How to Calculate ROI on an Automation Project Before You Build It

You want to automate something. Your CFO asks: “What’s the business case?” If you answer “it’s cool,” you won’t get budget. If you answer “we’ll save $40k/year and the project costs $8k,” you’ll get approved. Here’s the framework.

The Simple Formula

Let’s build this out with real numbers.

Component 1: Hours Saved

Step 1: Baseline the Current State

What task are we automating? How much time does it take?

Example: Invoice Processing

Current state:

• 50 invoices arrive per week
• Each invoice takes 8 minutes to process (read, extract line items, enter into system)
• Total time: 50 × 8 = 400 minutes/week = 6.7 hours/week

Who does it:

• One person spends 25% of their time on this
• Their salary: $60k/year loaded ($28.85/hour)

Step 2: Calculate Efficiency Gain

What percentage of the task will automation handle?

• Best case: “The system reads PDF, extracts line items, stores in database. We don’t touch it.” → 95% automation
• Realistic case: “The system extracts data, a person reviews it, approves it, submits it.” → 70% automation
• Conservative case: “The system extracts data, a person reviews it.” → 50% automation

Use the realistic case. Assume 10–15% manual review still required.

Component 2: Error Reduction Value

Current errors:

• 2–3% of invoices are processed incorrectly (wrong vendor, wrong amount, duplicated)
• Wrong invoices create rework: 15 minutes each
• Cost per error: 15 min × $28.85/hr = $7.21
• Annual error cost: 50 × 52 × 2.5% × $7.21 = $468/year

After automation: the system makes 0.5% errors (misses something, extracts wrong value), but it never duplicates or misroutes (structured validation). New error cost drops to $93.60/year.

For some workflows, error reduction is significant. For others, negligible.

Component 3: Speed-to-Customer Value

Does faster processing generate revenue or reduce costs downstream?

Example: Faster invoice processing leads to faster invoicing.

Current: Invoice received Wed → processed Fri → customer billed Mon → customer pays Wed = 7 days. Automated: Invoice received Wed → processed Wed (within hours) → customer billed Thu → customer pays Sat = 3 days.

Benefit: 4-day faster cash collection. On $500k/month invoiced, 4-day faster collection = $500k × (4/30) = $67k in average working capital freed up. At 5% cost of capital: $67k × 5% = $3,350/year.

Another example: Faster support triage reduces customer churn. If faster response prevents 1 churn per month at $10k customer lifetime value: 12 × $10k = $120k/year.

Component 4: Build Cost

Approach 1: Time-based (internal build)

Add contingency: 15–20% ($3–5k).

Approach 2: Vendor-based (outsource)

• Consulting shop (like Three Moons Network): $8–15k for a focused automation project
• Time: 3–4 weeks

Use Approach 2 if speed matters. Use Approach 1 if you want to retain knowledge. For this example: $10,000 build cost (3-week project with consulting partner).

Component 5: Annual Operating Cost

Infrastructure:

• Lambda: 50 invoices/week × 1 min runtime × $0.0000166667 per GB-sec (2GB) = ~$10/month = $120/year
• DynamoDB: storage + throughput = ~$5/month = $60/year
• API costs (Claude): 50/week × ~$0.10 per invoice = $260/year

Maintenance:

• Monitor alerts, fix bugs: 2 hours/month = $575/year
• Schema changes, updates: 4 hours/month = $1,150/year

Putting It Together

This doesn’t look great unless we’re looking long-term or increasing volume.

But: if the company processes 200 invoices/week instead of 50, labor savings jump to $40k/year, and payback becomes 6 months.

Always model for realistic scale, not just current volume.

When NOT to Automate

Unless you expect volume to grow 5x in the next year, error reduction is worth $10k/year (unlikely), or speed generates revenue.

If payback is > 2 years, challenge the premise. Either the labor savings are wrong, or this isn’t a good project.

Real-World Template

Use this template. Fill it out. Share it with your CFO.

If the numbers don’t work, don’t build. If they do, move forward confidently.