The hidden problem under AI projects

Many AI projects start with a tool question: which platform, model or automation should the business use? But the earlier question is often simpler: can the business explain the workflow the tool is supposed to support?

What to map first

  • Trigger: what starts the workflow?
  • Source: which files, records or customer details are authoritative?
  • Owner: who is accountable for the step?
  • Approval: who must review before action?
  • Risk: what data, commitment or professional boundary is involved?
  • Evidence: what needs to be kept for future review?

Why handoffs matter

The risky parts of a process are often the gaps between people and systems. A quote moves from email to CRM, a file moves from Drive to SharePoint, or a customer request becomes a task without a clear owner. AI cannot safely fix unclear handoffs by guessing.

What comes after mapping

Once the process is mapped, the company can decide whether AI should draft, summarise, classify, route, flag risk or prepare evidence. Each use case can then be tied to a human approval gate and a stop path.

A simple process mapping template

A useful first map does not need to be complicated. It should make the operating reality visible enough for leadership, staff and IT/MSP partners to challenge. Start with one workflow that matters commercially and is repeated often.

  • Workflow name: the plain-language name staff already use.
  • Trigger: the email, customer request, job status, renewal date or internal decision that starts work.
  • Inputs: files, forms, CRM notes, accounting records, customer history or staff knowledge used.
  • Owner: the person accountable for the next action and final outcome.
  • Systems: the tools touched during the work, including shared drives and spreadsheets.
  • Approval: who must review before customer communication, live-system change or commitment.
  • Risk: sensitive data, professional boundaries, financial exposure or reputational concern.
  • Evidence: what should be saved so the decision can be reconstructed later.

Example: customer follow-up before AI

A customer follow-up workflow may begin with an enquiry, quote request or missed response. The source material may be split across an inbox, CRM note, spreadsheet and shared folder. Before AI drafts a reply or prioritises follow-up, the company needs to know which source is authoritative, who owns the response, what claims are allowed and whether pricing, legal or privacy sensitivity is involved.

Once that is mapped, AI might assist by summarising the enquiry, classifying urgency, drafting an internal next-action note or preparing evidence for review. It should not automatically send a customer email or change a live system unless the owner has approved that specific action path.

Process mapping checklist

  • Can a new manager understand the workflow without asking the owner from memory?
  • Can staff find the source files and know which version is current?
  • Can IT identify the systems and access boundaries involved?
  • Can leadership see where decisions, approvals and exceptions happen?
  • Can professional advisers identify where their review may be needed?
  • Can AI assistance be limited to a narrow task with human review?

What not to do

Do not map only the ideal process from a policy document. Map what actually happens: the shortcuts, duplicated spreadsheets, missing handoffs and decisions that live in inboxes or memory. Do not let AI hide those gaps by producing polished drafts on top of weak operating structure. Do not treat a process map as final; it should be updated when exceptions reveal new risks or better ways of working.

Process mapping before AI is not bureaucracy. It is the control layer that lets a company decide where AI can safely assist and where people, advisers or IT providers must stay in charge.

For leadership readiness, read AI readiness for CEOs and General Managers. For the adoption signal behind workflow readiness, read the AI adoption gap guide. To turn a process map into a scoped engagement, review the Owner-led AI Readiness & Operating Discovery Audit.

A 90-minute mapping workshop

A practical workshop can start with one repeated workflow and three people who see it from different angles: a decision owner, a staff member who handles the work and a systems or IT contact. Spend the first 20 minutes listing the real steps, not the policy version. Spend the next 20 minutes marking files, systems and handoffs. Spend 20 minutes naming decisions, approvals and risks. Use the final 30 minutes to decide whether AI could assist one narrow task without creating external action.

The output should be simple enough to use the next day: a workflow name, owner, source list, approval point, risk note and AI candidate. If the group cannot agree on those fields, the company has found the real readiness issue. That is useful evidence, not failure.

Why evidence matters

AI-assisted work becomes much easier to review when the business keeps evidence of sources, approvals and changes. Evidence does not need to be heavy. It can be a decision log, source register, approval note or weekly review. What matters is that leadership can reconstruct why an output was used and what changed after exceptions.

Keep the map alive

A process map loses value if it is treated as a one-off diagram. Keep it alive by updating it when a handoff fails, when a source changes, when an approval owner moves or when AI output exposes a repeated ambiguity. The map should become a practical operating memory for the business, not a document that sits beside the work.