Gorgias Publishes 60 Days. Intercom Says Under an Hour and Documents 21 Steps.

Gorgias sells the fastest, simplest AI agent in this directory. One mandatory ecommerce integration, no custom development, a human implementation manager on two to five calls.

On its own onboarding page it publishes this:

“Average time to launch is 60 days, even for brands with multiple stores and complex setups.”— gorgias.com/onboarding, read 21 August 2026

Sixty days is the honest number for the easy case. It is also the only number of its kind anybody in this category publishes.

What everyone else says instead

Intercom says Fin sets up in under an hour. Zendesk says you can begin automating in just minutes. Both are true about one step and neither is true about a deployment.

The gap is not marketing dishonesty so much as a category error. The vendors are timing the configuration screen. Buyers are timing the project, which includes the content work before it and the testing after it, and those are where the weeks go.

Documented steps against claimed time

Deploying Fin over both chat and email is 21 documented steps by Intercom's own numbering, not one hour. Zendesk's step one is optimise your help centre content, which is a content project, and step two links out to five separate channel-setup guides. The minutes describe step three.

Zendesk's own testing guide is the honest document

The same vendor that says minutes on the setup page says this on the testing page:

“We recommend connecting to your CRM and embedding the widget onto a sandbox, staging, or testing environment of your website or help center.”— Zendesk, testing your AI agent, updated 28 May 2026

And on who should do the testing: ideally not the same people who built the flows, because they will be biased, so two types of stakeholder, process-focused and experience-focused. It also describes the whole exercise as your onboarding journey, ending when the agent is ready to graduate.

A staging environment, a CRM connection and two categories of non-builder testers is a cross-functional project. It is a good recommendation. It is not minutes, and it is on a different page from the one that says minutes.

Ada's documentation admits most customers never see the setup flow

This is the single most useful sentence we read all week, and it is buried in a getting-started page:

“Most people who obtain access to their AI Agent don't go through this setup process. Ada's Customer Experience (CX) team usually configures AI Agents in advance, adding Knowledge (at minimum), and sometimes also customizing the Persona.”— docs.ada.cx, read 21 August 2026

There is no self-serve door either: your Ada representative provides a link to your agent. So there is no clock to quote because you are not the one on it. The engineering is stated plainly elsewhere — external engineering teams often build or adapt APIs so Actions can retrieve data, update records or initiate workflows — with a real integration spec behind it, including authorisation URLs, token URLs, PKCE and refresh tokens for per-user access.

That is a backend workstream that has to finish before the agent can take a single action. It is not in anybody's time-to-value claim.

Two vendors put their documentation behind a login

docs.sierra.ai serves a login form. docs.decagon.ai redirects to one. Neither publishes a pricing page: both /pricing URLs return 404.

Sierra's login wall states the terms on which the documentation exists:

“All Sierra Products are confidential, and you will not use or disclose them other than to evaluate the Sierra Products.”— docs.sierra.ai, read 21 August 2026

So for these two, every column in the table above is blank because they publish nothing to fill it. That is not an oversight, it is a sales model, and it means you cannot scope the work before entering a sales process. Decagon does settle the engineering question by accident on a public product page, which describes integrating with CI/CD practices to test every agent version and triggering alerts to tools like PagerDuty. A product whose testing story is a CI pipeline and an on-call rotation is not a support-ops-only deployment.

The part nobody warns you about: testing is billable

You will run test conversations before launch. At three of these vendors that costs money, and only one of them makes it obvious.

Salesforce is the only vendor in this category that publishes, numerically, what testing costs. Its Flex Credits rate card prices a standard action at 20 credits in production and 16 in a sandbox, with voice actions at 30 and 24, and states that the sandbox multiplier applies to pre-production environments including scratch orgs. At $500 per 100,000 credits, a production action is ten cents and a sandbox action is eight. Testing runs at 80% of production rates.

Those credits also expire. They must be used before the order end date and no rollover is permitted, so an unused testing budget is not money you keep.

Intercom's rule is simpler and stated on both deployment pages: preview is free, and if you test Fin in your live environment you are billed per outcome — though no outcome counts if the person asked for a human.

Salesforce is nine objects for one chat window

Salesforce's own developer workshop opens by admitting it: there are quite a few pieces that need to come together to expose an agent. To put one agent on one web chat you create messaging settings, a routing configuration, a queue, a presence status, a permission set and its assignment, a presence configuration, a published Experience Cloud site to get the domain, a messaging channel, and then an embedded service deployment.

That is exercise two of four, in a workshop whose data track alone runs eleven exercises and whose agent track requires Apex and Flow actions. Anyone comparing this to Gorgias on features is comparing two products that do not have the same unit of work.

What to ask before you sign

• How many customers of our size launched in the last quarter, and what was the median elapsed time from contract to first live customer conversation?
• Which steps require an engineer, and how many engineer-days did that take at a comparable customer?
• Do test conversations consume our allowance, in production and in a sandbox? Get the answer in writing — two vendors here do not publish it.
• Is there a staging environment where the agent's configuration can be changed without touching live, or do we test against the published config?
• What does your onboarding contract oblige you to do, and what does it cost? Intercom's pricing page says no setup fees while its deployment services terms define paid Premier Onboarding with the amount unpublished.

One more, because it is the cheapest way to calibrate everything else: ask them to send the setup documentation before the call. Whether it arrives, and whether it needs a login, tells you most of what the answers above will say.

Step counts are the vendors' own numbering, not ours. Time claims are quoted from the vendor page that makes them. Where a field says not published, we looked and did not find it, rather than not looking.

WRITTEN BY AR · UPDATED 2026-08-21

I read the fine print. Vendor pricing pages, billing definitions, terms, funding filings and acquisition notices — then I do the arithmetic nobody publishes: what a platform actually costs at your volume, what its headline metric is really counting, and who owns it now. I do not run benchmarks, and no page here pretends otherwise.

Editorial policy