Enterprise-Only vs Everyone: Why the AI Support Market Is Splitting in Two

The AI customer support market is on track to reach $15.12 billion in 2026. The three most visible vendors — Decagon, Sierra AI, and Ada — have a combined valuation exceeding $15 billion. And yet, the vast majority of support teams cannot buy their products.

This is not a criticism of those companies. They made a rational choice: pursue the largest accounts, charge premium prices, and use logos like Chime, Duolingo, and Rocket Mortgage to build credibility. It works. But it has left a massive gap in the market — one that is now creating a structural split between enterprise-only AI and AI that everyone else can access.

If you lead a CX team — whether you have 10 agents or 500 — this split affects how you evaluate vendors. Here is what is happening and what to do about it.

The Numbers Behind the Split

Let us start with the pricing reality:

Vendor	Minimum Annual Contract	Typical Range	Pricing Model
Decagon	~$95K	$95K–$590K/yr	Annual contract
Sierra AI	~$150K	$150K–$350K+/yr	Annual contract
Ada	~$100K	$100K–$400K/yr	Annual contract
Forethought (pre-Zendesk)	~$40K	$40K–$160K/yr	Annual contract
Twig	$0 (free tier)	$5/ticket	Per-ticket
Intercom Fin	~$0.99/resolution	Variable	Per-resolution

Now consider what the typical support team looks like. According to Zendesk's 2025 CX Trends report, the median support team has 12–25 agents. These teams handle 2,000–15,000 tickets per month. Their CX technology budget — inclusive of their help desk, phone system, chat tool, QA platform, and workforce management software — is typically $50K–$200K per year total.

When one AI vendor's entry price exceeds your entire CX tech budget, you are not the target customer. And that describes the vast majority of support organizations.

Why Vendors Went Enterprise-First

The enterprise-first strategy is not irrational. There are good reasons the most funded vendors started at the top of the market:

Higher contract values justify high CAC. Enterprise sales cycles cost $30K–$80K in sales and marketing expense per deal. A $200K ACV makes that math work. A $20K ACV does not.

Fewer customers to support. Serving 200 enterprise accounts is operationally simpler than serving 5,000 mid-market accounts.

Stronger defensibility. A 12-month enterprise contract with custom integrations creates switching costs that protect revenue.

Venture capital incentives. When you raise $481 million (Decagon) or reach a $10 billion valuation (Sierra AI), your investors expect a path to hundreds of millions in ARR. Enterprise contracts get you there with fewer customers.

None of this is wrong. But it does mean that these vendors have optimized their product, pricing, implementation, and support model for a customer profile that represents less than 10% of the support teams in the market.

The Two Pricing Models

The market now offers two distinct pricing philosophies. Both serve teams of all sizes — the difference is how you buy, not what you get.

Dimension	Annual Contract	Usage-Based
How you pay	Fixed annual commitment, custom-negotiated	Per-ticket or per-resolution, scales with volume
Typical range	$95K–$590K+/year	$5/ticket to $500K+/year at scale
Implementation	Weeks to months	Minutes to days
Commitment	12-month minimum, custom SOW	Month-to-month or pay-as-you-go
Flexibility	High switching costs, deep lock-in	Low switching costs, fast iteration
Managed service	Dedicated CSM, custom configuration	Managed AI Specialists, pre-built integrations
Vendors	Decagon, Sierra AI, Ada	Twig, Intercom Fin

Note that Twig bridges both models — it offers per-ticket pricing but serves enterprise teams alongside Decagon and Sierra AI with the same autonomous AI agents, quality scoring, and enterprise-grade security. The distinction between these columns is purchasing model, not capability.

The Annual Contract Model

The annual-contract segment is well-served and getting more competitive. Decagon, Sierra AI, and Ada are all building sophisticated agentic platforms that can handle multi-step workflows — processing returns, modifying subscriptions, pulling data from internal systems, and escalating with full context.

These platforms are genuine. If you have the budget and the implementation resources, the technology works. Sierra AI reportedly hit $100 million in ARR by November 2025, which means large companies are buying and renewing.

The challenge in this segment is differentiation. When three vendors all promise "agentic AI for customer support" at similar price points, the decision often comes down to founder relationships, sales execution, and which vendor your board member knows. That is not ideal for buyers, but it is the current reality.

The Usage-Based Model

Usage-based pricing serves companies of all sizes that prefer flexible, outcome-aligned spending over annual commitments. Buyers include:

Series A–C startups with 10–30 agents and a CX leader who also manages the help desk, QA, and training.
Mid-market SaaS companies with 30–80 agents that have outgrown basic chatbots but want predictable unit economics.
E-commerce brands with seasonal volume spikes that make annual contracts risky.
B2B companies with technical support needs that require deep product knowledge, not just FAQ deflection.
Enterprise teams that want to deploy quickly and scale based on results rather than committing to a six-figure contract upfront.

These teams share common priorities:

Operational efficiency. They want AI that works without dedicating an engineering team to building and maintaining it.
Cost accountability. They want to tie AI spend directly to outcomes — paying per ticket resolved rather than committing to a fixed annual contract regardless of results.
Speed to value. They need results in days, not months. A 90-day implementation timeline is a dealbreaker when you are trying to hit next quarter's efficiency targets.

The Opportunity Beyond Annual Contracts

Teams that prefer usage-based pricing are not a small niche. They represent the majority of the market by company count, even if they represent a smaller share by revenue.

Consider the math. There are approximately 30,000 SaaS companies with more than 10 employees worldwide. The vast majority have customer support teams. If you add e-commerce, financial services, healthcare, and other verticals, the total addressable number of support organizations that could benefit from usage-based AI pricing is well into six figures.

