The interactive market map for AI sales agents and customer-facing agents — the voices, the tools, the leaders, the technology underneath, the money, the claims they make, and the part the decks leave out.
A neutral landscape, not a pitch. Tap any card to open it — website shot, the headline claim, and the KPIs that vendor sells on. Figures are sourced and dated; where a category is contested, it says so.
01 — The map
Two sides, eight lanes
“AI sales agent” and “customer-facing agent” are not one market — they split into lanes with different buyers, latency budgets and success metrics. The lanes share a tech stack but little else. Vocabulary in the definitions.
Revenue side — agents that go and sell
Lane 1 · Outbound / AI SDR
Prospect & book
List, research, personalise, send, follow up, book. The most hyped — and most contested at cold scale.
Lane 2 · Enrichment / signal
Find & enrich
Data, intent and de-anonymisation that feed the machine. The picks and shovels.
Lane 3 · Inbound qualification
Qualify inbound
The layer that replaces the demo form — qualify in real time, route the hot ones.
Lane 4 · Support / deflection
Resolve & deflect
Answer, resolve, escalate the rest. Priced per resolution; clearest ROI, biggest rounds.
Service side — agents that talk to your customers
Lane 5 · Agent assist
Whisper to humans
AI beside the rep — prompts, summaries, next-best-action. The defensible augmentation lane.
Lane 6 · Contact centre
Run the front door
IVR replacement, routing, workforce management with voice AI up front. The platform play.
Lane 7 · Voice infrastructure
Build your own
ASR, TTS, orchestration and per-minute voice APIs everything else is assembled from.
Lane 8 · Incumbent platform
The CRMs strike back
The giants bundling agents into software you already own. Distribution is the weapon.
02 — The voices
The named agents
The category gave its agents names and faces — a tell that these are sold as “digital workers,” not features. The personas you'll meet across the decks.
Ava
ArArtisan
All-in-one outbound platform; Ava runs the full SDR motion over a 300M-contact database.
Alice
1111x
Digital workers Alice (outbound) and Julian (inbound phone). a16z- & Benchmark-backed.
“Hire Ava, the AI BDR.” Automates the end-to-end outbound workflow — discovery, research, personalised send, follow-up, booking. Publicly softened from ‘replace your reps’ to augmentation through 2025.
KPIs they sell on
meetings bookedreply ratecost per meetingemails sent
11
11xAlice
+
Digital workers Alice (outbound) and Julian (inbound phone). a16z- & Benchmark-backed.
“Autonomous digital workers” across email, LinkedIn, SMS, WhatsApp and voice. A 2025 TechCrunch report alleged overstated customer claims; community testing put churn at 70–80%.
“AI agents for the enterprise workflow.” Embedded in the system of action.
KPIs they sell on
tasks automatedtime saveddeflection %
04 — Ranking
The heavyweights
Ranked by disclosed valuation / deal size — a money-and-traction ranking, not a quality verdict. It tilts to the customer-facing side, which is where the largest rounds sit.
01
Si
Sierra
Support / deflection
$15.8B
May 2026 · $950M round · ~$150M+ ARR
02
De
Decagon
Support / deflection
$4.5B
Jan 2026 · $250M Series D
03
Cl
Clay
Enrichment / signal
$3.1B
Aug 2025 · $100M Series C
04
Pa
Parloa
Contact centre
$3.0B
2025 · $350M Series D · $560M+ raised
05
Co
Cognigy
Contact centre
$955M
Jul 2025 · acquired by NICE (~25× revenue)
The outbound / AI-SDR names (11x, Artisan) raised fast in 2024–25 but don't make this list on disclosed, current valuation — and several saw their numbers questioned (see the reality check). The durable capital favours support, infrastructure and the incumbents.
05 — What's hot / new
The last twelve months
The moves that reshaped the board — the rounds, the first big acquisition, and the narrative correction on the sales side.
May 2026
Si
Sierra raises $950M at $15.8B
The CX-agent bellwether nearly doubles its valuation in eight months; reported $150M+ ARR.
Jan 2026
De
Decagon hits $4.5B
$250M Series D one year after a $1.5B round; revenue reportedly tripled in Q3 2025.
2025
Pa
Parloa reaches $3B
$350M Series D, $560M+ raised in under four years — EU's contact-centre challenger goes global.
Jul 2025
Co
NICE buys Cognigy for $955M
Consolidation begins — a CCaaS incumbent pays ~25× revenue to own the agent layer.
FY2025
Sa
Agentforce passes ~$800M ARR
Salesforce's 'quarter of Agentforce' — the incumbent bundles agents into the CRM at scale.
