DEV Community

Prince Raj
Prince Raj

Posted on

Why mid‑market firms should pay per resolved ticket, not per seat

#business #ai

Why mid‑market firms should pay per resolved ticket, not per seat

Most leaders still frame AI as "build vs buy." That framing misses the point in 2026. The surprising reality: for many mid‑market companies the right choice is not to build or to buy feature add‑ons from incumbent SaaS — it's to buy outcome‑priced vertical agents (per ticket, per matter, per encounter). This is counterintuitive because it asks you to trade a software subscription for a labor‑priced contract. The evidence shows that, when vendors own the workflow and charge per outcome, time‑to‑value shortens, integration costs fall, and ROI becomes visible.

The common assumption and why it fails

Leaders assume incumbents (Salesforce, ServiceNow, HubSpot) are the safe bet: they already run your workflows and can bolt AI features onto existing licenses. That seems lower risk than hiring engineers or trusting a niche vendor.

But three forces have flipped the economics. First, inference costs collapsed dramatically (a16z documents a fall from roughly $60/M output tokens in 2021 to ~$0.05–$0.10/M by late 2024), removing the model itself as a durable moat. Second, AI‑native products run materially lower gross margins (52–65% vs ~80–90% for classic SaaS per ICONIQ/Bessemer), and incumbents are passing those costs through via per‑seat hikes and caps — effectively a stealth margin tax. Third, vertical agents capture the labor dollars being displaced and can price outcomes, which aligns incentives and amortizes integration across many similar buyers.

A concrete contract example

Outcome‑priced agent contract (typical market shape):

  • Price: $X per resolved ticket (or $Y per matter / $Z per encounter).
  • KPIs: resolution accuracy ≥ 95% on agreed categories, <N% escalation rate to human, <T days to integrate with existing CRM/ERP.
  • Ramp: pilot converts to production with fixed onboarding fee amortized over first 6 months.

Why this beats a seat: imagine a support function where each resolved ticket costs $12 in labor. A vertical agent that charges $6–8 per resolved ticket turns labor line items into a predictable unit cost, with the vendor accountable for quality and throughput. Incumbent add‑ons, by contrast, often increase subscription spend and leave integration work and quality on your plate.

Short case sketches (numbers matter)

  • Sierra (support): outcome‑priced, hit ~$100M ARR in ~7 quarters. Fast ramp suggests customers are getting predictable ROI quickly.

  • Harvey (legal): outcome pricing for law firms, ~$300M ARR by Q2 2026 — buying by outcome maps directly to billable hours replaced.

  • Klarna (cautionary): initially claimed an OpenAI assistant replaced 700 agents and saved ~$40M/year, then rehired humans after CSAT dropped. Klarna later clarified the AI handles work equivalent to ~800 FTEs but with humans in the loop. This shows pure replacement narratives fail without workflow‑level accountability.

These examples show two things: vertical agents scale when their unit economics align with buyer labor economics, and pure tool installs fail when quality or workflows aren't owned end‑to‑end.

The mechanism that makes outcome pricing repeatable

Outcome pricing works repeatably when three capabilities are standardized:

  1. Vertical integration adapters. Successful agents ship a suite of connectors and mapping logic for CRMs, ticketing systems, and document stores so integration is weeks, not months.

  2. Vertical ontologies and templates. Agents reuse industry‑specific taxonomies (support intents, legal matter types, clinical encounter codes) so much of the training and routing is shared across customers.

  3. Aggregated signal. Firms like Sierra and Harvey pool anonymized interaction data and distill it into shared improvements, which reduce per‑customer fine‑tuning costs and improve quality faster than isolated in‑house models.

These three forces let vendors keep onboarding costs low, meet SLA‑style guarantees, and therefore offer per‑outcome pricing that is attractive to mid‑market buyers.

When this is not the right move

Exceptions matter. If you have decades of proprietary transaction logs or regulated data that cannot leave premises (pharma R&D, certain fraud systems, clinical diagnostics), an internal build with private models can still win because no vendor can replicate your dataset. Also, highly regulated firms may need internal models for auditability and residency constraints.

Quick cost math to test a buying decision

Use three numbers to decide fast:

  • Your current labor cost per unit of outcome (e.g., labor per ticket).
  • Vendor outcome price.
  • One‑time integration fee amortized over 12 months.

If vendor price + amortized integration < current labor + marginal incumbent seat increase, outcome pricing wins. Note the break‑even scale math: in‑house projects often require $1.5M+ upfront and only break even above 100–200M tokens/day (DreamFactory); most mid‑market companies never reach that volume.

Honest caveats

The MIT NANDA finding that 95% of GenAI pilots delivered zero measurable P&L impact is based on public deployments, interviews, and surveys rather than a randomized trial — treat it as strong suggestive evidence, not a causal law. Some vertical agents are unproven at very large scale and their margins will compress as API and inference economics evolve. Finally, careful contracting and right‑sized SLAs are essential: outcome pricing transfers risk to the vendor, but buyers must verify data handling, compliance, and escalation terms.

What to do next

For CFOs and COOs at $50M–$1B firms: run a two‑week financial test. Calculate labor per outcome, get an outcome‑priced pilot quote from a vertical agent, and compare landed cost vs your incumbent add‑on (including expected seat increases and integration work). If the vendor meets KPIs and the math is favorable, outcome pricing will usually deliver faster payback.

Closing

The surprising truth for 2026: the real decision isn't build or buy — it's whom you buy outcomes from. For many mid‑market firms, paying per resolved ticket from a vertical agent will be cheaper, faster, and less risky than either building internally or buying AI add‑ons from incumbents.

Sources

Top comments (0)