What it is
Ada is an enterprise AI customer-service platform — AI agents that hold customer conversations across voice, email, chat, WhatsApp, SMS, Instagram, in-app messaging, and custom channels, then act on the backend systems the conversation is about. Founded in Toronto in 2016 by Mike Murchison and David Hariri, Ada has raised over $200M from Accel, Tiger Global, Bessemer Venture Partners, and Spark Capital, and runs at roughly 705 employees as of March 2026. It reports 550+ AI agents deployed and 6.4 billion interactions handled since 2016, with customers including Square, Pinterest, YETI, Monday.com, and Cebu Pacific.
In February 2026 Ada shipped its Reasoning Engine: one intelligence layer that carries the same knowledge, business logic, policies, and safeguards across every channel, instead of per-channel bot builds that drift apart. The named pieces are knowledge ingestion, API-backed actions, multi-layer safeguards, and Coaching — the feedback loop where CX staff correct agent behavior without a rebuild.
Why it shows up in CX and support stacks
- The bill does not climb as the agent improves. Ada charges per conversation handled, whatever the outcome. Every other enterprise-grade name on the shortlist meters resolutions: Zendesk at $1.50 per automated resolution on committed volume ($2.00 pay-as-you-go), Intercom Fin at $0.99 per outcome. On a resolution meter, raising automation from 25% to 75% triples the invoice on flat volume. On Ada’s meter it does not move. That is a procurement difference, not a positioning one.
- It replaces the automation layer, not the system of record. Ada sits on top of the help desk you already run — Zendesk (Guide, Talk, Support, Chat, Messaging), Salesforce, Freshworks, Genesys, Dixa, Gladly, Gorgias, Help Scout, Kustomer, NICE CXone, Twilio Flex, Amazon Connect, and Aircall. Ticketing, routing, and reporting stay where they are.
- Voice is a first-class channel, not a bolt-on. Ada’s contact-center integrations connect voice conversations to the same reasoning layer serving chat, so a phone-first support org is not buying a chat product and hoping.
- There is an official MCP server. Documented at
docs.ada.cx, it exposes most dashboard operations to Claude, ChatGPT, or any MCP client with OAuth or API-key auth — CSAT trends and automated-resolution drivers queried conversationally instead of through the dashboard.
Pricing reality
Sales-led, nothing published. The model is conversation-based: you pay for each conversation the agent handles, resolved or not. Vendr’s dataset of 112 purchases puts the median buyer at $73,500 per year, with bands around $30K–$60K for starter deployments, $70K–$150K mid-market, and $150K–$300K+ enterprise; first-year totals including implementation run $45K–$100K small and $200K–$400K+ at the top. Multi-year commitments are reported to move price 20–30%.
Ada publishes its own worked comparison: $0.35 per conversation against $1.50 per resolution, on 500,000 annual conversations growing 10% a year while automation climbs from 25% to 75%. Under those assumptions the resolution meter costs more than triple by year three. Run it with your own numbers before you accept the conclusion — the arithmetic only favors Ada when your automation rate is actually rising. If your automated-resolution rate is flat at 25%, per-resolution billing is cheaper, and Ada’s model is the more expensive one.
Best for
Enterprise CX and support leaders running high volume across voice and chat on an existing help desk, who need an AI-support line item finance can forecast a year out and whose automation rate they expect to push upward. It is the right pick specifically when predictable spend and multi-channel consistency matter more than paying only for outcomes.
Skip it if your volume cannot pay back a $45K first year, if your support is chat-only help-center deflection where a lighter tool wins on cost, or if your automation rate is stuck and a resolution meter would bill you less.
Versus the alternatives
The incumbents are the platforms you already pay for. Zendesk and Intercom Fin have the installed base; choose them when the AI attached to your existing help desk clears your bar and adding a vendor is not worth the deployment. The fastest-growing entrant is Decagon — $250M Series D in January 2026 at a $4.5B valuation, 100+ new enterprise customers signed in 2025 — and it wins when you want the aggressive deflection curve and will accept outcome-based billing to get it. Sierra is the pick when the agent must execute transactions inside PCI-scope flows. Ada’s case against all three is the same one: it is the only one of the four whose invoice is decoupled from how well the agent performs.
Watch-outs
- Per-conversation billing charges you for the failures too. A conversation the agent botches and escalates to a human costs the same as one it handles cleanly, so a bad quarter is billed at full price. Guard: negotiate a floor on automated-resolution rate with a credit or renegotiation trigger attached, and instrument escalation rate from week one so you can prove the number.
- The 84% automated-resolution rate is Ada’s figure across its own deployments. It is a ceiling from a favorable sample, not a baseline you inherit. Guard: pilot one high-volume journey, measure containment and reopen rate against your current tooling for a full quarter, and set the contract’s volume commitment off your measured rate rather than the marketing one.
- The Reasoning Engine is five months old. Anything built on the previous per-channel model is a migration, and behavior tuned in the old builder does not carry over untouched. Guard: ask for the migration path in writing before signing, keep the legacy flows running in parallel through the first peak-volume period, and use Coaching to close behavior gaps rather than rebuilding under deadline.