoft supports tenant, group and user spending controls, usage reporting, alerts, budgets and $0.01-per-credit PayGo billing for Cowork. |
The scorecard says Microsoft has already reached the right pricing direction. Its remaining problem is buyer simplicity, not meter selection.
The evolution is instructive because each reset followed a change in either distribution or product behaviour.
Microsoft has a powerful incentive to get the transition right. Its FY2025 10-K reported $87.8 billion of Microsoft 365 Commercial products and cloud-services revenue, up from $77.0 billion in FY2024, and explicitly identified Microsoft 365 Copilot as part of that business.
A pricing reset that destroys seat expansion would therefore attack an enormously valuable installed-base model. A reset that ignores AI compute is equally dangerous.
Microsoft’s July 2026 earnings make the second problem concrete. Amy Hood said Microsoft 365 segment gross-margin percentage had decreased slightly as Copilot usage increased, while GitHub Copilot margins “improved through the quarter with the June business model change to usage-based pricing.”
That is unusually direct evidence. Flat pricing becomes harder to sustain precisely when an assistant starts doing more work.
The autonomy score puts copilots in the middle, where outcome pricing goes too far
The Agentic Monetization Spectrum, or AMS, helps separate a true copilot from software that merely carries the same AI label. It evaluates a product on three dimensions. Zero-human ability asks how much useful work the system can complete without a person continuously steering it. Operational domain asks how broad the system’s authority is across tools, data and business processes rather than inside one narrow feature. Output/cost ratio asks whether the economic value of the work produced is large relative to the variable cost required to produce it. As scores rise, pricing can move further from access and closer to consumption or outcomes because the machine increasingly performs the work itself.
Microsoft now spans two different positions on that spectrum. Everyday Copilot in Word, Excel, Outlook and Teams remains deeply human-led, while Cowork is designed for complex, multi-step tasks that can run for much longer and invoke models, organisational context and tools. Microsoft explicitly prices Cowork based on model use, context retrieval, tool calls and runtime.
That distinction produces a useful AMS score.
| AMS dimension | Core Microsoft 365 Copilot | Copilot Cowork | Pricing implication |
|---|---|---|---|
| Zero-human ability | 2/5 | 4/5 | Core use still revolves around a person prompting, reviewing and acting; Cowork can carry multi-step work much further independently. |
| Operational domain | 3/5 | 4/5 | Core Copilot spans major Microsoft 365 apps; Cowork can combine work data, tools and longer task sequences. |
| Output/cost ratio | 3/5 | 4/5 | Everyday assistance can be supported inside a predictable subscription; long-running tasks have enough compute variation for Microsoft to meter them separately. |
| Total | 8/15 | 12/15 | The product family crosses the point where one meter is no longer sufficient, but the core offer has not become an outcome product. |
The AMS therefore supports a seat-first, consumption-second architecture, not pure consumption and certainly not outcome pricing for the whole product.
Outcome pricing sounds attractive because it promises perfect alignment. In a copilot, however, the question “who produced the outcome?” has no clean answer.
Suppose a salesperson uses AI to research an account, draft an email and prepare for a call. The opportunity closes three months later after discovery, negotiation, legal review and human relationship-building. Charging the copilot a percentage of the deal would attribute too much causal power to one contributor.
The research evidence points in the same direction. The 2026 P&G experiment found that AI could improve idea generation while human judgement retained value in selection, and the knowledge-work study found performance could worsen when people applied AI outside its capability frontier.
Outcome pricing becomes credible when those hand-offs disappear.
The market is converging on seats for assistance and usage for machine work
No major vendor has settled the market by itself. Yet several large B2B products are independently arriving at variants of the same design.
As of August 2026, the important distinction is not whether a vendor uses one price card or two. The question is what happens when AI shifts from being available to a person toward consuming substantial resources on that person’s behalf.
| Vendor and product | Meter as of August 2026 | Primary-source evidence | What the design tells us |
|---|---|---|---|
| Microsoft 365 Copilot | Per-user subscription for Copilot; Copilot Credits for Cowork usage | Microsoft says Cowork requires a Copilot user licence and charges credits based on model use, retrieval, tools and runtime. | Human collaboration stays on the seat; delegated work becomes variable. |
| GitHub Copilot | $19/user/month Business, $39/user/month Enterprise, with pooled AI Credits and paid usage beyond allowances | Current GitHub documentation lists 1,900 credits per Business user and 3,900 per Enterprise user; credits are pooled across the organisation. | The coding seat remains the anchor while expensive agent and model use gets a budget. |
| Google Workspace with Gemini | AI bundled into Workspace per-user plans; Business Standard $14/user/month on annual commitment | Google’s August 2026 pricing includes Gemini AI in Gmail, Docs, Meet and other Workspace tools. | When AI is part of everyday collaboration, bundling can increase the value of the underlying seat. |
| Zoom AI Companion | Included at no additional charge with eligible paid Zoom Workplace services | Zoom’s current product page says AI Companion is included with eligible paid plans; Zoom first adopted that approach in September 2023. | Low-friction copilot features can protect and strengthen the core seat rather than require a separate AI meter. |
| Salesforce Agentforce | Per-user access or consumption; $125/user/month flat-fee access and $500 per 100,000 Flex Credits on its current Agentforce pricing | Salesforce offers both flat user access and Flex Credits for customer- and employee-facing agents. | Employee assistance can remain user-priced while higher-volume digital labour can move toward actions and credits. |
The convergence is stronger than the headline prices suggest. None of these vendors needs to abandon the seat simply because inference has a variable cost.