The enterprise AI vendors are targeting the top 2,000–5,000 of these. Everyone else is making do with:

Platform-native AI (Zendesk AI, Intercom Fin) that works for simple use cases but lacks depth.
DIY solutions built on GPT-4 or Claude APIs, which require ongoing engineering maintenance.
Nothing — still running traditional support operations and waiting for a viable option.

This is the gap. And it is closing, but slowly.

What Teams Using Usage-Based AI Actually Need

Having spoken with hundreds of CX leaders across company sizes, the requirements for teams choosing usage-based pricing are remarkably consistent:

1. Transparent Pricing

Annual contracts with custom pricing require a sales call, a demo, a proposal, and a negotiation. Mid-market CX leaders want to see pricing on a website and model the cost against their ticket volume before they talk to anyone.

Per-ticket pricing ($5/ticket with Twig, for example) or per-resolution pricing ($0.99 with Intercom Fin) lets you calculate your cost today, this quarter, and this year. You can start small and scale. You can stop if it does not work without eating an annual contract. See Twig's pricing page for an example of this model.

2. Fast Time to Value

"We can have you live in 30 minutes" versus "implementation typically takes 6–8 weeks" is the difference between trying something this afternoon and scheduling a kickoff call for next month.

Fast setup does not mean low quality. It means the vendor has invested in pre-built integrations, managed onboarding, and sensible defaults so you do not have to configure everything from scratch.

3. Managed Operations

Enterprise vendors sell you a platform and expect you to operate it — configure workflows, train models, monitor quality, tune prompts, build escalation rules. That is fine when you have a dedicated AI operations team. It is not fine when your CX leader is also managing scheduling, training, and QA.

Managed AI services — where the vendor handles the ongoing tuning, monitoring, and optimization — are essential for the mid-market. This is the difference between buying a tool and buying an outcome.

4. Quality Visibility

The biggest fear CX leaders have about AI is not that it will be too expensive. It is that it will give bad answers and they will not know until customers complain.

Quality measurement must be built into the product, not bolted on. Self-evaluation, multi-dimensional quality scoring (accuracy, completeness, tone, policy compliance, etc.), and per-response visibility are not premium features. They are baseline requirements.

Twig's 7-dimension quality scoring is one approach. What matters is that you can see how every AI response performs, not just an aggregate satisfaction score.

5. Platform Agnosticism

Mid-market companies change platforms more often than enterprise companies. A team might start on Freshdesk, move to Zendesk, and then switch to Intercom as they grow. If your AI support layer only works on one platform, you lose your investment every time you switch.

Vendors with broad integration support — 30+ integrations across help desks, CRMs, knowledge bases, and internal tools — provide flexibility that platform-native AI cannot.

The Economics Are Changing

The cost structure of AI support has shifted meaningfully in the past 12 months:

Cost Component	2024	2026
LLM inference (per 1M tokens)	$15–$60	$1–$10
RAG infrastructure	Custom build required	Commoditized, vendor-managed
Quality evaluation	Manual QA, sampling	Automated, per-response
Integration development	3–6 months per integration	Pre-built connectors, days

Falling inference costs, maturing RAG infrastructure, and pre-built integrations mean that the cost of delivering AI support has dropped dramatically. Vendors that built their pricing models in 2023 — when inference was expensive and RAG was artisanal — are charging for a cost structure that no longer exists.

This is why per-ticket and per-resolution pricing is gaining traction. The unit economics now support a model where vendors can profitably serve smaller accounts without six-figure minimums.

How to Evaluate Usage-Based Vendors

If your team prefers usage-based pricing over annual contracts, here is what to prioritize:

Priority	What to Look For	Red Flag
Pricing transparency	Published pricing on website; per-ticket or per-resolution model	"Contact sales for pricing" with no ballpark
Setup speed	Live deployment in hours or days, not weeks	"Typical implementation is 6–8 weeks"
Managed services	Vendor handles tuning, monitoring, and optimization	"Here's the platform; your team manages it"
Quality measurement	Per-response quality scoring, automated self-evaluation	Quality measured only by CSAT or aggregate metrics
Integration breadth	30+ integrations across platforms	Deep integration with one platform only
Contract flexibility	Monthly or per-ticket billing; cancel anytime	12-month minimum with auto-renewal
Security	SOC 2 Type II; clear data handling policies	No compliance certifications; vague data policies

For a deeper evaluation framework, see our 15 questions every Head of Support should ask.

Where This Goes Next

The split in the AI support market is not temporary. It reflects fundamentally different pricing philosophies — annual contracts versus usage-based models — each with distinct trade-offs. Annual-contract vendors will continue to grow by winning large committed accounts. Usage-based vendors like Twig will grow by serving teams of all sizes that prefer flexibility and transparent pricing.

Some vendors bridge both worlds. Twig's per-ticket model scales from a 10-agent startup to a 500-agent enterprise team — the pricing model is different, not the capability. The question is not "which segment are you in?" but "which pricing and deployment model fits how your organization buys software?"

For now, if you are a CX leader evaluating options, the practical advice is straightforward: focus on the pricing model and deployment approach that matches your organization's buying process. Check Twig's customer stories for examples of teams across the size spectrum, and evaluate whether the per-ticket model works for your volume.

The AI support market is large enough for both models to thrive. The mistake is assuming that a six-figure annual contract is the only path to enterprise-grade AI support.

Market data sourced from Grand View Research, Zendesk CX Trends 2025, Statista, and publicly available vendor information as of March 2026.