ASAPP

What it is

ASAPP sells an AI agent for the very large contact centre, built by a company that spent its first decade selling AI to the humans in that contact centre. GenerativeAgent handles voice and chat conversations end to end: it reads your knowledge base, calls your back-office APIs, and resolves the contact or hands it to a person. It sits inside ASAPP’s Customer Experience Platform (CXP), launched late 2025, which on 27 April 2026 gained five purpose-built agents that work on the agent itself: Discovery (finds automatable intents in your conversation history), Developer (builds a GenerativeAgent flow from plain-language instructions), Simulation (stress-tests it against edge cases before launch), Insights, and Optimization.

The company was founded in New York in 2014 by Gustavo Sapoznik and has raised roughly $400M according to funding trackers. Named customers are airlines, telcos and cable providers: American Airlines, JetBlue, Dish, Astound Broadband. Katie Stein, previously CEO of Atain and chief strategy officer at Genpact, became CEO on 5 August 2026, succeeding Priya Vijayarajendran, who was CEO at the April CXP launch.

Why it shows up in CX stacks

• Approver mode is the safety story regulated buyers ask for. In Human-in-the-Loop Agent (HILA) approver mode, GenerativeAgent drafts every reply into a queue that sits in your agent desk as an iframe app; the human accepts, edits, transfers or ends each message before the customer sees it. Admins choose which tasks need full supervision. One supervisor approving drafts across several conversations is a staffing model you can run during a sensitive rollout, then relax intent by intent.
• It plugs into the telephony you already run. The documented voice paths are Genesys Cloud (AudioHook, call stays inside your Genesys environment), Amazon Connect (contact-flow hand-off or PSTN transfer), and Twilio, plus a direct API that takes transcripts from your own speech recognition. Chat works beside Salesforce or Zendesk messaging.
• The test loop is built in. The Simulation agent and the continuous red-teaming ASAPP added to CXP in 2026 matter more than the build tool: a large airline cannot learn about prompt-injection or refund-policy edge cases from live customers.

Pricing reality

ASAPP publishes no rate card on its own site, but its AWS Marketplace listing does. On the 1-month contract tab, GenerativeAgent bills $0.70 per digital interaction and $1.00 per voice interaction, where an interaction is one unique contact that closes after 24 hours of inactivity. The listing will not sell fewer than 500,000 interactions, which puts the floor at $350,000 all-digital or $500,000 all-voice at those rates. 12- and 24-month terms are discounted against them; the exact annual rate comes through a private offer. No refunds.

The unit is the thing to model. ASAPP bills per interaction handled, not per resolution, so a contact the agent escalates to a human still costs the full rate. At 50% containment the effective price per resolved contact is $1.40 on chat and $2.00 on voice. Compare that against Intercom Fin at $0.99 per resolution, or against Decagon’s reported platform fee plus roughly $0.50 per resolution, only after you have normalised the unit.

Best for

Heads of customer operations at enterprises handling well over a million contacts a year, voice-heavy, in airlines, telecoms, insurance or financial services, already on Genesys Cloud or Amazon Connect, who need a human approval step before the AI talks to customers unsupervised. It fits best when procurement can buy through AWS Marketplace and draw down an existing AWS commitment.

Skip it if you handle fewer than about 500,000 contacts a year, because the listing minimum alone outruns your volume; if your queue is chat-only and email-heavy; or if you want a self-serve trial, since approver mode cannot even be tried in ASAPP’s Previewer and must be enabled by the account team.

Versus the alternatives

The incumbent share sits with the CCaaS suites, NICE CXone and Genesys Cloud, whose bundled AI agents ride on a contract you already signed. Test those first if your contact volume is mostly simple intents; a second vendor has to beat “already paid for”. Sierra is the fastest-growing entrant and the pick when you want outcome-based billing, so that escalated contacts are not billed. Decagon wins on chat-first, mid-enterprise queues with a lower platform fee. Cresta is the pick if improving human agents is half the plan, and Parloa or PolyAI if the job is voice only. ASAPP’s case is narrow: a published per-interaction price you can buy on AWS, and supervised drafting for the regulated intents.

Watch-outs

• Leadership changed mid-2026. A new CEO in August, three months after a platform relaunch, can move the roadmap and the pricing posture. Guard: put a price cap on renewal, a data-export clause covering transcripts and flow definitions, and termination rights on change of control into the contract.
• Approver mode auto-approves by default. The auto-accept timeout defaults to 30 seconds: if the reviewer is idle, the draft goes to the customer. Guard: for intents that carry regulatory or refund risk, lengthen or disable auto-accept and staff the HILA queue to the volume you route through it.
• Escalations are billed. Every handled contact counts, contained or not. Guard: set a containment target per intent before signing and route only intents that clear it in simulation; low-containment intents should never reach GenerativeAgent.
• NICE and Five9 are not on the documented voice-integration list. Guard: if you run either, ask for a reference customer live on the same CCaaS and budget for the direct-API route before you commit to the 500,000-interaction minimum.