How to Choose an Enterprise AI Customer Service Vendor (2026 Checklist)

Four platforms in this directory have been acquired since 2024. Zendesk alone bought Klaus, Ultimate and Forethought, the last of those closing fifteen days after it was announced.

If you are signing a three-year enterprise contract, that is not industry trivia. It is the most under-weighted variable in your evaluation.

You cannot compare on price, so compare on something else

Decagon, Sierra, Ada, PolyAI, Netomi and the CCaaS suites all price by negotiation. You will not get a number without entering procurement, which means shortlisting on cost is impossible and biases early evaluation toward whoever publishes.

So compare on the things that are knowable: capitalisation, architecture, and what happens when the agent is wrong.

1. Will they still be independent?

Sierra and Decagon are capitalised well enough that few plausible acquirers exist. That is a genuine procurement consideration, not a vanity metric.

2. Procedures or improvisation?

The two most credible independents reach the same architectural answer from different directions. Decagon configures explicit operating procedures. Ada forces the AI through no-code business rules step by step before it resolves anything.

Both are choosing constrained execution over free generation. For anything touching money or compliance that is the correct choice, and a vendor that cannot show you the constraint mechanism is telling you something.

3. Voice, or only chat?

Sierra is the only platform here treating voice and chat as one agent with one configuration. If phone volume is material, that is the single strongest differentiator available and comparing Sierra's rate against chat-only tools is not comparing like with like.

If phone is not in scope, you are paying for breadth you will not use, and Decagon or Ada is the better shape.

4. What does the pricing model do at your volume?

Sierra charges per resolved interaction at a reported $1.50, negotiated. Fin publishes $0.99. Outcome pricing sounds aligned and is, but it also means costs rise as the agent improves. At 10,000 monthly resolutions a $1.50 rate is $15,000 a month, and you are negotiating that against one of the best-funded companies in the sector.

5. Questions for the call

• What event triggers a billable resolution, and what happens when a customer abandons mid-conversation?
• Show me the mechanism that stops the agent improvising outside policy.
• What is your resolution rate confirmed by the customer, not inferred from silence?
• What is the implementation timeline, in weeks, with named dependencies on our side?
• Who owns you in three years?

Funding figures are drawn from press reporting and company announcements, cited below. We have not tested any enterprise platform — they are sales-led and we do not accept vendor-configured environments.

What enterprise agentic actually costs

None of these vendors publish, which makes the reported figures worth collecting in one place — treated as indicative rather than quoted.

The first two are platform fees before usage, which is a different commitment shape from a per-resolution rate. A $50,000 floor means low volume is expensive per unit and the vendor has no incentive to help you stay small.

The implementation line

At this end of the market implementation is a separate quote and it can rival first-year licences. Ask for it explicitly rather than as an estimate, and ask who does the work — vendor, partner, or you.

Vendor-led implementation is usually faster and creates a dependency. Partner-led costs more and leaves you with someone who can change things later. Doing it yourself is cheapest and is how projects stall.

Five clauses worth arguing over

• The billable event definition, in writing, including abandonment and repeat contact within a window.
• Rate protection at renewal, with a cap. A published rate is not a promise about year three.
• What happens on acquisition. Four platforms in this category changed hands since 2024 and one no longer exists standalone.
• Data portability on exit — conversations, configurations and the knowledge you fed it.
• Whether failed containment is billed. Paying for a conversation the agent could not handle is common and negotiable.

On the acquisition clause specifically: Salesforce signed to buy Fin in June 2026, NICE bought Cognigy in 2025, Zendesk bought Forethought in March 2026. If your term is three years, ownership changing during it is closer to likely than not.

The trade you are actually making

You cannot shortlist enterprise agentic vendors on price because none publish one. Shortlist on independence, on whether the architecture constrains the agent to your procedures, and on whether voice is in scope. Sierra if phone matters and budget is not the constraint. Decagon if auditability matters most. Ada if you need channel and language breadth with a no-code owner. And ask every one of them who owns them in three years.

WRITTEN BY AR · UPDATED 2026-07-28

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