The Better the Product, the Fewer Seats Sold? AI Is Forcing Software to Reinvent Its Business Model

11/17 2026 585

Per-seat subscription is one of the most popular business models of the internet SaaS era. Software vendors are responsible only for "feature availability," not for guaranteeing final business output. But with the spread of AI applications and FDEs, the business model is changing.

One core logic here is that if the core value of an AI product is to help customers greatly improve workforce efficiency—so that work originally requiring 100 people can now be completed by only 30—then if software vendors continue to "charge per head," does that not mean the smarter their AI becomes and the more labor it saves customers, the fewer account seats they can sell?

Moreover, the low marginal costs and high gross margins of traditional software are still difficult for today's AI products to achieve. For today's AI products, every time a user issues a complex instruction or invokes a long-chain reasoning process, the backend is in real time burning expensive GPU compute and tokens. Continuing the old-era models of free land-grabbing or fixed flat-rate monthly packages makes it easy for high-frequency heavy users to punch through profit margins.

Therefore, a new endpoint is being anchored by a large number of AI companies: "how many things AI has actually accomplished."

I. From grabbing 1% of software tool fees to capturing 50% of outsourcing labor fees

Tool software companies feel most deeply how AI is reshaping profit models.

In the mobile internet era, the commercialization challenges of photo-editing software and office document software were quite significant. Mass users were accustomed to using basic features for free, and their willingness to pay for memberships for "a few more nice filters" or "a few more exquisite layout templates" often had an obvious ceiling.

But after AI became fully integrated into workflows, companies in these tracks such as Meitu and Kingsoft Office ushered in new growth.

According to Meitu's financial report for the first half of 2026, revenue from its imaging and design products business increased by 30.9% year on year to 1.77 billion yuan, global paid subscription users exceeded 18.44 million, driving the company's adjusted net profit up 39.5% year on year.

Similarly, Kingsoft Office's third-quarter report for 2026 disclosed that its revenue for the first three quarters exceeded 5.08 billion yuan, up 21.57% year on year, and AI features directly drove simultaneous jumps in personal paid conversion rates, average revenue per paying user, and the WPS 365 enterprise business.

Why is it that the same group of users who used to carefully calculate tens of yuan for software memberships are now willing to renew long term without hesitation? The core reason is certainly not the label of AI, but that at the moment users pay, the benchmark in their minds has changed.

When an ordinary user buys a photo-editing software membership, what they buy is "a leisure feature to smooth and filter selfies," and in their minds they compare it with other free photo-editing apps on the market. Today, however, a large number of those flocking to pay for "Meitu Design Studio" are small and medium-sized sellers on Taobao, Douyin, and cross-border e-commerce.

When a seller uploads a casually taken flat-lay photo of clothing, AI can automatically generate within seconds high-definition outdoor product images of models with different skin tones wearing the clothes, and even automatically complete the editing of spoken-word sales videos. At that moment, what the seller compares in their mind is the detail-page service that originally would have cost thousands of yuan and several days.

WPS AI follows a similar logic. When AI can directly generate a draft report based on an enterprise's historical documents, extract meeting minutes, and build a departmental knowledge base, what it delivers is no longer a text formatting tool, but a computer secretary.

Once the benchmark changes, enterprises' willingness to procure also changes accordingly.

Enterprise software procurement generally uses the "IT informatization budget" controlled by the CIO. But since AI serves the business, when enterprises pay for agents this year, payment decision rights are rapidly shifting from IT departments to specific business departments, and the relevant budgets can be allocated directly from human resources budgets, equipment operations and maintenance budgets, and marketing placement budgets.

For example, when a large manufacturing enterprise introduces an intelligent agent for predictive equipment maintenance, it spends the "O&M service fee" originally used to hire external maintenance teams; when a financial institution purchases an intelligent marketing outbound call system, it uses the "business commissions and market promotion fees" originally paid to outsourced telemarketing teams; when a state-owned enterprise deploys an intelligent agent for bidding compliance review, what it saves is the labor-hour cost of temporarily transferring dozens of business backbone staff to review bids full time.

