Break the budget into work, not features
A quote should separate discovery, implementation, testing, training and ongoing support. Discovery maps the present process and confirms which steps are stable enough to automate. Implementation includes integrations, rules, interfaces and access control. Testing covers malformed inputs, missing data, duplicate events and recovery after a failure. Support covers changes to connected systems and the business rules themselves.
A single approval flow inside one tool can be modest. A process spanning an inbox, a CRM and an accounting system has more failure points even if the screen looks simple. Ask what is included in the price and which changes would require a new scope.
Estimate the cost of exceptions
Count the cases that do not follow the happy path. Does an invoice lack a purchase order? Can two staff members approve the same request? Does a customer submit the same form twice? Every exception needs a rule, a review path or a deliberate decision to leave it manual.
For a pilot, collect a sample of real cases and label each as routine, ambiguous or exceptional. This reveals whether a rules-based workflow is enough or whether staff review must remain central. It also prevents a deceptively cheap quote that ignores the hard cases.
Check payback without invented ROI
Use your own numbers: monthly volume × minutes saved per case × loaded hourly cost. Subtract the time staff will still spend reviewing exceptions, plus the expected monthly running cost. Compare that net monthly benefit with the full setup cost. Add qualitative benefits such as fewer missed handoffs, but do not turn them into a percentage without measurement.
Example: if a team handles 300 requests each month and spends eight minutes copying data on each one, the current copying effort is 40 hours. That is a workload estimate, not a savings promise. A pilot should measure how many minutes are actually removed and how much review time appears elsewhere.
Choose the first workflow
Prioritize a process with frequent, consistent inputs, an owner who can approve rules, and a clear way to measure the result. Defer processes that change weekly or depend on unrecorded judgment. Request a small release with an exception queue, audit trail and handover notes before funding a wider rollout. If AI is being added to existing tools, read our integration guide to understand the additional data and review work.