GitHub provides perhaps the cleanest supporting architecture. Its Business and Enterprise plans still begin with a user licence, but included AI credits are pooled across the organisation, allowing heavy users to draw from the unused allocation of lighter users. Customers can then budget additional paid use.
Microsoft should pay close attention to its own GitHub business.
The practitioner commentary also reflects the shift from AI as feature to AI as colleague. Nadella described Microsoft’s commercial direction in July 2026 as “per-seat-plus-consumption”, while Marc Benioff said in February 2026 that Salesforce was “bringing humans and agents together on one trusted platform”. Salesforce reported Agentforce ARR of $800 million at the time, up 169% year over year.
Google offers a useful counterpoint because stronger AI does not automatically require a separate AI invoice. In Alphabet’s July 2025 earnings call, Sundar Pichai reported that “Gemini in Google Workspace is saving employees nearly three hours per week” at BBVA, which was rolling the product out to 100,000 employees. Google now includes premium AI in its per-user Workspace plans.
Zoom went further. In September 2023, Chief Product Officer Smita Hashim described including AI Companion in paid accounts as “disrupting the industry’s pricing model”. Zoom continues to include AI Companion with eligible paid Workplace plans rather than charging for each summary or prompt.
Taken together, those examples reject a common assumption in AI pricing: variable cost does not mean every interaction should become a billable event.
Microsoft gets the meter right but still exposes buyers to a double-payment problem
Microsoft gets three things right.
First, it protects the seat. By July 2026 Microsoft had more than 30 million paid Microsoft 365 Copilot seats, customers with more than 50,000 seats had grown more than sevenfold year over year, and enterprise customers deploying Copilot to most information workers had grown nearly 75% quarter over quarter.
Second, it has proof that the seat can create substantial labour value. Microsoft reported in July 2026 that NHS England planned a 505,000-person rollout after a trial measured average time savings of 43 minutes per employee per day. The same earnings call named KPMG at more than 276,000 professionals and HSBC at 200,000 seats.
Third, Microsoft now protects itself against the cost of autonomy. Cowork tasks consume credits according to four concrete cost drivers, and customers can set spending limits at tenant, group and user level.
What Microsoft gets wrong is the seam between the two charges.
A customer that has already bought a Copilot seat must also start paying once that copilot becomes useful enough to handle long-running delegated work. Economically, that makes sense. Psychologically, the transition can feel like paying first for the worker and then again when the worker becomes productive.
GitHub’s pooled allowance offers a better answer. Give the seat meaningful included agent capacity and let the enterprise use that capacity wherever demand appears.
The distinction matters because copilot adoption is uneven. The QJE field study found significant variation in productivity gains across workers, while Microsoft itself reports adoption and usage at the user level. A fixed amount of autonomous capacity assigned rigidly to every employee would leave credits stranded with light users while power users hit overages.
A pooled allowance solves that problem without abandoning the seat.
Microsoft’s next pricing reset should therefore be one committed architecture: every paid Copilot seat buys the collaboration layer plus a defined contribution to a tenant-wide Cowork credit pool; usage beyond that pool is charged at the existing credit rate. Administrators should see the consumption in one currency, with the cost of each delegated task visible before or immediately after execution.
Such a structure would do more than make billing friendlier. The base licence would answer, “How many employees work with Copilot?” The credit pool would answer, “How much work are we delegating to AI?” Those are different economic questions and deserve different meters.
Outcome measures should remain central to value selling rather than invoicing. Microsoft can prove a deployment saves minutes, increases throughput or shortens a workflow without charging a customer for every dollar of downstream benefit. The peer-reviewed evidence on human-AI performance reinforces why: collaboration improves outputs, but the human frequently remains responsible for judgement and final action.
The next pricing reset should make collaboration predictable and delegation elastic
Copilot vendors face a temptation to overcorrect. Per-seat SaaS looks old; outcome pricing looks new. Neither observation is a pricing strategy.