2025
11
AI-SDR: replacement → augmentation
The loudest outbound brands walk back 'fire your reps' after a churn-and-claims reckoning.
06 — Market status
The money
Estimates vary wildly by analyst and by how the category is drawn — “AI agents,” “conversational AI” and “AI for customer service” overlap and double-count. Treat the direction as solid and any single number as one house's view.
$7.8B→$52.6B
AI-agents market, 2025→2030 (46% CAGR)
MarketsandMarkets
~$4.7B
Voice-agent revenue, 2025
Gartner, via Perspective AI
$2.9B
VC into CX agents (cumulative)
Hogan / industry tally
$15.8B
Sierra valuation, May 2026
CNBC · Sacra
$4.5B
Decagon valuation, Jan 2026
Sacra
60%
seller tasks automatable by 2028 (forecast)
Gartner
07 — Claims & KPIs
What they promise
The metrics the category sells on. Two different scoreboards for the two sides — and a caveat: vendors quote their best deployments, and ARR / churn reporting in the SDR lane has been openly questioned.
deflection / containment %automated resolution ratecost per resolutionCSAT (vs human parity)average handle time (AHT)first-response timeQA coverage %languages supported
Read the claim, then the basis. “90% resolution” or “3× pipeline” are real for some deployments on warm, bounded problems. Ask: which deployment, which baseline, measured how. Open any card above to see the headline claim each vendor leads with.
08 — The technology
What's under the hood
A modern voice agent is a stack, not a model. What's sold as “an AI agent” might be any one layer — or a vertical product assembled from all of them. The layers tell you who owns the hard part.
Layer 1 · Ears
Speech-to-text
Transcribes the caller in real time. The latency floor of the system.
Deepgram · Whisper · AssemblyAI
Layer 2 · Brain
The LLM
Understands, reasons, decides the next move and which tool to call.
Claude · GPT · Gemini
Layer 3 · Voice
Text-to-speech
Speaks the reply — naturalness and sub-300ms latency are the battleground.
ElevenLabs · Cartesia
Layer 4 · Glue
Orchestration
Turn-taking, interruptions, tool calls, state — the conductor of the call.
LiveKit · Pipecat
Layer 5 · Memory
Retrieval & tools
Grounds the agent in your data and lets it act — CRM, knowledge base, APIs.
RAG · MCP · CRM APIs
Why it matters: the LLM is increasingly a commodity the agent vendors rent. The defensible work is orchestration, latency engineering, retrieval over the customer's own data, and the evaluation harness that proves the agent is safe to let loose. “Built on GPT” is not a moat.
09 — Reality check
What the decks omit
A neutral map owes you the counter-narrative. The technology is real and improving fast — but 2024–25 also delivered a hype correction that reshaped how the category sells itself.
The reckoning
AI-SDR claims got audited
A 2025 TechCrunch investigation reported 11x claimed customers it didn't have; community testing put AI-SDR churn in the 70–80% range, and ARR figures were widely flagged as misleading. Cold outbound is the hardest place for an agent to win.
The pivot
Replacement → augmentation
The two loudest AI-SDR brands — 11x and Artisan — both walked back “fire your reps” toward “augment your team.” The market that endured leans copilot and human-in-the-loop, where ROI lands without betting the function on full autonomy.
Where it works
Warm, bounded, high-volume
The wins concentrate: support deflection (Klarna-scale, resolution-priced), warm inbound qualification, agent assist, data enrichment. Bounded problems with abundant ground truth — not open-ended cold persuasion.
10 — Definitions
The vocabulary
The terms that recur across this map, in plain language.
AI sales agent
Software that autonomously runs sales tasks — prospecting, research, outreach, qualification — with a named persona and some independence. The revenue side.
Customer-facing agent
An agent that talks to your customers — deflecting tickets, answering calls, qualifying inbound. The service side. Increasingly priced per resolution.
AI SDR / BDR
An agent built to replace or augment the sales-development rep. The most hyped — and most contested — category.
Deflection / containment
The share of inbound an agent resolves without a human. The headline KPI for support agents and the basis of per-resolution pricing.
Agent assist
AI alongside a human — live prompts, summaries, next-best-action. The augmentation lane; defensible because humans stay in the loop.
Latency budget
How fast an agent must respond. Real-time voice targets sub-500ms; async doesn't care. Mismatching it to the job is the classic error.
Orchestration
The glue running turn-taking, tool calls and state of a voice agent — e.g. LiveKit, Pipecat — between ASR, the LLM and TTS.
Outcome / resolution pricing
Paying per resolved issue or booked meeting, not per seat. Now the de-facto model in support; aligns vendor incentive to result.
