AI-Native Services vs SaaS: Why Outcomes Beat Subscriptions

In early 2025, Sequoia Capital published a thesis that rattled the SaaS establishment: services are the new software. The argument was simple but radical. The total addressable market for global services is roughly $4.6 trillion. The total addressable market for software is around $800 billion. For two decades, venture capital focused almost exclusively on the smaller market because software had near-zero marginal costs. Services required humans, and humans do not scale.

AI changed the math. When you can automate 70 to 90 percent of the cognitive work inside a service delivery workflow, you unlock the economics of software inside a market that is 5.75x larger. That is not an incremental improvement. That is a category redefinition. And it is already happening across accounting, legal, insurance, compliance, and a dozen other verticals where the incumbents are still selling seats and subscriptions.

The core insight is that customers never wanted software. They wanted outcomes. A CFO does not want an accounting platform. She wants accurate books, filed taxes, and clean audits. A general counsel does not want a contract management tool. He wants reviewed contracts with flagged risk clauses, delivered on time. SaaS forced buyers to purchase tools and then hire people to operate those tools. AI-native services skip the middleman entirely and sell the outcome.

This is not a theoretical shift. Companies like Basis (AI-native accounting), Harvey (AI-native legal), and Tennr (AI-native medical records processing) have raised hundreds of millions of dollars on this thesis. They are not building better software for accountants, lawyers, or medical administrators. They are replacing the labor those professionals perform, packaging the output as a service, and charging for results. If you are planning a new venture or rethinking an existing product, this is the most consequential strategic question you can ask right now: should you sell a tool, or should you sell the work the tool does?

The numbers tell an uncomfortable story for pure SaaS. Global IT software spending is projected at roughly $800 billion to $1 trillion depending on whose estimate you trust. Global spending on professional and business services exceeds $4.6 trillion. That is not a small gap. It is a canyon. And the reason SaaS captured only the smaller market is that software companies could only sell tools, not labor replacement. AI closes that gap.

Consider accounting. U.S. businesses spend approximately $140 billion per year on accounting and bookkeeping services. The entire accounting software market, including Intuit, Xero, Sage, and every startup in the category, is worth about $18 billion. The software captures roughly 13 cents of every dollar spent on accounting. The other 87 cents goes to human labor: data entry, reconciliation, categorization, report generation, tax preparation, and audit support. An AI-native accounting service that automates 80 percent of that labor is not competing for the $18 billion software market. It is competing for a meaningful slice of the $140 billion services market.

The same pattern repeats in legal ($350 billion in U.S. legal services vs. $30 billion in legal tech), insurance ($45 billion in claims processing labor vs. $12 billion in insurtech software), and compliance ($25 billion in consulting fees vs. $8 billion in GRC software). In every case, the service TAM dwarfs the software TAM by 3x to 10x.

The implication for founders is direct. If you are building a SaaS product in a vertical where most of the customer spend goes to labor rather than tools, you are fishing in the wrong pond. The bigger opportunity is to absorb that labor spend by delivering the outcome directly. This is why AI agents for business are becoming central to so many go-to-market strategies. Agents do not just assist workers. They replace entire workflow steps, which means you can capture service-tier revenue instead of software-tier revenue.

The fundamental economic difference between a SaaS business and an AI-native service business comes down to what you charge for. SaaS charges for access: a seat, a license, a subscription. The value is implicit. You are paying for the right to use the tool, and the vendor hopes you find it useful enough to keep paying. AI-native services charge for outcomes: a completed tax return, a reviewed contract, a processed claim, a resolved support ticket. The value is explicit and measurable.

Let us compare the unit economics of both models using a concrete example. Imagine you are building for the mid-market accounting vertical.

SaaS model: You sell accounting software at $200 per user per month. A typical mid-market customer has 3 to 5 users, generating $600 to $1,000 in monthly recurring revenue. Your gross margin is 85 percent because your marginal cost of serving an additional seat is near zero. That sounds great until you realize the customer still needs to hire accountants at $60,000 to $90,000 per year to operate your software. Your product is a cost center layered on top of another cost center.

