Is your organisation ready for AI agents?

21 questions across 7 layers: data, technology foundation, knowledge, trust & governance, processes, people, customer. Result straight away, no sign-up, no e-mail required.

Scale 1–5: 1 = the description on the left, 5 = the one on the right. If you do not know, choose “Don’t know”.

Data

Is there a data catalogue or glossary, and can an agent use it, e.g. via MCP?

1 — No catalogue and no glossary

5 — An up-to-date catalogue or glossary that agents can reach through an interface (e.g. MCP)

Where does the data the agent reads come from, and how fresh is it?

1 — Nobody knows where it comes from or when it was last updated

5 — Known source and lineage; freshness is measured and fits the process

Do access rules cover agents: identity and least privilege?

1 — Agents run on people’s accounts or shared keys

5 — Every agent has its own identity and the least privilege it needs, reviewed

Technology foundation

Can agents connect to the systems they need (APIs, integrations, MCP) without moving data by hand?

1 — The agent gets data copied by hand or from one-off exports

5 — Documented, maintained integrations with every system the agent needs

Is agent activity logged and monitored: calls, costs, errors?

1 — No logs; we learn about errors from users

5 — Logs, costs and errors collected in one place, with alerts and an owner

Are new agents built on a shared platform rather than every pilot from scratch?

1 — Every pilot on a different tool, built from scratch

5 — A shared platform and reference architecture; a new agent reuses ready parts and moves from pilot to production along a set path

Knowledge

Do key terms and metrics have a single definition and an owner?

1 — Each department counts its own way; nobody owns the definitions

5 — One definition of every key metric, with a named owner

Are the procedures the agent is to carry out written down and current?

1 — Procedures live in people’s heads

5 — Written down, current, with an owner and a review date

Who approves a change of definition, and how do we detect contradictions?

1 — Nobody; contradictions surface by chance

5 — An explicit change-approval process and regular contradiction checks

Trust & governance

Decision rights

1 — Implicit, decided on the fly

5 — Written down and reviewed quarterly

Review process

1 — Ad hoc, depends on individuals

5 — A separate, scheduled process with metrics

Organisational structure

1 — No owner of agents

5 — An explicit operating model and owners

Processes

Is the process the agent works in described end to end, with the agent’s and people’s steps?

1 — The process is not described; the agent is to “help” without defined steps

5 — An as-is and to-be map with the agent’s steps, hand-offs to people and a process owner

Do we know how the process outcome is measured before and after the agent goes live?

1 — No baseline; impact judged by gut feeling

5 — Baseline measured before go-live, measured regularly after, with a target

Do we build repeatable steps as reusable parts (e.g. a library of agent skills)?

1 — Each process automated separately, from scratch

5 — A maintained library that further processes reuse

People

Workflow design

1 — Agents bolted onto the old process

5 — The process redesigned around the human–agent division of work

Role clarity

1 — Roles did not change after go-live

5 — New roles described and communicated

Training focus

1 — Operating the tool

5 — Judging output, orchestration and judgement

Customer

Do we know what value agents should create for customers, not just internal savings?

1 — The only goal is internal cost savings

5 — Written customer-value hypotheses, with a metric and a plan to test them

Does the customer know when an agent is serving them, and can they easily reach a person?

1 — The customer does not know it is an agent; no way to reach a person

5 — Clear notice about the agent (where applicable, also the AI Act Art. 50 obligation) and an easy hand-off to a person

Do customer feedback and complaints about agents reach the team that develops them?

1 — We do not collect them; complaints get lost in general support

5 — Feedback and complaints about agents collected separately, analysed and turned into changes

0 of 21 answered

The result is indicative and is not legal advice. Attestia is a compliance automation tool.