Ada
Enterprise AI customer service across voice, chat and messaging, priced per conversation or resolution
Enterprise
Decision snapshot
Fast-read signals for fit, pricing, and trust.
- Category
- Customer Engagement & CRM
- Best for
- customer service chatbot
- Pricing
- Enterprise
What is this tool?
The unit Ada sells is a resolved customer contact, not a seat, and that changes what an evaluation looks like. You need enough volume across enough channels for a percentage point of automation to be worth something, someone inside the business to own the agent, and a figure for what a contact costs you today — because no price, rate or plan is published anywhere, and every route ends in a demo. Security reviewers should expect the same: the trust centre lists its attestations but leaves residency, encryption, retention and model training to the documents behind them.
By QT Desk
Read full overviewHide full overview
What this platform counts is resolutions, not conversation flows. An agent draws on the systems you connect, answers on whichever channel the customer used, and is shaped by playbooks and coaching rather than by decision trees; simulations let a change be rehearsed before it goes live. That is an operating model, not a feature, and it assumes a named person inside the business whose job includes watching how the agent behaves and correcting it. Bought without that person, the platform still works and simply stops improving.
Commercially, nothing is published. The vendor's own pricing URL is a demo-booking page, and the platform page states the shape of the model rather than a rate: you pay for every conversation the agent has with an end user, and a resolution-based model is also offered. Under either, the invoice follows volume — which makes the performance centre a financial instrument as well as an operational one, since the report that tells you what was resolved is the report that explains what you were charged. It also means there is no licence fee to hold a quote up against. The only number that makes a proposal legible is your own: what a contact costs you to handle today, across the channels you would be automating. Work that out before the demo, because you will not be able to judge the answer without it.
On security the vendor publishes documents rather than statements. The trust centre carries a SOC 2 Type 2 report and a SOC 3 report covering October 2023 to September 2024, a HIPAA report dated September 2024, GDPR, and a PCI attestation, with the retention policy and the list of sub-processors available as separate files. What it does not print on the page itself is where data is hosted, how it is encrypted, how long it is kept, or whether customer conversations are used to train models — the section for AI training data exists but is not filled in. For a platform that will read your customers' messages, those four answers belong in the review before the pilot, not after.
Is this right for you?
Quick fit check against how you actually work.
Choose this if
- Your customers arrive through several channels and you want one agent handling all of them
- You are buying an outcome — contacts resolved without a person — rather than seats for your team
- Your support systems are already in place and the agent needs to work inside them
- You have someone who will own the agent: tune playbooks, review coaching and run simulations before releases
- Your procurement will want the attestation reports in hand before a pilot, which the vendor makes available to request
Consider alternatives if
- You need to size the cost before talking to sales, because no price, plan or rate is published anywhere on the vendor's site
- You want to trial it yourself: every published route ends in a demo booking, and no self-serve sign-up or trial is offered
- Your security review needs data residency, encryption detail, a retention period or a position on model training in writing, because the trust centre points at documents rather than stating them
- You are a small team wanting a chat widget on one channel, which is not the shape of this platform
- You need a per-seat cost to compare against your current help desk, since this is billed by conversation or by resolution instead
Key features
What you can actually do with this tool.
One agent across every channel support runs
Voice, email, chat, Messenger, WhatsApp, SMS, Instagram, in-app and custom channels are handled by the same agent rather than by separate bots. An organisation whose customers arrive by four routes maintains one set of behaviour instead of four.
Simulations before changes reach customers
Changes can be rehearsed in simulation before they meet a customer, which is what makes it reasonable to adjust an agent weekly rather than quarterly.
Coaching and playbooks instead of decision trees
The platform is organised around guiding an agent's behaviour — playbooks for how to handle a class of request, coaching for correcting it — rather than around scripting flows. A rules-based bot fails predictably and is repaired by editing a rule; this fails in the way a new colleague does, and is repaired by changing the guidance.
Connected to the systems of record
Zendesk, Salesforce, ServiceNow, Genesys, Amazon Connect, Twilio Flex, Freshworks, Aircall and GitHub are named, with open APIs, SDKs and MCP behind them. Automated resolution depends on the agent being able to read and write in those systems, so this list is the feasibility check, not a nice-to-have.
Measurement built into the platform
A performance centre sits beside the conversation hub, reporting what the agent handled, what it escalated and where it failed. It is the instrument you tune playbooks and coaching against, so the quality of the reporting sets how quickly the agent gets better.
Integrations
Tools and platforms this connects with.
ZendeskSalesforceServiceNowGenesysAmazon ConnectTwilio FlexFreshworksAircallGitHub
Frequently asked
Common questions about this tool, answered.
What decides how much of our contact volume it can actually automate?
What the agent can read and write in your systems, more than anything in the product itself. It resolves contacts rather than deflecting them, so a refund, an address change or an order status only gets closed if the agent can perform that action where the record lives. The vendor names connectors for the common help desks, CRMs and contact-centre platforms and offers open APIs, SDKs and MCP behind them — but the useful exercise before a demo is to take your ten highest-volume contact reasons and ask, for each, which system holds the answer and whether the agent may write to it. That list, not the connector list, is the automation ceiling you are buying against.
What does owning this agent actually involve?
More than configuring it once. The day-to-day is writing and revising playbooks for classes of request, correcting the agent through coaching when it handles one badly, running a simulation before a change reaches customers, and reading the performance centre to find where it is escalating or failing. That is a recurring job rather than a project — closer to managing a new colleague than to maintaining a rules engine, and it does not have an obvious owner in most support organisations until someone is given it. Teams that buy this without naming that person tend to get the automation rate they launched with and no more.