Start with a real request

Pick one recent approval and follow it from the first question to the final decision. Ask the requester and the approver to show what actually happened: the form, the email, the spreadsheet and the conversation that never made it into the record. A tidy policy diagram rarely tells the whole story.

Consider a studio buying 24 desk trays. The purchase is within the team’s budget, but only 20 are in stock. The useful question is not simply whether the system can approve the order. It is who can accept a split delivery, what they need to know and how that decision reaches purchasing.

Who requests, and what are they asking for?

Identify who can start a request and on whose behalf. A colleague raising a purchase for another team needs the right cost centre and delivery contact. Make those responsibilities explicit rather than expecting the approver to reconstruct them.

Write down the minimum information needed for a decision: item, quantity, cost, reason, delivery date and budget owner. Mark which values come from a trusted system and which a person must supply. Validate missing information at the point of entry, with a clear way to correct it.

Who decides, and where does their authority stop?

Name a role for each decision, not just the person doing it today. Define value thresholds, delegated cover and situations that need a second review. A request should not disappear because someone is on leave.

In the tray example, the budget owner can approve the spend, but the project lead must accept the later delivery of four items. Those are different decisions. Recording both prevents a financial approval from being mistaken for permission to change the delivery plan.

What does the decision-maker need to see?

Put the relevant information beside the decision: the request, budget context, stock position, supporting document and previous decisions. Show when the information was last checked. A stock figure from yesterday may be less useful than a clear warning that a live check is unavailable.

Decide what must be captured when approval is given: the decision, the person, the time, any conditions and the version of the request they reviewed. If the quantity or price changes afterwards, define whether the approval is still valid.

Map the exception before the happy path is finished

List missing data, rejected requests, conflicting records, unavailable approvers and failed connections. Give each exception an owner and an observable state. “Needs information” is useful only if someone knows what is missing and who will provide it.

For the stock mismatch, pause the order and ask the project lead to choose a split delivery, a substitute or a later date. Keep the answer with the order. When the customer accepts 20 now and four later, the invoice should reflect the agreed shipment rather than silently charging for all 24.

Decide what to automate first

Automate a rule when its inputs are reliable, its owner can explain it and the exception path is understood. Start with checks for completeness, routing and reminders. Keep judgement with a person when context changes the answer.

Before building, test your map with three examples: an ordinary request, a rejection and a messy exception. Ask a colleague who was not in the workshop to follow each one. If they cannot tell what happens next, refine the process first.

Take this into your next workshop

Bring one completed request and one that stalled. Agree the requester, decision roles, required data, thresholds, escalation route and final record. Measure waiting time and rework before the change so you can tell whether the new workflow helped.

The result should be a shared map that the team recognises, with enough detail to design a useful system and enough flexibility to handle the work as it really happens.