The ROI framework we use to decide what to automate (and what to leave alone)

“Should we automate this?” is the wrong question. The right question is “of the 47 things we could automate, which three will pay back fastest?”

Here is the framework we use, end-to-end, on every discovery engagement.

Step 1: Inventory the candidates, aggressively

Walk the floor. Shadow the ops team. Read the Slack channels. Pull the SaaS bills.

Candidates come from:

• Spreadsheets anyone opens more than 10 times a week.
• Messages in Slack or WhatsApp that are essentially “ping when done”.
• Copy-paste between tools.
• Emails that look 80% the same.
• Any process a team insists “only X knows how to do.”
• Any metric leadership wants to see but no one reports consistently.

Typical discovery yields 30–60 candidate automations. This is normal. Most teams have never written them down in one place.

Step 2: Score each candidate on four axes

For every candidate, we score:

Frequency (how often it happens)

• 5 — Multiple times per day
• 4 — Daily
• 3 — Weekly
• 2 — Monthly
• 1 — Quarterly or less

Duration (how long it takes a human each time)

• 5 — 60+ minutes
• 4 — 20–60 minutes
• 3 — 5–20 minutes
• 2 — 1–5 minutes
• 1 — Under a minute

Variance (how repeatable is it, really)

• 5 — Nearly identical every time
• 4 — Mostly the same, minor variation
• 3 — Pattern-based but judgment required
• 2 — High variation, frequent exceptions
• 1 — Every case is genuinely different

Criticality (what happens if it breaks)

• 5 — Customer-facing, money-moving, or compliance-critical
• 4 — Operations stops without it
• 3 — Noticeable slow-down
• 2 — Mild inconvenience
• 1 — Nobody notices for a week

The raw annual hour saving from automation is roughly:

frequency_multiplier × duration_in_minutes ÷ 60 × weeks_per_year

Where frequency_multiplier is the annualized count. A duration-5, frequency-5 task running 10× per day, 5 days per week, 50 weeks per year = 2,500 runs × (for example) 45 minutes = ~1,875 hours per year. That is a full-time person.

Step 3: Score the build cost honestly

Not every candidate is cheap to automate. We grade effort on a 1–5 scale:

• 1 — A few hours. Existing integration, trivial logic.
• 2 — A day or two. Clean APIs on both sides.
• 3 — A week. Real data work, some edge cases.
• 4 — 2–4 weeks. Custom integrations, significant logic, UI required.
• 5 — More than a month. New subsystem, significant data work.

Step 4: Compute the payback

Annual value = (hours saved per year) × (blended hourly cost of the person doing it).

Build cost = (effort score) × (your blended engineering rate for that effort level).

Payback period (months) = build cost ÷ (annual value ÷ 12).

Anything with a payback under 3 months is an obvious yes. 3–6 months is a “probably, if it fits the roadmap.” Over 12 months usually means we have the scope wrong or we are solving the wrong problem.

Step 5: Apply the strategic multipliers

Pure ROI is not the only lens. We multiply by:

Confidence (×0.5 to ×1.5)

How sure are we that this automation will work without unforeseen complexity? Brand-new technology with no production track record gets a 0.7. Patterns we have shipped 5+ times get a 1.2.

Capability build (×0.8 to ×1.4)

Does building this unlock the next 3 automations? If yes, ×1.3. A beautiful one-off that teaches us nothing, ×0.9.

Change management (×0.6 to ×1.2)

Will the team actually adopt it? If it requires retraining 40 people and a cultural shift, ×0.7. If it is invisible to everyone except the person currently doing the work, ×1.2.

Regret (×0.7 to ×1.4)

How bad is “we should have done this 6 months ago”? Things that compound (better data quality, better routing) get a regret multiplier. Things that are static (one-off reports) do not.

Step 6: Pick the top 3. Ship them. Re-prioritize.

We rank the full list by adjusted score, pick the top three, and commit to shipping them before touching anything else. Every 4–6 weeks we re-score — because shipping the first three usually unlocks or invalidates items 4–10 on the list.

The counter-intuitive finding

Running this exercise with clients, we consistently find that the automation people were most excited about is rarely in the top three. What shows up in the top three is almost always an unglamorous hygiene task — onboarding, invoice chasing, ticket routing — that nobody had the patience to fully document.

Glamour does not scale. Arithmetic does.

The shortcut if you do not want to run this yourself

Our paid discovery engagement is exactly this framework, applied with a discipline most internal teams struggle to maintain (because they are the ones currently drowning in the work we are measuring).

If you want to do it yourself, the framework above is enough. If you want us to do it with you, book a call — we will run the process, hand you the ranked backlog, and you can decide to build it with us, build it internally, or not build it at all.

Either way, the list is yours to keep.