Draw the path of one request. A call may read information from a diary, create a task in a work queue and send a confirmation. Mark which system owns each piece of information and who can update it.
Start with the smallest access needed. A workflow that only checks availability can begin with read access. Add an update only where you have decided what to change and how to handle failures or duplicates.
Try an unavailable system and a record that has changed since the initial request. Decide whether to retry, hand work to a person or leave the original record intact. Give the customer a clear status in each case.
Find the record that controls each step
Take one example and write down which systems a colleague opens. To move a booking they may read a diary, check a policy document, update a customer record and send a confirmation. Identify which system holds the authoritative time, which source owns the rules and where a person can tell whether the change was made. Make those roles explicit before discussing connections.
Give each system an owner who can explain access and limitations. A system may show bookings but have no practical way to reserve a slot automatically. Decide what the workflow can do with current access and agree how to hand off the parts that need a person. Start with the path the team can actually support.
Plan what happens when a write fails
Reading information and making updates have different consequences. Start with the minimum access needed to answer the request. If the workflow later writes to a system, list the fields it can change and what counts as success. An accepted request in one place does not prove the second system recorded the update.
Decide how to detect retries and duplicated requests. When someone calls twice about the same appointment, link the work to its reference where you can. If an update times out, check the current record before trying again. Give the support team a clear way to see an uncertain state and choose whether to retry or recover manually.
Compare what customers and staff see
Walk through the end of the process from both sides. A customer should see the correct details in their confirmation while staff can find the matching record and the follow-up status. Include a failure case where one system changes and another does not. Decide who owns the reconciliation and what the customer should be told while it is pending.
Write the result in a simple workflow map. For every read or write, give the source, condition, expected result and fallback. Review that map with the people who own the systems and the people who do the work. The map should give everyone a way to spot when a new field, permission or decision needs agreement.
Use the first incident to improve the map
When the first record fails to update, review it with the system owner. Check which component reported success, where the first discrepancy was visible and what a colleague needed to correct it. Document that recovery in the same map as the routine workflow. The next person supporting the connection should be able to reproduce the check without needing to know who originally built it.