AI-native service model: You charge $50 per completed month-end close, $30 per reconciled account, or $500 per prepared tax return. The same mid-market customer that generated $800 per month in SaaS revenue now generates $2,000 to $5,000 per month in service revenue because you are capturing both the software value and the labor value. Your gross margin is lower, perhaps 60 to 70 percent instead of 85 percent, because you have real inference costs, quality assurance overhead, and human-in-the-loop review for complex cases. But your absolute gross profit per customer is 2x to 4x higher.

The math gets even more interesting at scale. SaaS businesses grow revenue by adding seats, which requires the customer to hire more people. AI-native service businesses grow revenue by processing more work, which scales independently of headcount. When your customer wins a big new client and needs twice the accounting throughput, the SaaS vendor collects maybe one or two additional seats. The AI-native service vendor collects twice the outcome fees. Your revenue scales with their business volume, not their payroll. For a deeper look at how to structure these pricing tiers, see our guide on how to price AI features.

The biggest objection to the services model has always been margins. Traditional professional services firms operate at 25 to 40 percent gross margins because humans are expensive. SaaS businesses enjoy 75 to 90 percent gross margins because software is cheap to replicate. AI-native services sit in the middle, but closer to SaaS than to traditional services, and the gap is closing fast.

Here is how the cost structure actually breaks down for an AI-native service business operating at scale. For every $100 in revenue, you are looking at roughly $8 to $15 in LLM inference costs (and falling 3 to 4x per year on a capability-adjusted basis), $5 to $10 in infrastructure and tooling, $5 to $8 in quality assurance and human review for edge cases, and $3 to $5 in customer-facing operations. That puts your cost of goods sold at $21 to $38, yielding gross margins of 62 to 79 percent.

The key to reaching and sustaining 60 percent or higher margins is a disciplined automation ladder. At launch, you might automate 60 percent of the work and have humans handle the rest. Your margin is thin, maybe 45 percent. Over the first 12 months, you instrument every human intervention, identify patterns, and build automation for the most common exception cases. By month 12, you have pushed automation to 80 percent and margins to 60 percent. By month 24, with enough training data from production work, you reach 90 percent automation and 70 percent margins. Basis, the AI-native accounting firm, has publicly discussed following exactly this trajectory.

Three tactics that move the margin needle fastest:

• Model routing. Not every task requires GPT-4o or Claude Sonnet 4. Route simple categorization and extraction tasks to smaller, cheaper models like Claude Haiku or GPT-4o-mini. Reserve frontier models for complex reasoning steps. This alone can cut inference costs by 40 to 60 percent without meaningful quality degradation.
• Batch processing. Real-time inference is expensive. If your service SLAs allow for 4 to 24 hour turnaround rather than instant results, batch your inference calls during off-peak hours when API pricing is lower, and use asynchronous processing to maximize throughput per dollar.
• Fine-tuned domain models. After processing thousands of domain-specific tasks, you have a gold mine of training data. Fine-tune smaller models on your production data to handle routine cases at a fraction of the cost of general-purpose frontier models. A fine-tuned Llama 3.1 70B can often match GPT-4o quality on narrow, well-defined tasks at one-tenth the inference cost.

The margin story gets even better when you factor in the learning curve. Every task your system processes generates data that makes future tasks cheaper and more accurate. SaaS products do not get cheaper to operate over time. AI-native services do. That compounding cost advantage is one of the most underappreciated aspects of this model.

One of the sharpest criticisms of AI-native services is the moat question. SaaS businesses defend themselves with switching costs, network effects, and data lock-in. If your product is "AI does the work," what stops a competitor from doing the same work with the same foundation models?

The honest answer: traditional SaaS moats are weaker than people think, and service moats are stronger. Let me explain.

Data flywheel moat. Every outcome you deliver generates proprietary training data. After processing 50,000 tax returns, your system has seen edge cases that no competitor starting from scratch will encounter for years. You can fine-tune models on this data, build specialized classifiers, and create decision trees that encode domain expertise no foundation model possesses out of the box. This moat deepens with every customer and every task. It is the AI-native equivalent of a network effect, and it is extremely difficult to replicate.

