Start with time, not tools
Most AI automation projects start with “we should use AI for something”. They end with a chatbot nobody uses. The ones that work start with a ledger of where hours go, and only then ask what AI can do about it.
This audit takes one week of light logging and one afternoon of scoring. At the end you have a ranked list of automations with real payback numbers.
Step 1: The Time Ledger
For 5 working days, every person on the team logs recurring tasks. Not projects. Recurring, repeatable tasks. Use this format:
| Task | Who | Times / week | Minutes each | Trigger | Output goes to |
|---|---|---|---|---|---|
| Copy leads from form to CRM | Sam | 25 | 4 | Form submit | HubSpot |
| Draft first reply to inbound quote | Priya | 12 | 15 | Email arrives | Gmail |
| Weekly client status email | Sam | 8 | 20 | Friday | Client |
| Chase unpaid invoices | Ade | 10 | 6 | 7 days overdue |
Multiply times by minutes. Most 10-person teams find 30–60 hours a week on this list.
Step 2: The FRIC score
Score each task 1–5 on four dimensions:
- F, Frequency. How often does it happen? Daily = 5, monthly = 1.
- R, Rules. How describable is the decision? “Always do X” = 5, “depends on gut” = 1.
- I, Inputs. Is the data already digital and reachable by API? Yes = 5, on paper or in someone’s head = 1.
- C, Cost. Hours × hourly cost × 52. Bucket: under $2k/yr = 1, over $20k/yr = 5.
Add them up. Anything 16+ is an automation candidate. Anything under 10 is not worth it yet.
Step 3: The six that pay back fastest
Across the businesses we have automated, these six recur and pay back in under 90 days.
1. Inbound lead triage and enrichment
Form or email arrives → AI extracts fields, enriches with company data, scores fit, creates a CRM record, drafts a first reply for approval. Build: 1–2 weeks. Typical saving: 8–15 hrs/week.
2. Document intake
PDFs, invoices, contracts, applications arrive → AI extracts structured data, validates, files it, flags anomalies. Build: 1–2 weeks. Saving: 5–20 hrs/week depending on volume.
3. Support first-response
Ticket arrives → AI classifies, pulls the relevant doc or account data, drafts a reply. Human approves for the first 60 days, then auto-sends for high-confidence categories. Build: 2–3 weeks. Saving: 10–25 hrs/week.
4. Reporting and status updates
Pull from your systems on a schedule → AI writes the weekly client / investor / team update in your voice → human edits for 3 minutes. Build: 1 week. Saving: 4–8 hrs/week.
5. Collections and follow-up sequences
Invoice overdue, quote unanswered, trial ending → AI writes a context-aware follow-up, escalates tone over time, stops when a reply arrives. Build: 1 week. Saving: 3–6 hrs/week plus faster cash.
6. Content repurposing
One long-form input (call recording, blog post, doc) → AI generates the derivative assets (summary, social posts, email, FAQ) into your queue for review. Build: 1 week. Saving: 4–10 hrs/week.
Step 4: Build, buy or Zapier?
- Zapier / Make when: fewer than 5 steps, no judgment calls, under 1,000 runs a month, and nobody will be embarrassed if it breaks on a Friday.
- Buy a SaaS when: the problem is generic (invoicing, scheduling) and a product does 90% of it.
- Build custom when: the workflow is your competitive edge, needs your data, involves judgment, or runs at volume. Custom automations cost more up front and dramatically less at month 12.
Step 5: ROI math that survives a CFO
Annual saving = hours/week × 48 × loaded hourly cost
Build cost = one-off
Run cost = (API tokens + hosting + maintenance) × 12
Payback months = build cost ÷ ((annual saving − run cost) ÷ 12)Example: lead triage saving 10 hrs/week at $60/hr = $28,800/yr. Build $9,000, run $1,800/yr. Payback: 4 months.
90-day rollout
- Days 1–7: Time Ledger.
- Days 8–10: FRIC scoring. Pick the top 2.
- Days 11–40: Build #1 with human-in-the-loop approval. Measure.
- Days 41–70: Build #2. Move #1 to auto for high-confidence cases.
- Days 71–90: Review the ledger again. The list will have changed.
What to do next
Run the ledger this week. If you want us to score it with you and estimate the top two builds, book a call. We maintain every automation we ship, so you are not left holding a Zap.