Field note / Automation
What should an owner-led business automate first?
Automate a repeated, high-cost workflow first when its inputs are stable enough to describe, a person owns the exceptions, and the result leaves a visible receipt. The best first automation is rarely the most impressive demonstration.
/
The decision in one line
Choose a workflow that matters, repeats, can be bounded, and will fail visibly.
A first build should create leverage and teach the business how to operate automation responsibly before it receives broader scope.
Begin with the cost of friction
Look for time, delay, lost opportunity, inconsistency, or avoidable rework.
A task can be irritating without being important. Estimate how often it occurs, who touches it, what waits while it happens, and what a miss costs the business or customer.
The best candidate does not need a perfect financial model. It needs a plausible operating consequence large enough to justify the attention required to map, build, and maintain it.
Test the shape of the work
Repetition helps only when the pattern is stable enough to describe.
A workflow with a recognizable trigger, consistent inputs, a finite set of actions, and a clear finish is easier to automate safely. Exceptions do not disqualify it, but they need a path.
If the work changes every time or depends almost entirely on tacit judgment, begin with better intake, knowledge retrieval, or preparation rather than automatic action. That often leads toward private AI or a focused knowledge surface first.
- A trigger the system can identify
- Inputs with a defined source of truth
- A named person for review or exceptions
- A receipt that shows the work completed
Preserve human authority
Separate preparation from decisions and decisions from external action.
AI can read, classify, summarize, or prepare a draft inside a workflow. Those capabilities do not automatically justify permission to send, approve, purchase, file, delete, or change a customer record.
A sound first automation makes those boundaries explicit. The system pauses for review where authority, ambiguity, privacy, or relationship requires a person. ShepBuild’s automation and integrations work starts from that same control model.
Make failure observable
A silent automation is operational debt.
The team should be able to see whether work is waiting, running, complete, or failed. Errors need a responsible destination, and important actions need enough context to reconstruct what happened.
Start with one complete flow, operate it long enough to learn, and expand only when the evidence supports the next step.
Original decision tool
Score the workflow before you automate it.
The ShepBuild first-automation scorecard is a diagnostic, not a client result. Give one point for each condition below, then write down the current baseline before changing the workflow. A high score identifies a testable candidate; it does not prove return on investment.
Use one real owner-led workflow, such as missed-inquiry routing, quote follow-up, or a recurring document handoff. The scorecard follows the same practical sequence described in Microsoft's process-automation planning guidance: understand the current process and business value before choosing automation. For AI-assisted steps, the NIST AI Risk Management Framework provides a primary-source structure for governing, mapping, measuring, and managing risk.
- Consequence: a miss creates meaningful delay, cost, rework, or lost opportunity
- Repetition: the workflow occurs often enough to measure before and after
- Boundary: triggers, inputs, actions, and exceptions can be stated clearly
- Owner: a named person handles review, ambiguity, and failure
- Receipt: the team can see waiting, running, complete, and failed states
Useful answers before a tool is chosen.
01Should the first automation be the most impressive demo?
Usually not. The best first automation is a repeated workflow with meaningful cost, a stable enough pattern, a clear owner, and a result the business can observe.
02What if the work depends heavily on judgment?
Begin with better intake, knowledge retrieval, or preparation rather than automatic action. Automation should prepare and route; authority should remain explicit. Compare that approach with agents, automation, and knowledge.
03How do we know the automation is working?
Make failure observable. The team should see whether work is waiting, running, complete, or failed, and important actions should leave enough context to reconstruct what happened.
One problem. One useful first build.
Bring one repeated workflow and test it against the four-part filter.
The AI Build Session maps value, repetition, stability, and visibility into a first-build decision.
Start a conversation