The economically sound move is to follow the work. When AI sits beside a human, the seat still maps naturally to the buyer’s organisation, permissions, adoption programme and productivity budget. Once the AI leaves that interaction and runs substantial work on its own, consumption provides the better link to both customer value and supplier cost.
Microsoft’s own results now demonstrate both sides. Paid Copilot seats are scaling, increased Copilot usage has put some pressure on gross margin, and GitHub’s move toward usage-linked charging improved margins during the June 2026 quarter. Monetizely’s position is therefore not that the SaaS seat is disappearing. The seat is becoming the admission price to human-AI collaboration, while consumption becomes the meter for machine labour performed beyond it.
For teams pricing copilots now, four decisions follow.
Design the business model around the human-to-AI relationship before optimising inference revenue. A product whose daily value comes from making accountants, developers, lawyers or sellers better should maximise penetration of the relevant workforce. Metering every prompt can suppress the very usage needed to create that value.
Create one explicit boundary between assistance and delegated execution. Normal chat, drafting, summarising and interactive analysis should sit inside the seat. Work that runs asynchronously, calls several tools or consumes materially more model resources should draw from a common usage balance.
Pool autonomous-work allowances at the company level. GitHub’s August 2026 design already pools AI Credits for Business and Enterprise organisations, which better accommodates the uneven usage patterns seen in workplace AI research.
Keep outcomes in the ROI case until the AI genuinely owns them. Use time saved, cases handled, code shipped or revenue influenced to justify the price. Move the invoice to a per-outcome basis only when human intervention is sufficiently small that both parties can agree what the AI actually delivered.
Assumptions
AMS scores are Monetizely assessments on a 1-to-5 scale, not vendor-reported measures. Prices are USD list prices unless a promotion is explicitly identified and reflect primary sources available through 13 August 2026. Product packaging and AI-credit schedules can change quickly; the analysis treats Microsoft 365 Copilot as the primary pricing teardown and uses the other vendors as evidence for the market architecture rather than as feature-for-feature substitutes.
Footnotes
Monetizing Agentic AI: https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
Microsoft, “Furthering our AI ambitions - Announcing Bing Chat Enterprise and Microsoft 365 Copilot pricing”, 18 July 2023: https://blogs.microsoft.com/blog/2023/07/18/furthering-our-ai-ambitions-announcing-bing-chat-enterprise-and-microsoft-365-copilot-pricing/
Microsoft, “Expanding Copilot for Microsoft 365 to businesses of all sizes”, 15 January 2024: https://www.microsoft.com/en-us/microsoft-365/blog/2024/01/15/expanding-copilot-for-microsoft-365-to-businesses-of-all-sizes/
Microsoft FY2025 Form 10-K, SEC: https://www.sec.gov/Archives/edgar/data/789019/000095017025100235/msft-20250630.htm
Microsoft, “Copilot Cowork is now generally available”, 16 June 2026: https://www.microsoft.com/en-us/microsoft-365/blog/2026/06/16/copilot-cowork-is-now-generally-available/
Microsoft Fiscal Year 2026 Fourth Quarter Earnings Conference Call, 29 July 2026: https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q4
Microsoft 365 Copilot business pricing, accessed August 2026: https://www.microsoft.com/en-us/microsoft-365-copilot/pricing
GitHub, Copilot billing for organisations and enterprises, accessed August 2026: https://docs.github.com/en/copilot/concepts/billing/organizations-and-enterprises
Google Workspace, AI tools and pricing, accessed August 2026: https://workspace.google.com/solutions/ai/
Zoom, “Zoom introduces Zoom AI Companion”, 5 September 2023, and current Workplace product pricing: https://news.zoom.com/zoom-ai-companion/
Salesforce, Agentforce pricing, accessed August 2026: https://www.salesforce.com/agentforce/developers/vibe-coding/pricing
Salesforce Fourth Quarter Fiscal 2026 Results, 25 February 2026: https://investor.salesforce.com/news/news-details/2026/Salesforce-Delivers-Record-Fourth-Quarter-Fiscal-2026-Results/default.aspx
Erik Brynjolfsson, Danielle Li and Lindsey Raymond, “Generative AI at Work”, Quarterly Journal of Economics, 2025: https://academic.oup.com/qje/article/140/2/889/7990658
Fabrizio Dell’Acqua et al., “The Cybernetic Teammate: A Field Experiment on Generative AI and Teamwork”, Organization Science, 2026: https://pubsonline.informs.org/doi/abs/10.1287/orsc.2025.20702
Fabrizio Dell’Acqua et al., “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality”, Organization Science, 2026: https://pubsonline.informs.org/doi/10.1287/orsc.2025.21838
Alphabet 2025 Second Quarter Earnings Call, 23 July 2025: https://abc.xyz/investor/events/event-details/2025/2025-Q2-Earnings-Call/