Once this macro account is calculated clearly, one can understand where the incremental growth of AI business models comes from: in global economic data, enterprise spending on IT software procurement each year accounts for only about 1% of GDP, while spending on employee salaries, labor outsourcing, and professional services accounts for more than 50% of GDP.

When AI enables products to complete a specific job end to end, software products leap through the dragon gate and swim squarely into an ocean of manpower and service budgets dozens of times larger than before.

II. From "collecting rent by seat" to "sharing accounts by delivery results"

Once a product cuts into manpower and labor budgets, the old rule of "charging a fixed annual fee per account seat" is bound to loosen. That is because when enterprises buy tools, they accept bearing the risk of use themselves, but when buying labor and services, what they have always cared about is settling accounts according to how much has been accomplished.

Thus we see that enterprise AI products are setting off a RaaS (Result as a Service, pay by effect) trend.

According to the latest forecast released by IDC, by 2028 the traditional per-seat charging model in the software industry will be broadly rewritten, and 70% of software vendors worldwide will restructure their business models, shifting toward value-oriented pricing around business results, transaction volume, or automation outcomes.

In the Chinese market, as many as 66% of decision-makers among surveyed enterprises explicitly stated that, compared with buying account seats, they prefer to procure AI products "by business outcome," a proportion even noticeably higher than the global average.

Silicon Valley has made many attempts in this regard. Sierra, an AI customer service unicorn founded by OpenAI board chairman and former Salesforce co-CEO Bret Taylor, thoroughly abandoned from its first day the traditional customer service software industry rule of "charging by the number of human seats."

The contract terms Sierra offers large enterprise customers such as Nike and Sonos are very straightforward: it charges only by the number of issues AI resolves independently; if AI fails to resolve the consumer's issue and the conversation is forced to be transferred to a human customer service agent midway, then this conversation is not charged.

With this pricing model highly aligned with customer interests, Sierra's valuation quickly exceeded $10 billion in less than two years since founding, and annual recurring revenue (ARR) crossed the $100 million mark.

Even Salesforce, the founding patriarch that personally established the industry iron law of "per-seat subscription," after launching its core product Agentforce, also proactively broke its own ancestral rule. It first tested "charging $2 per AI conversation," and then during 2025 to 2026 fully upgraded to a more flexible "Flex Credits" system linked to results: enterprises consume credits according to the actions actually executed by agents and the business milestones achieved, welding software budgets directly to business output.

Finance, accounting, e-commerce, and content marketing are all areas where this model takes effect.

In the traditional enterprise ERP and financial software field, Yonyou Network officially launched BaaS 3.0 (intelligent finance outsourcing service) and the FDE (forward-deployed engineer) model in 2026. In the past, enterprises buying financial software still needed a bunch of supporting components; now AI agents take over the automated closed loop from invoice recognition and intelligent bookkeeping to tax filing, and the logic of enterprise payment has directly changed from "buying a financial software license" to "paying by document processing volume and financial compliance delivery results."

In addition, vendors providing AI virtual human livestreaming and TikTok short video matrix services are changing their business models to extremely low-threshold or even zero-cost broadcasting, taking a commission based on the proportion of GMV actually generated by agents in sales; some overseas marketing companies charge in overseas short-video customer acquisition based on guaranteed views brought by a single video, valid inquiry leads, or lead conversion and retention.

In this way, AI not only eliminates customers' hesitation and defensiveness when procuring new technology, but also gives technology vendors their first opportunity to share in the rich cake of customers' business growth.

III. When "zero marginal cost" disappears, how should products and organizations be restructured?

Of course, there is no such thing as a free lunch.

Crossing from "selling tools" to "selling results" opens up more than ten times the market space, but it also sets new thresholds for all AI products: one is how to cover underlying token compute costs, and the other is how the organization's DNA can shift from "sell and stop" to "deep accompaniment."

The cost issue needs no elaboration. Why do today's AI operating companies rarely dare to promise unlimited use of all top models and agent features for a monthly fee of a few dozen yuan?