Workflow integration moat. When your service plugs directly into the ERP of a customer, bank feeds, payroll system, and tax filing infrastructure, switching costs are real and high. You are not just another subscription they can cancel. You are embedded in their operational workflow. Ripping you out means finding someone else to do the actual work, migrating process-specific configurations, and risking errors during the transition. That stickiness exceeds what most SaaS tools achieve with data lock-in alone.

Quality and trust moat. In professional services, reputation and track record matter enormously. An AI-native accounting service that has processed 10,000 month-end closes with a 99.7 percent accuracy rate has a credential that no new entrant can claim on day one. Trust compounds. Referrals compound. In verticals like legal, insurance, and compliance where errors carry regulatory consequences, the switching cost is not just operational. It is psychological. Nobody wants to be the person who moved to an unproven service and caused an audit failure.

Regulatory and compliance moat. In regulated industries, your compliance posture becomes a moat. If you have SOC 2 Type II, HIPAA BAAs, state-specific certifications, and a track record of passing audits, you have built a barrier that takes competitors 12 to 24 months and significant capital to replicate. The compliance bar in sectors like healthcare, financial services, and insurance is high enough to deter casual entrants.

The bottom line: AI-native service businesses build moats differently than SaaS businesses, but the moats are real and, in many cases, deeper. A SaaS product with commodity features and no data advantage can be cloned in months. A service business with proprietary training data, embedded integrations, and a proven quality track record cannot.

Not every market is right for the services model. Here is a practical framework for deciding which approach fits your opportunity.

Build an AI-native service when:

• Total customer spend on human labor for the task is 3x or more what they spend on software tools. This is the TAM signal. Accounting, legal, insurance claims, compliance audits, medical coding, and bookkeeping all qualify.
• The output is standardized enough to quality-check programmatically. Tax returns have deterministic rules. Contract reviews have checklists. Reconciliations have matching logic. If you can write automated QA checks for 80 percent of the output, you can build a scalable service.
• Customers already buy the outcome from humans. If there is an existing market of firms selling the deliverable (completed books, reviewed contracts, processed claims), you have proof that buyers value outcomes, not tools.
• Regulatory requirements create trust barriers that favor established players. Industries where compliance matters reward companies that invest early in certifications and audit readiness.

Build SaaS when:

• The primary value is creative, strategic, or inherently collaborative. Design tools, project management, and communication platforms still work better as tools because the human judgment is the point, not the bottleneck.
• The output is too subjective or varied to quality-check at scale. If every deliverable requires a senior expert to review, your margin structure will look like a traditional services firm, not an AI-native one.
• Network effects drive the value. Platforms like Slack, Figma, and Notion get more valuable as more people use them. A service business does not benefit from network effects in the same way.
• The buyer wants control and customization over process. Some customers, particularly enterprises, want to own their workflow and will resist handing it to a service provider, regardless of efficiency gains.

The most interesting opportunities right now sit at the intersection: service-wrapped software. You build proprietary software that powers the service delivery, but the customer never sees or operates the software directly. They get outcomes. You own the platform. This gives you the margin profile of a software company, the TAM of a service company, and the defensibility of both. That is the model Sequoia is betting on, and it is the model we help clients architect every week.

If you are looking for where to build, these are the verticals where the services-over-SaaS thesis is playing out fastest and where the opportunity is still wide open.

Accounting and bookkeeping. This is the poster child. U.S. businesses spend $140 billion on accounting services annually, there is a severe CPA shortage (the profession lost 17 percent of its workforce between 2020 and 2024), and the work is highly structured and rules-based. Basis has raised over $100 million to build an AI-native accounting firm. Pilot was an early mover with a tech-enabled services approach. The opportunity is enormous for vertical-specific players: AI-native accounting for e-commerce sellers, for restaurants, for medical practices. Each niche has unique chart-of-account structures, tax rules, and compliance requirements that general-purpose tools handle poorly.

