The first workflow should be useful enough to matter and contained enough to control.
Choosing only by staff frustration can lead to a minor convenience project. Choosing only by financial value can produce a risky first build that never reaches daily use.
Build a candidate list
Ask each team to name recurring work that creates delay, re-entry, checking, chasing, or avoidable decisions.
Describe each candidate as a complete workflow, not a task. “Prepare a client file for review” is more useful than “summarize documents” because it includes the source records, missing information, review, and destination.
Score six factors
Use a one-to-five score for each factor.
| Factor | What to examine |
|---|---|
| Frequency | How often the workflow runs |
| Burden | Staff time, queue age, rework, and delay |
| Stability | Whether the normal path and rules are understood |
| Data readiness | Whether the source records are accessible and reliable enough |
| Risk | Consequences of error, privacy exposure, or incorrect action |
| Measurability | Whether the business can compare before and after behavior |
Do not treat the total as an automatic decision. Use it to expose disagreements and assumptions.
Look for a strong first candidate
A good first workflow usually has:
- Regular volume
- Visible manual burden
- A clear owner
- A stable normal path
- Identifiable exceptions
- Accessible source data
- Reviewable outputs
- A measurable result
The Federal CIO Council’s robotic process automation white paper describes rules-based automation as a fit for routine, repetitive, high-volume work with structured inputs and stable rules. AI can extend the range of inputs, but it does not remove the need for a defined task and operating context.
What should freight forwarders automate first?
Choose one queue that repeatedly makes your team check, copy, or chase the same information. Common candidates are missing proof of delivery, late shipment updates, and invoices that do not match the agreed rate. Pick the candidate with accessible records, clear rules, and someone responsible for reviewing the result.
A proof-of-delivery preparation workflow can match a document to a load, flag missing fields, and place incomplete cases in a review queue. Staff still decide whether the evidence is acceptable and whether billing can proceed. If service exceptions dominate the day instead, start with that queue. The right first workflow depends on your operation.
Check your existing transport software before adding another tool. CargoWise, for example, describes container event alerts and exception workflows. Use that capability where it meets the need. Scope additional AI around remaining work such as interpreting inconsistent notes or preparing an explanation from conflicting records.
The shipment exception guide compares the options. Try the Load Exception, POD Verification, and Invoice Reconciliation demonstrations to see which review task resembles your own. They use fictional scenarios and keep the next action with a person.
Check whether AI is necessary
Before selecting a model, test simpler answers:
- Remove the step.
- Standardize the input.
- Fix the system configuration.
- Connect two existing tools.
- Add a clear queue or owner.
- Use deterministic rules.
The NIST AI RMF Core includes consideration of viable non-AI alternatives. This protects the business from adding model cost and uncertainty where ordinary software is enough.
Avoid these as a first project
Do not begin with a workflow that:
- Has no accountable owner
- Changes every week
- Depends on undocumented judgment
- Uses inaccessible or unreliable records
- Creates serious harm if the output is wrong
- Requires several departments to agree before testing can begin
- Has no baseline or success measure
These workflows may still deserve attention. They need process, governance, or data work before automation.
Define a narrow pilot
The pilot should test the main uncertainty.
Limit the scope by record type, team, volume, action, or permission. Start with read-only access or draft outputs when possible. Keep a person in control of external messages, financial actions, or sensitive decisions.
Set acceptance criteria before launch:
- Quality threshold
- Maximum review time
- Allowed exception rate
- Cost per completed case
- Required audit record
- Stop conditions
Decide what happens next
After the pilot, choose one of four actions:
- Expand because the workflow met the standard.
- Revise because a specific gap is fixable.
- Keep the workflow limited because the value is narrow.
- Stop because the cost, risk, or review burden is too high.
The first automation should teach the business how to select, test, own, and measure future systems. That operating discipline is more reusable than the first tool.
Use the AI workflow audit checklist to collect the baseline. Then use the workflow automation ROI guide to define the proof.
If your team knows the recurring problem but needs help defining the solution, review our implementation services or describe the workflow. You do not need to choose an AI model or write a technical brief before that conversation.
Questions this guide answers
What should a business automate first?
Start with a frequent, measurable workflow that follows a reasonably stable path, uses accessible records, creates visible staff burden, and can keep risky exceptions under human review.
Should a business automate its most expensive workflow first?
Not always. A costly workflow may also be unstable, sensitive, or full of exceptions. The best first project balances business value with feasibility and controlled risk.
What should not be automated first?
Avoid starting with rare work, unclear ownership, disputed rules, poor source data, sensitive judgment, or a workflow that the business cannot measure.