Is your service agent‑ready? The Agent Readiness Standard
An open standard that scores whether a customer’s AI assistant can get the job done with your service, safely.
Version 1.0 · 18 November 2026 · 5 min read
The short answer
The Agent Readiness Standard (ARS) scores whether a company’s customer service is ready for its customers’ personal AI assistants. A declared AI assistant tests the live service from the outside, the way a customer would send it. The result is a score out of 100 across five groups, and a level from 1 to 3.
Customers are starting to hand errands to AI assistants of their own: track this order, move that booking, get my money back. When one of those assistants contacts a company, two things can go wrong. It gets nowhere: a wall, a loop, an answer it can’t use. Or it gets too far: an order cancelled or account details shared on the assistant’s word alone, without the customer ever confirming. ARS measures both.
Want to know where your own service stands? Get a free ARS snapshot.
Why this standard exists
Benchmarks for AI agents mostly test models on simulated tasks. They show what a model can do in a lab. They don’t show what a customer’s assistant gets from a real company on a normal afternoon, in the customer’s own language.
ARS tests the other side: the service a company runs today, whoever built it, whether AI or people answer the chat. It asks one question: when a customer’s AI assistant contacts us, does it get the job done, safely?
It is built for more than English. The test runs in the language a company’s customers use. Version 1.0 launches with English and Arabic (Modern Standard Arabic), German and French are in review, and more languages follow. The checks on consent, data and disclosure follow principles that data-protection and AI-transparency rules in Europe and the Middle East share: ask before acting, ask only for what you need, and tell people when they are dealing with a machine.
What ARS scores
| Group | Points | The question |
|---|---|---|
| A. Let agents in | 15 | Can an assistant find your service, reach it, read your help, and find a direct route if you publish one? |
| B. Give answers agents can use | 25 | Are the answers right, exact and the same every time, in the customer’s language? |
| C. Get the job done | 25 | Are status checks, changes, cancellations and refunds done or routed, with no loops? |
| D. Protect the customer | 25 | Does the service check it is really the customer, ask them before anything irreversible, and leave the assistant unsteered? |
| E. Be open | 10 | Does it say it is AI, explain where data goes, and never ask for passwords in chat? |
The five groups hold 20 dimensions. Each one, with what scores 0 and what scores 5, is in the specification.
Three levels
| Level | Name | In short |
|---|---|---|
| 1 | Findable | An assistant can reach your help and read it |
| 2 | Usable | A full test scores 60 or more, the answers hold up, and there is no critical finding |
| 3 | Agent-ready | A full test scores 75 or more, jobs get done, the customer is protected, and there is no serious finding on protection or openness |
How a test works
A declared AI assistant, acting for a reserved test customer, works through 11 everyday jobs on the live service: a status check, a change, a cancellation, a refund, an account question and more. It never pretends to be a real customer, never confirms anything irreversible and never asks to be put through to your staff. When the service asks for the customer’s own login or confirmation, the test stops, because that is the right answer.
Every job, judging rule and safeguard is on the method page.
What ARS is not
Not a certification. A result reads “scored against ARS 1.0 on” a stated date. There is no badge.
Not a vendor ranking. ARS scores deployments, not vendors. It never compares the companies that build AI agents.
Not legal advice. A good score does not show that a service meets any law or regulation.
Open by design
Free to use. The text is licensed under CC BY 4.0. The rubric file, ars-1.0.json, is licensed under Apache 2.0 and published in full.
Versioned. Version 1.0 is dated 18 November 2026. Every change is recorded in the changelog.
Open for comment. Comments are open for 60 days, by email to hello@onclarity.com. Version 1.1 follows in late January 2027.
Advisory group. We are inviting five to seven outside members: CX leaders, a data protection officer, a privacy lawyer and a language-technology academic. Names are published once people accept. To take part, write to hello@onclarity.com.
Disputes and re-tests. Any company that is scored can dispute a finding with evidence within 180 days of the test, or ask for a re-test after making fixes.
Where OnClarity stands
OnClarity sells AI agents, so we have a stake in how they are judged. That is why the method is open and anyone can rerun it, why ARS never ranks vendors, and why we will publish our own AI agent’s result, tested with the same method as anyone else’s.
Questions people ask
What is agent readiness?
Agent readiness is how well a company’s customer service works when a customer’s own AI assistant contacts it on their behalf. An agent-ready service lets the assistant in, gives it exact and consistent answers, gets the job done, and asks the customer before anything irreversible happens.
How do you test if customer service is ready for AI assistants?
Send a declared AI assistant through everyday jobs on the live service: a status check, a cancellation, a refund, an account question. Then score what happened. ARS does this across 20 dimensions in five groups, and stops whenever the service rightly asks for the customer’s own login or confirmation.
Is the Agent Readiness Standard a certification?
No. ARS is an open scoring standard, not a certification, and there is no badge. A result says that a service was scored against ARS 1.0 on a given date, with the level, the score and the findings. Any company can rerun the method and check the result.
About
The Agent Readiness Standard is published by OnClarity. OnClarity (www.onclarity.com) is an AI customer experience platform that helps teams resolve, score and understand every customer conversation. The standard is free to use and adapt, with attribution.
Text licensed under CC BY 4.0. Rubric file ars-1.0.json licensed under Apache 2.0.
