Choose the first automation by measuring repetitive work, exception handling and the cost of mistakes. A strong candidate has a clear trigger, stable inputs, an accountable owner and an outcome you can verify. High transaction volume alone is not enough: an unclear process can become more difficult to repair when it runs automatically.
Let’s plan your first automation workflow.
Bring your workflow, constraints and open questions. We’ll help you define the scope and work through the next decision.
Observe the complete process before selecting a tool
Follow several authorised examples from request to completion. Record who enters information, who checks it, where it waits and how unusual cases are resolved. Include the informal spreadsheet or phone call that makes the process work today.
Separate time spent doing repetitive work from time spent waiting for a policy decision. Automation may reduce copying without changing an approval bottleneck. That can still be useful, but the expected outcome should reflect the real constraint.
Compare candidates with a baseline worksheet
| Factor | Evidence to collect | What it tells you |
|---|---|---|
| Frequency | Cases per period and peak pattern | Potential reach and required capacity |
| Handling effort | Active minutes across roles | Work that might be reduced |
| Exceptions | Types, frequency and resolution steps | Human review and recovery scope |
| Input quality | Missing fields and disputed records | Preparation needed before automation |
| Error impact | Cost of a wrong or repeated action | Required controls and approval |
| Verifiability | Authoritative completion record | Whether the result can be trusted |
Use the worksheet for comparison, not as a universal scoring formula. A low-volume but high-risk process may need controls urgently while being unsuitable for unattended execution.
Model a small, testable first release
Consider a hypothetical purchase-request workflow. The first release could validate required fields, route a request to the correct approver and remind an owner about a pending decision. It need not automatically approve spending or create a payment.
Define the state transitions: submitted, awaiting clarification, approved, rejected and cancelled. Decide who can move a request and what evidence is retained. Specify duplicate submission, absent approver and expired request behaviour before connecting the workflow to downstream systems.
Measure improvement without hiding transferred work
Compare the same type of case over comparable periods. Track completion time, human handling effort, exception workload and correction rate. If one team saves time but another spends longer clearing exceptions, include both effects.
For an illustrative internal calculation, value hours actually removed from the workflow, then subtract platform, support and review effort. Do not present that model as guaranteed cash savings: staffing costs may remain unchanged even when useful capacity is released.
Automation questions
Does business process automation require AI?
No. Explicit rules can handle structured routing and validation. Use AI only where the task needs it and where its output can be evaluated and reviewed appropriately.
Should exceptions be automated first?
Usually start by understanding them. Some exceptions can be classified with clear rules; others require a person with authority to decide.
Use the exception queue guide to define that boundary. Bring a measured workflow to Drema's process automation team to discuss a bounded first release.
