How to prioritize automation projects by ROI

Rank automation candidates with a simple ROI formula — (hours recovered × loaded cost) − implementation cost — adjusted for dependency risk and cross-team impact, not stakeholder volume in meetings.

This guide supports the automation roadmap framework. Use it after discovery produced a candidate list with evidence.

Step 1: Estimate hours recovered

For each candidate, document:

Field	Example
Manual step description	Re-key policy data from email to PMS
Minutes per occurrence	12
Occurrences per week	35
FTEs affected	4

Hours saved per week = (minutes × occurrences × FTEs) ÷ 60
                     = (12 × 35 × 4) ÷ 60 = 28 hours/week

Round down 20% for adoption slippage unless you have pilot data.

Step 2: Convert to annual recoverable OpEx

Loaded hourly rate = annual fully loaded comp ÷ 2,080
Annual recoverable ≈ hours saved per week × 52 × loaded rate

Example: 28 hrs/wk × 52 × $45/hr loaded = $65,520/year

Use role-specific loaded rates — claims handler vs underwriter comp differs.

Step 3: Score impact, effort, and risk (1–5)

Score	Impact (recoverable value)	Effort (implementation)	Risk (failure modes)
5	>$150K/yr or strategic	Single system, rules-based	Proven pattern, low change resistance
4	$75–150K/yr	2 systems, API available	Moderate training
3	$35–75K/yr	3 systems, custom integration	Cross-team adoption needed
2	$15–35K/yr	Legacy, no API	Political sensitivity
1	<$15K/yr	Multi-year platform change	High attrition during build

Priority index (simple): Impact × 2 − Effort − Risk
Higher is better. Tie-break with strategic alignment (customer-facing latency, compliance).

Step 4: Worked example

Three candidates from the same discovery cycle:

ID	Pain	Hrs/wk	Annual $	Impact	Effort	Risk	Index
A	Intake OCR + validation	18	$42K	3	2	2	2
B	Unified submission layer	45	$105K	5	4	3	3
C	Email approval bot	6	$14K	1	2	4	-4

Sequence: A first (quick win, positive index), B next (highest total recovery), C deprioritize (low $, high risk).

Step 5: Common prioritization mistakes

1. Automating broken process — if the step should not exist, eliminate it; do not bot it
2. Ignoring handoffs — local automation that pushes work downstream reduces net ROI
3. Using gross hours without loaded cost — 10 hrs/wk at $80/hr loaded ≠ 10 hrs at $25/hr
4. Skipping maintenance — budget 15–25% annual run cost for RPA and integrations
5. One-time ranking — re-score quarterly; hidden bottlenecks shift

Linking ROI to evidence

Every line item should cite discovery evidence:

• "Claims + underwriting both reported manual doc re-verification (n=12 interviews)"
• "Cross-validated: same pain, three roles"

ROI without evidence is fiction. Evidence without ROI is unreadable. Pair them in the backlog table.

When discovery pre-ranks for you

Manual scoring works for 10–20 candidates. At workforce scale, AI process discovery deduplicates signals and outputs a pre-ranked backlog — you still apply the rubric above to validate and socialize with engineering.

Your next action

Export your current automation wish list. Score the top five with impact, effort, and risk. Kill anything with index below zero unless strategically mandated. Publish the ranked list to engineering this week.