Because deep reasoning models and multi-agent collaboration mean continuous, large-scale context and constant real-time connections, all of which require higher marginal costs. If the crude pricing of the past internet era, "fixed monthly fee with all included," continues to be used, then as long as 10% of heavy users run complex tasks around the clock, they can pierce 80% of the product's gross margin space.

In order to find a balance between ensuring certainty of user experience and locking in marginal compute costs, a three-tier hybrid pricing method of "base subscription + compute credit pool (Credits) + result premium" is becoming an experimental configuration for global AI products in 2026:

The first tier is the "base subscription fee": charging a relatively affordable fixed monthly or annual fee, corresponding to value such as long-term cloud memory storage, personal/enterprise knowledge base hosting, workflow orchestration interfaces, and unlimited responses from lightweight on-device/small models. This money is used to cover basic R&D and keep users within the ecosystem.

The second tier is the "flexible compute credit pool": when users need to invoke high-end deep reasoning models, generate high-resolution audio and video, or launch long-chain agent tasks, credits in the package are dynamically deducted according to task complexity; after credits are used up, users can buy top-up packs as needed. This avoids exposing ordinary users to obscure underlying technical bills of "how much per million tokens," while precisely making high-value, high-consumption heavy tasks pay for their own compute costs.

The third tier is "result sharing": in business scenarios with clearly quantifiable results such as customer service resolution, lead conversion, and transaction facilitation, process billing is skipped directly, and commissions are charged according to the final business indicators achieved.

What is deeper than pricing model adjustment is the restructuring of corporate organizational DNA forced by the business model.

For a SaaS company, the most core department is often the strong sales army, with the core being selling subscriptions. But under AI's new rules of "billing by consumption" or even "paying by result," "selling the product" is only the first step of a long march; only by helping customers truly produce business results can vendors get paid.

This requires AI product companies to complete three structural turns:

First, product selection must prioritize scenarios that are "low-ambiguity, strongly closed-loop, and easy to attribute."

Not all industries can immediately make pay-per-result work. Intelligent customer service, financial document review, coding assistance, e-commerce photo editing, and precision marketing for financial management can run out front because the delivery standards in these scenarios are extremely objective, black and white, with no room for wrangling. The more quantifiable the results and the clearer the boundaries of attribution, the stronger the explosive power of the AI business model.

Second, product design must shift from "providing a menu of features" to "encapsulating industry know-how."

In the past, making software meant placing hundreds or thousands of buttons on the interface and letting customers figure out how to combine and use them themselves; now, to be responsible for results, products must internalize the work experience of top industry experts into the agent's workflow in advance.

Just like the FDE (forward-deployed engineer) boom that has emerged in the industry this year, large-model engineers who understand technology must be tied together with industry veterans who understand business, going deep into customers' real business flows to clean data and optimize strategies, turning industry tacit rules into highly deterministic standard actions executable by AI.

Third, spawn a whole new ecosystem of AI dynamic pricing and billing infrastructure.

When software dynamically settles accounts in real time based on invocation counts, token consumption, and business achievement rates, enterprises' internal financial reconciliation and billing systems may also need to be overhauled. Overseas, startups such as Paid—founded by the former CEO of sales automation unicorn Outreach—have recently emerged, dedicated to providing billing infrastructure for AI companies. These ventures specifically help AI products design hybrid pricing strategies, monitor each customer's task profit margin, and automatically generate outcome-based bills.

Every generational leap in underlying technology ultimately crystallizes into a new unit of pricing. Once it was whole machines or hours; later, copies and licenses; then consumption and attention. Today, AI has merely found a new scale: charging by the "work closed loop" that machines independently complete.

Because technology vendors have proactively taken over the "tool use efficiency risk" that customers previously bore alone—directly delivering determined business results through the intelligence and compute of their own models—AI products are, in return, able to break out of the narrow software tool box and share in the giant pie of the global manpower and labor services market, which is dozens of times larger.

Only when products truly dare to be accountable for results can AI truly become the load-bearing wall of the industry for the next 20 years.

Source: Songguo Finance

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.