“We are ready for AI” often means leadership watched a demo. Readiness, in an operational sense, means you can name process owners, produce sample records, list forbidden automated actions and grant sandbox access without three weeks of committee. This checklist helps internal teams and external partners start discovery on facts rather than assumptions. It complements technical work in Connecting AI Automation With Existing Business Software.
Aligning leadership expectations
Executives sometimes expect step-change cost reduction; operators hope for fewer late nights. Write shared expectations: automation should make performance visible and remove repeatable drag, not eliminate judgement or hide understaffing. Misaligned expectations destroy programmes even when technical delivery succeeds.
1. Clarify strategic intent (without buzzwords)
Write one paragraph answering: which customer or staff experience should improve, and how will we notice in eight weeks? Avoid goals like “adopt AI” without operational anchors. Link intent to a primary workflow, enquiry handling, document intake, operational reporting, not to every department at once.
2. Inventory systems and data flows
List systems of record, channels, and shadow tools (spreadsheets, personal inboxes). Note where data duplicates and which source wins conflicts. Identify whether vendors allow API or export access under your contracts. Gaps here predict integration risk more reliably than model selection.
3. Assign process ownership
Each automated workflow needs a named owner responsible for policy updates, approval backlog and quarterly review. IT may build it; operations must own behaviour. Without ownership, prompts and rules drift the moment the builder leaves.
4. Document the current-state workflow
Sketch steps from trigger to outcome, including exceptions handled manually today. Compare with desired state minimally, remove handoffs, not accountability. If current state is disputed between departments, resolve that before automation funding. Related: Common Automation Mistakes.
5. Define approval and forbidden actions
Produce a threshold table: what may run unattended, what requires review, what must never automate. Finance, legal and customer leads sign it. Details in Human Approval in AI Workflows.
6. Improve entry-point data quality
Validate required fields on forms; merge duplicate CRM records you already know about; standardise picklists agents will rely on. AI magnifies messy catalogues into confident-sounding errors.
7. Establish metrics and baselines
Before launch, measure time-to-acknowledge, routing accuracy sample, backlog age or rework rate, whatever matches your intent. Post-launch dashboards should show the same numbers, not new vanity metrics.
8. Plan security and access
Service accounts, least privilege, credential rotation and logging retention should be agreed with IT or competent advisors. Discuss subprocessors and data regions with providers; Miguelito AI does not claim blanket GDPR, ISO or SOC certification on this website, project agreements define specifics.
9. Prepare staff and customers
Staff need training on overrides and exception paths, not only happy demos. Customers benefit from honest acknowledgement templates about when they will hear from a person. Change management is not marketing fluff; bypassed tools return shadow processes.
10. Choose pilot scope deliberately
Use scoring from How to Identify Processes Worth Automating to pick one pilot with manageable risk and visible benefit. Define exit criteria: what would cause you to pause rollout?
Readiness self-assessment table
| Question | Not ready | Ready enough to pilot |
|---|---|---|
| Process owner named? | No accountable role | Named with deputy |
| Sample data available? | Cannot export examples | Anonymised samples in sandbox |
| Approval policy written? | “Manager decides” | Signed threshold table |
| Baseline metrics captured? | None | Two weeks of data |
| Integration path verified? | Assumed from marketing | API or export tested |
Engaging an implementation partner
Bring the checklist outputs to consultation. Useful partners ask about queues and permissions before quoting timelines. Explore Custom AI Business Platforms when SaaS cannot fit; explore Custom AI Agents when language variability dominates. Definition of terms: What AI Business Automation Actually Means.
After go-live
Schedule day-7 and day-30 reviews: sampling outputs, approval backlog, error categories. Expand scope only when metrics improve or stabilise within agreed tolerance. SMEs should guard capacity. See AI Automation for Small Businesses.
Documentation deliverables to request
From internal teams or partners, expect: workflow diagram with systems; data map; approval table; runbook for common failures; evaluation set locations for any agents; and credential rotation instructions. Store them where the process owner, not only IT, can access during leave.
Procurement and legal touchpoints
Involve procurement early when new subprocessors process personal data. NDAs and data processing terms should precede sharing production exports in discovery. Miguelito AI scopes legal specifics per project; this article does not substitute for your counsel.
Red flags that delay discovery
- No executive sponsor willing to resolve cross-department disputes.
- Refusal to provide any sample records, even anonymised.
- Demand for fully autonomous customer commitments on day one.
- Simultaneous ERP migration and AI rollout without phased plan.
Executive briefing pack (one slide each)
Sponsors benefit from a tight pack: intent and metric; chosen workflow; approval summary; systems touched; timeline with named internal time commitment; risks and deferrals. Avoid technology diagrams without ownership labels, boards approve outcomes and accountability, not middleware brands.
Refresh the pack when scope changes so finance and operations stay aligned on what was promised versus what shipped.
Training plan outline
Train three cohorts: approvers (queues and accountability), frontline staff (what changed in daily tools), and sponsors (metrics and escalation). Length beats hype, forty-five minutes hands-on beats a two-hour slide deck. Record short screen captures for hires who join mid-year.
Data protection impact assessment (proportionate)
Even SMEs benefit from a short note: what personal data flows through automation, lawful basis discussion with competent advice where needed, retention periods, and subprocessors involved. Miguelito AI discusses minimisation in discovery; your organisation remains controller for customer data in typical B2B setups.
Update the note when scope expands, especially if agents begin processing attachments that may contain special-category data inadvertently.
Vendor access requests
Prepare a standard access packet: NDA status, sandbox accounts, IP allowlisting needs, and emergency contact. Repeated ad hoc requests delay discovery and frustrate IT, especially in SMEs without a ticket system.
Readiness is iterative: each discovery cycle should leave your data map, approval table and metrics clearer, even when a phase ends in “not yet” rather than a build contract.
Boards and investors sometimes press for AI visibility; a prepared organisation can respond with a phased plan tied to measurable operations, not a rushed chatbot that bypasses records and approvals.
Share the readiness checklist with your implementation partner early; gaps discovered in week one of discovery should not surprise executives in week six of build.
Assign one person to maintain the checklist after go-live, readiness is not a pre-project document alone but a living record of owners, systems and deferrals.
Include customer support and finance in readiness sign-off even when the pilot is “technical”; they encounter automation outcomes first when something misfires.
Limitations
Checklists cannot replace sector-specific legal advice or union consultation where applicable. Remote delivery from Miguelito AI in Paphos, Cyprus suits organisations comfortable with collaborative video discovery and documented runbooks.
When your internal checklist is partially complete and you want help closing gaps, request a business consultation. We will suggest preparation work if discovery would otherwise waste your time and ours.