What Should You Automate First? A Practical Scoring Framework

Choosing an automation target by frustration alone creates a weak business case. The loudest task in your operation may happen twice per month. A quiet task may consume 80 staff hours, delay customer responses, and introduce costly errors every week.

Start with workflow economics. Then check whether your systems, data, and risk controls can support automation.

Define one unit of work

Avoid targets such as “automate customer service” or “use AI for sales.” They cover too many decisions, systems, and failure modes.

Name a specific unit of work instead:

• Qualify a website enquiry and assign it to a salesperson
• Read an invoice and enter approved fields into accounting software
• Follow up a missed call within two minutes
• Classify a support email and draft a response
• Send appointment reminders and process rescheduling requests

A useful workflow definition includes a trigger, required inputs, decision steps, output, owner, and exception path. If your team cannot describe those parts, map the process before evaluating technology.

Measure the baseline

Collect enough data to understand current performance. One representative month usually gives you a useful starting point, though seasonal workflows need a longer window.

Record:

Measure	Question
Volume	How many times does this workflow run each week or month?
Labour	How many staff minutes does one run consume?
Delay	How long does work wait before someone handles it?
Error rate	How often does staff correct, repeat, or escalate it?
Error cost	What does one error cost in labour, refunds, churn, or risk?
Revenue effect	Does speed or consistency affect conversion, retention, or capacity?

Use measured values where possible. If your systems do not capture them, sample 20 to 50 cases and label assumptions. A visible estimate supports a better decision than an exact-looking number with no source.

Calculate current monthly labour:

Monthly labour hours = monthly volume × minutes per case ÷ 60
Monthly labour cost = monthly labour hours × loaded hourly cost

Loaded hourly cost includes salary, payroll costs, and relevant overhead. Automation will not remove every minute. Estimate how much work remains for review, exceptions, and system maintenance.

Score business value

Give each factor a score from 1 to 5. Write one sentence beside each score explaining the evidence.

Frequency

A daily task offers more repeatable value than an annual task. High frequency also gives you more observations for testing and improvement.

Labour consumed

Measure total monthly effort, not annoyance per case. A five-minute task repeated 1,000 times deserves more attention than a two-hour task repeated twice.

Cost of delay

Fast handling matters for missed calls, new leads, urgent support, booking changes, and operational handoffs. Score delay highly when waiting loses revenue or creates downstream work.

Error cost

Separate common harmless errors from rare expensive ones. Automation can help when staff make predictable transcription or routing mistakes. High-consequence errors also increase implementation risk, so you will score them again under risk.

Average these factors to get a Value score.

Score feasibility

A valuable workflow can still make a poor first project. Check whether your current operation gives an automated system enough structure.

Process consistency

Score workflows higher when staff follow a stable sequence and agree on the expected output. If five people handle the same input in five different ways, standardize the process first.

Input quality

Structured fields, consistent forms, and accessible documents improve feasibility. Scattered notes, missing customer details, and unreadable scans add uncertainty.

System access

Check whether each system provides an API, webhook, reliable export, or supported integration. Browser automation can bridge some gaps, but interface changes make it more fragile than a direct integration.

Exception rate

Review a sample of real cases. Count how many follow the normal path and group exceptions by type. A workflow with a stable main path and a clear human handoff can work well even when some cases need review.

Average these factors to get a Feasibility score.

Score operational risk

Risk should reduce priority, even when projected savings look strong.

Consider:

• Customer harm if the system acts incorrectly
• Privacy, contractual, or regulatory requirements
• Financial authority given to the system
• Difficulty reversing an action
• Quality of human review and fallback paths
• Business impact during an outage

A draft response that staff approve carries less risk than an agent that sends refunds without review. A scheduling workflow can usually fall back to manual handling. A pricing or compliance decision may require strict controls and a narrower scope.

Average these factors to get a Risk score, where 5 means highest risk.

Rank opportunities without pretending the score is precise

Use a simple weighted score:

Priority score = (Value × 2) + Feasibility - Risk

Value receives extra weight because automation should serve a measurable business outcome. The formula creates a shortlist, not a final decision. Review assumptions, dependencies, and strategic importance before funding work.

Consider this illustrative service-business backlog:

Workflow	Value	Feasibility	Risk	Priority
Missed-call follow-up	5	4	2	12
Invoice data entry	4	5	2	11
Support email drafting	4	3	4	7
Custom pricing approval	3	2	5	3

Missed-call follow-up ranks first because response speed affects opportunity capture, the trigger is clear, and staff can handle uncertain conversations. Custom pricing approval stays low because inputs vary, commercial judgement matters, and a wrong action can damage margin.

Your evidence may produce a different order. Keep notes beside every score so another person can challenge the reasoning.

Choose technology after choosing the workflow

Use the least complex tool that can handle the work reliably.

Workflow characteristic	Likely approach
Fixed trigger, structured data, explicit rules	Deterministic workflow automation
Variable text, documents, or conversations	AI-assisted workflow with bounded tools
High-consequence decision	Human approval before action
Unclear or inconsistent process	Process redesign before automation

Many useful systems combine these approaches. Rules can validate required data, AI can classify an unstructured message, and a person can approve unusual cases. Treat AI as one component inside a controlled workflow.

Run a narrow pilot

Start with one workflow, one team, and a defined class of cases. Agree on success measures before building.

A practical pilot includes:

1. Baseline volume, labour, delay, error rate, and business outcome.
2. Target result, such as response time under two minutes or 50 staff hours returned per month.
3. Approved inputs, actions, and systems.
4. Human review rules and an immediate manual fallback.
5. Logs for every decision, action, exception, and correction.
6. Review date after enough cases have passed through the workflow.

Compare pilot results with the baseline. Include staff review time, software cost, failed cases, and maintenance effort. Expand only when the measured result supports the business case.

Shortlist your first project

Your strongest first automation candidate should have clear volume, visible cost, stable inputs, a repeatable path, manageable exceptions, and a safe fallback. It should also matter to a business metric your team already tracks.

Score three to five workflows rather than debating one idea in isolation. Relative comparison exposes weak assumptions and helps your team fund the project with the clearest return and lowest avoidable risk.