An AI data readiness checklist helps a New Zealand business decide whether its source information is clean, accessible, safe, and governed enough for an AI agent to use. The checklist should test source-of-truth quality, permissions, privacy risk, tool access, approval ownership, logging, and the first workflow to automate.
Data readiness is not a legal compliance guarantee, and it does not mean every system should be connected to AI. It is a practical operating review before you pay for automation. If you need a broader workflow review, start with the AI Automation Assessment NZ guide or the $1,000 AI Agent Assessment.
AI data readiness means your business can clearly name the information an AI agent may use, where it lives, who owns it, how reliable it is, what sensitive data must be protected, and what a person must approve before AI output affects customers, staff, records, publishing, or money.
For a small business, data is ready enough when the source is trusted, the task is narrow, access is limited, output can be reviewed, and the next decision is clear: build now, prepare first, wait, or avoid automation.
AI agents are only as useful as the context, permissions, and review process around them. When the data layer is unclear, automation can move a messy workflow faster without making it safer.
A useful agent should know which CRM field, document, spreadsheet, website page, policy, or decision record is the source of truth. If two sources disagree, it should flag the conflict.
Duplicates, stale records, missing owners, outdated pricing notes, and inconsistent labels make AI-prepared work harder to trust.
Customer details, staff information, finance records, commercial decisions, and private documents should not be exposed casually. Business.govt.nz's safe AI guidance is useful background; the checklist turns those concerns into workflow decisions. If AI will touch personal information, run an AI Privacy Impact Assessment NZ before connecting tools.
Use this checklist before connecting an AI agent to live business systems.
Write the job in one sentence. "Prepare a weekly list of stale sales opportunities for review" is clearer than "use AI in sales".
List the approved sources the agent may use: CRM records, website pages, offer notes, SOPs, spreadsheets, policies, meeting notes, task boards, or operational exports.
Find the quality issues that would damage the workflow: duplicate companies, empty next-action fields, old lead stages, or missing permission notes.
Every important source needs a named owner. Without ownership, the agent inherits the same confusion people already manage manually.
Separate read-only access from draft preparation and live updates. Start with the lowest useful permission, especially around customers, pricing, public copy, finance, and sensitive records.
Mark information that should be minimised, masked, excluded, or handled only in an approved environment. Private business data does not need to train a public model.
Some systems have APIs, permission controls, audit logs, or clean exports. Others rely on messy spreadsheets or copy-paste, which may mean preparation comes first.
Name the person who checks accuracy, tone, offer details, privacy risk, and source use before the output becomes action.
Record the source used, output prepared, human decision, exceptions, and system changes so early pilots can be reviewed and improved.
Build if the data is good enough for a narrow, reviewable pilot. Prepare first if ownership, access, or source quality is missing. Wait or stop if risk outweighs benefit.
Different AI workflows need different source checks. Do not apply one generic readiness score across the whole business.
Check contact records, deal stages, last-touch notes, offer details, opt-out context, and owner fields. AI can prepare research and follow-up drafts, but messages and commitments should stay human-approved. See AI CRM Automation for NZ Sales Teams.
Check whether numbers come from approved systems, spreadsheets, notes, project tools, or dashboards. A reporting agent should show sources, flag missing data, and prepare a decision brief rather than make management decisions.
Check public pages, approved service copy, current pricing, analytics exports, Search Console notes if available, and content standards. Publishing, pricing, proof claims, legal copy, and DNS changes need human approval. For broader patterns, see AI Workflow Automation Examples for NZ Businesses.
Check whether policies, processes, templates, and previous decisions are current. A Strategy or Systems agent is more useful when it can answer from approved documents instead of scattered chat history.
Data readiness does not have to become a six-month cleanup project. Fix the issues that would make the first workflow unreliable or risky.
Clean the fields the agent will read or prepare: company, contact, owner, stage, last interaction, next action, offer, permission notes, and important tags.
Move active offers, policies, SOPs, templates, and knowledge documents into a known location. Archive older versions so outdated material is not treated as current.
Write down what the agent may prepare, what it may update, what a person must approve, and what it must never do alone. The AI Approval Gates for Business Automation guide gives examples.
If the workflow only needs approved website pages and a CRM export, do not connect inboxes, finance systems, private folders, or production settings.
The AI Agent Assessment turns data readiness into a practical build decision. It does not assume automation is the answer.
We review the recurring work, where it stalls, which sources it uses, and what makes the current process hard to delegate.
We identify the documents, systems, exports, fields, credentials, and permission boundaries needed for a controlled first agent.
We define what the agent may prepare, what it may update, what requires review, what must never happen automatically, and who owns each decision.
The roadmap shows whether to build the agent now, prepare the data first, wait for a clearer business case, or avoid automating that workflow. If readiness is strong, the AI Agent Implementation Plan NZ explains the next build sequence.
Know the workflow, source-of-truth records, relevant documents, tool permissions, sensitive data, approval owner, logging method, and review standard.
Yes, if the workflow is narrow and uncertainty is visible. The agent should flag missing, stale, duplicate, or conflicting information instead of pretending the data is clean.
No. Your CRM needs to be good enough for the chosen workflow. Clean the fields the agent relies on first, then keep human approval in place.
Assign a named owner for each source and workflow: the sales lead, operations manager, marketing lead, account manager, or privacy/security owner.
No. Data readiness checks whether the information and access around a workflow are usable and safe enough. An AI agent assessment turns the workflow into an implementation decision.
Do not connect AI to every tool at once. Choose one workflow, check the sources, limit access, define the approval owner, and make a clear build decision.
Use the $1,000 AI Agent Assessment to turn this checklist into a practical now-next-later roadmap: what to build, what to prepare first, what needs approval, and what should not be automated yet.
