Market Map: AI Agents for Customer Support
1. Thesis
The market is restructuring from seat-based helpdesk software to outcome-priced autonomous resolution, collapsing the historic separation between "software vendor" and "labor provider" — the winners will be paid per resolved conversation, not per agent login, which threatens the core economics of every incumbent ticketing platform.
---
2. Segments
A. Incumbent Helpdesks Retrofitting AI
Companies: Zendesk (AI Agents, via Ultimate acquisition), Salesforce (Agentforce), Intercom (Fin), Freshworks (Freddy AI), HubSpot (Breeze)
Dynamics: Defending seat-based revenue while cannibalizing it; their moat is owning the system of record and workflow, but their pricing transition (Intercom's Fin at ~$0.99/resolution is the reference point) is destabilizing their own ARR models. Intercom has arguably moved fastest and most credibly.
B. AI-Native Resolution Platforms
Companies: Sierra (Bret Taylor's company), Decagon, Ada, Forethought, Lorikeet, Gradient Labs (less sure — smaller/newer)
Dynamics: Built agent-first with no legacy revenue to protect; selling directly on autonomous resolution rates and outcome pricing. Sierra and Decagon are the momentum leaders with large enterprise logos and aggressive valuations. Key risk: they sit on top of incumbents' data and could be squeezed.
C. Voice-First Agents
Companies: PolyAI, Parloa, Cresta (agent-assist + voice), Replicant, Retell AI (infrastructure-leaning)
Dynamics: Voice is the largest cost pool (contact centers spend most on phone support) and the hardest technically (latency, interruptions, telephony integration). Later to inflect than chat but larger prize; consolidation with chat-first players is likely.
D. Agent-Assist / Copilot Layer
Companies: Cresta, Assembled (Assist), Observe.AI, Level AI, ASAPP
Dynamics: Augments human agents rather than replacing them — easier enterprise sale, but structurally a transitional category. As full autonomy improves, assist tools must move up to orchestration or get absorbed.
E. BPO / Outsourced Labor Being Disrupted (and Responding)
Companies: Teleperformance, Concentrix, TaskUs, Foundever; disruptor hybrids like Crescendo (AI-plus-human managed service)
Dynamics: ~$300B of labor spend is the real TAM. BPOs are racing to become "AI-delivered service" companies before AI-native vendors eat their volume; public BPO valuations have already been compressed on this fear.
---
3. Where Value Accrues
*Segment B (AI-native platforms) captures the most new value, but Segment A retains the most absolute value if they execute.*
The deeper answer: value accrues to whoever converts labor budget into software revenue. A resolution priced at $1 replacing a $6–12 human interaction creates enormous margin room. AI-natives (Sierra, Decagon) are winning greenfield enterprise deals because they have no seat revenue to protect and can price on outcomes from day one. However, incumbents own the data, routing, and knowledge base — Intercom's Fin shows an incumbent can win if willing to self-cannibalize. The BPOs are the value donors, not winners, unless a Crescendo-style hybrid proves the managed-service model scales.
---
4. White-Space Opportunities
1. Cross-vendor agent QA, trust, and audit layer. As companies deploy autonomous agents from multiple vendors, no independent player provides continuous evaluation, hallucination/compliance monitoring, and regulator-ready audit trails across agents — the "Datadog + auditor for support agents." (Vendors self-report resolution quality today, which is a conflict of interest; Maihem and others touch AI testing broadly (less sure) but none owns this vertical.)
2. Agent-to-agent commerce/support protocol. Customers will increasingly send their own AI assistants to interact with companies' support agents. Nobody owns the machine-to-machine layer — authentication, structured negotiation (refunds, cancellations, disputes), and settlement between a consumer's agent and a brand's agent. Whoever standardizes this becomes the payment-network equivalent for post-sale service.