Most operators trying to figure out "what should we automate with AI" make the same two mistakes. They start with the workflow that's loudest, not the one with the highest ROI. And they go too granular too fast, picking a specific tool before understanding the workflow shape.
This is the audit we charge $1,495 to do externally. The DIY version below gets you 70% of the way for the cost of one focused afternoon. If your time is worth more than $1,500, hire someone (us or otherwise). If you're an operator who likes to run their own analysis, this works.
Step 1: Map the recurring work
Open a blank doc. List every workflow at your team that runs at least 5 times per week.
Don't list "tasks." List workflows. A workflow has: - A trigger (something that starts it) - A sequence of steps - A clear output - An owner
Examples of real workflows: - New customer signs up → CS welcomes them → CS schedules onboarding call → onboarding happens → success metric tracked - Sales call happens → notes captured → action items logged in CRM → follow-up email sent → meeting scheduled - Inbound lead arrives → routed to AE based on territory → AE reaches out within 24h → response logged
Most teams have 15-25 of these workflows. List all of them. Don't filter yet.
Step 2: Score each workflow on 4 dimensions
For each workflow, give it a 1-10 score on:
A. Frequency. How often does this workflow run? 1 = monthly, 5 = weekly, 10 = multiple times daily.
B. Time cost per instance. How long does the workflow take, end to end? 1 = 5 minutes, 5 = 30 minutes, 10 = 4+ hours.
C. Tool count. How many systems does it touch? 1 = one tool, 5 = three tools, 10 = five or more tools across teams.
D. Cost of getting it wrong. What happens if the workflow fails? 1 = nothing, just inconvenient, 5 = customer-noticeable but recoverable, 10 = legal/financial/reputation exposure.
Add the four scores. The total ranges from 4 to 40.
Step 3: Filter by the build/buy/skip rubric
For each workflow, apply this filter:
Score 30+ AND tool count ≥ 5 AND wrongness cost ≤ 7: Strong build candidate. High frequency, multi-tool, but recoverable if AI gets it wrong. This is where productized AI agents shine.
Score 30+ AND wrongness cost ≥ 8: Augment with human gate. AI handles the mechanical layer, human reviews before completion. Don't fully automate. Examples: anything customer-facing, anything legal, anything financial that can't be reversed.
Score 20-29 AND tool count ≥ 4: Productized fits. Standard AI agent SKUs ($2,995-$4,995) handle this category cleanly. Inbox Triage, Meeting Notes, CRM Hygiene, Proposal Drafter all live here.
Score 20-29 AND tool count ≤ 3: SaaS or Zapier fits. Buy off-the-shelf or connect existing tools. Don't custom-build.
Score below 20: Skip for now. Workflow isn't running often enough or doesn't cost enough to justify automation. Revisit when frequency or impact grows.
This filter typically narrows 15-25 workflows down to 3-5 real candidates.
Step 4: Calculate ROI for the top 3
For the top 3 workflows from Step 3, do the math:
Annual time cost: Frequency per week × time per instance × 50 weeks = hours/year.
Annual dollar cost: Hours/year × loaded hourly rate of the people doing the work. (Use $50-100/hr for admin, $100-200/hr for senior individual contributors, $300-500/hr for execs/founders.)
AI build cost: Productized agent ($2,995-$4,995) + ~$1,500/year tool cost. Custom build ($8K-$30K) + ~$3,000/year tool cost.
Year-one ROI: (Annual dollar cost - AI build cost) / AI build cost.
If the year-one ROI is below 2x, deprioritize. If it's between 2-5x, build when timing is right. If it's above 5x, build now.
Most teams find their top workflow has a year-one ROI between 8x and 30x. That's the one to build first.
What this audit catches that "what should we automate" doesn't
Three patterns this approach surfaces:
1. The workflow you complain about isn't usually the highest ROI. The loudest pain points are often medium-frequency, single-tool problems. The highest-ROI workflows are usually quieter, repetitive, multi-tool, costing 2 hours each but happening 50 times a week.
2. The workflow nobody mentioned is often the winner. Post-meeting CRM logging is a good example. Nobody complains about it because each instance is small. But 8 minutes × 80 calls/week × 50 weeks = 533 hours/year. At blended $200/hr, that's $107K of recovered capacity.
3. The workflow you tried to automate before but it didn't work usually had the wrong tool, not the wrong workflow. Productized AI agents in 2026 work for things that Zapier in 2022 couldn't handle.
What to do with the audit output
Three options:
1. DIY build: The simpler workflows can be built with off-the-shelf tools (Zapier, n8n, Make.com) plus light AI integration. Works for workflows scoring 20-29 on the rubric.
2. Productized build: Workflows scoring 30+ that fit the standard SKU categories (inbox, CRM, meetings, proposals, hygiene), buy a productized agent. $2,995-$4,995 per SKU, live in 7 days.
3. Custom build: Workflows scoring 30+ that are unique to your team or industry. Custom development, $8-30K depending on scope.
The audit usually surfaces 1 productized build + 1 custom build as the "ship in next quarter" plan. Most operators come out with a clear sequence and stop spinning their wheels on "where to start."
The honest caveat
This audit takes 90-120 minutes if you do it well. It's a thinking tool, not a script. The hard part is being honest about which workflows are actually painful versus the ones that are just visible.
If you don't have the time or want a second opinion, the paid Audit Pro is $1,495 (refundable on any Starter SKU you commission). Output is a written report within 24 hours of the call. Most operators choose paid because the second opinion catches blind spots.
Get started
Want a real number for your specific situation?
30-minute audit call walks through your workflows and outputs a fixed price for the 2-3 things worth automating first.