Legal operations. Contract review, regulatory compliance, litigation support, and legal research are all tasks where AI can handle 70 to 85 percent of the work. Harvey raised $300 million to build AI for law firms. But the bigger opportunity may be in building AI-native legal services that bypass law firms entirely for routine matters: standard contract generation and review, trademark filings, corporate governance documentation, and compliance monitoring. The key constraint is the unauthorized practice of law (UPL) rules, which vary by jurisdiction and limit what non-lawyers can deliver. Structure your service carefully.

Insurance claims and underwriting. Claims processing is a $45 billion annual cost center for U.S. insurers. Most of it is document review, data extraction, coverage verification, and payment calculation. These are tasks that AI handles exceptionally well. Companies like Tractable (auto claims) and Shift Technology (fraud detection) have proven the technology works. The next wave is full-service AI-native claims processing, where an AI system receives a claim, extracts all relevant information, verifies coverage, calculates the payout, flags fraud indicators, and presents a decision-ready package to a human adjuster. That workflow eliminates 60 to 80 percent of the manual labor in claims operations.

Compliance and regulatory reporting. Every financial institution, healthcare provider, and publicly traded company spends significant money on compliance. Much of that spend goes to consulting firms and in-house compliance teams performing repetitive monitoring, documentation, and reporting tasks. AI-native compliance services can continuously monitor regulatory changes, map them to existing policies, flag gaps, generate updated documentation, and prepare audit-ready reports. The compliance consulting market alone exceeds $25 billion in the U.S., and the work is structured enough for high automation rates.

Other verticals worth watching: medical coding and billing ($15 billion in services spend), tax preparation ($14 billion), real estate transaction coordination ($9 billion), and HR administration ($30 billion). The pattern is consistent: wherever you find large pools of human labor doing structured, rules-based cognitive work, you will find an AI-native service opportunity. For a deeper exploration of how to structure agentic pricing in these verticals, check out our guide to agentic SaaS pricing models.

If you are convinced the opportunity is real, here is how to move from thesis to execution without overbuilding or underdelivering.

Step 1: Pick a narrow vertical and a specific outcome. Do not try to build "AI-native accounting." Build "AI-native month-end close for Shopify merchants doing $1M to $10M in annual revenue." The narrower your initial scope, the faster you reach the quality bar that customers will actually pay for. You can expand later.

Step 2: Start with a human-in-the-loop delivery model. Your first 50 to 100 customers should receive a service that is part AI and part human review. This lets you gather the training data you need to automate further, identify the edge cases that will break a fully automated system, and build trust before you remove the safety net. Your margins will be thin early. That is fine. You are investing in the data flywheel.

Step 3: Instrument everything. Every human intervention should be logged with the reason for the intervention, the input that triggered it, and the correction that was made. This data is what you use to push automation from 60 percent to 90 percent over the first 18 months. Without it, you are stuck with expensive human labor forever.

Step 4: Build pricing around outcomes from day one. Do not charge per seat or per month. Charge per deliverable: per tax return, per reviewed contract, per processed claim, per completed reconciliation. Outcome pricing aligns your incentives with the customer, makes your value proposition concrete, and lets you capture more revenue as you become more efficient. Your cost per outcome drops as automation improves, but your price per outcome stays stable or increases with quality improvements.

Step 5: Invest in compliance early. If you are operating in a regulated vertical, get SOC 2 Type II, relevant industry certifications, and proper data handling agreements in place before you scale. These are table stakes for enterprise customers, and they take 6 to 12 months to obtain. Treating compliance as an afterthought is the single most common mistake we see from technical founders entering regulated markets.

The shift from SaaS to AI-native services is not a trend. It is a structural change in how technology companies create and capture value. The $4.6 trillion services market is opening up to software-style economics for the first time, and the founders who move now will build the defining companies of the next decade.

If you are exploring an AI-native service model and want help with architecture, pricing, or go-to-market strategy, we work with founders on exactly this every week. Book a free strategy call and let us figure out the right approach for your vertical.