How to Audit Your CRM for AI Opportunities (in 90 Minutes)

Most “AI strategy” engagements start the same way: a four-week discovery phase, a thirty-page report, and a backlog of ideas that nobody has time to actually build. By the time the recommendations land, the people who could implement them have moved on to the next quarter’s fires.

You don’t need that. You need ninety minutes, a spreadsheet, and the discipline to ask five honest questions about your own org.

Here’s the audit we run on the first day of every engagement — translated into something you can do yourself, before you ever hire anyone.

Why the usual approach fails

Most AI audits look for “areas of opportunity.” That phrase is the problem. It’s too vague to act on, so the report gets full of recommendations like “Explore predictive lead scoring” or “Investigate generative AI for case summarization.” All technically true. None of them tell you what to build first.

The audit below produces a different output: a ranked list of specific workflows, each with an estimated build time, an estimated time saved per week, and a clear yes/no on whether AI is the right tool. You can hand it to a developer (or to us) on Monday morning and they can start shipping by Wednesday.

What you need before you start

• Admin access to your Salesforce org (or whatever CRM you use)
• A spreadsheet (Google Sheets, Excel, or Airtable — doesn’t matter)
• 90 minutes of uninterrupted time
• Your team’s last two sprint retros or status reports — anywhere they’ve complained about manual work

That’s it. No workshops. No stakeholder interviews. You already know where the pain is — you just need a structured way to write it down.

Step 1 — Make a list of every workflow that touches the CRM

Open your spreadsheet. Make one row per workflow. Don’t filter, don’t prioritize, don’t worry about whether AI is involved. Just write down every recurring thing that happens in or around the CRM.

Examples to seed the list:

• Inbound leads get qualified and assigned
• Sales reps log meeting notes
• Support cases get routed to the right team
• Quotes get drafted and sent to customers
• Renewals get tracked and reminders go out
• Reports get pulled for the weekly sales meeting
• Data gets cleaned up before quarterly close
• New rep onboarding (the “where do I find X” questions)
• Approvals route through managers
• Invoicing data flows to accounting

Aim for 25–40 rows. You’ll rarely come up short — most teams have more workflows than they realize.

Step 2 — Tag each workflow on three axes

Add three columns to your spreadsheet:

Frequency — how often does this workflow run?

• daily (multiple times per day)
• weekly (a few times per week)
• monthly (a handful per month)
• quarterly (rare)

Manual effort — how much human time does each instance take?

• < 5 min
• 5–30 min
• 30 min – 2 hours
• > 2 hours

Repetition score — how similar is each instance to the last one?

• 5 — nearly identical every time (e.g. “qualify lead against ICP rules”)
• 3 — similar structure, different content (e.g. “draft a quote”)
• 1 — every instance is different (e.g. “investigate a one-off support escalation”)

The combination of these three columns is the entire signal. High frequency × high effort × high repetition is exactly the workflow AI is best at automating. Low repetition is where humans should stay.

Step 3 — Calculate weekly hours burned

Add a column called weekly_hours_burned. For each row, do the math:

weekly_hours_burned = instances_per_week × avg_minutes_per_instance / 60

Convert your frequency tag to instances per week (e.g. daily ≈ 25, weekly ≈ 5). Use the midpoint of your manual effort bucket.

You’ll find that 80% of the time burn is concentrated in 5–8 workflows. This is the Pareto cut, and it’s the most useful number on the spreadsheet. Anything below the cut is a distraction. Anything above it is worth a deeper look.

One client we ran this with expected lead triage to be the biggest line item. It came in fourth. The biggest drain was actually “manually pulling weekly pipeline reports” — 14 hours a week across the team, which they’d never tracked because no single person owned it.

Step 4 — Score AI fit on each high-burn workflow

For each row in the Pareto cut, ask three questions and score 1–5:

Pattern recognition — does this workflow involve recognizing a pattern from text, structured data, or images? (Lead scoring: yes. Booking a meeting: no.) Score 5 if pattern recognition is the entire job.

Structured output — does the workflow produce output that fits a clear schema? (Extracting fields from a contract: yes. Writing a creative blog post: no.) Score 5 if the output is fully structured.

Tolerance for review — can a human review the output before it takes effect, and does that review take less time than doing the work from scratch? Score 5 if review is fast and the human is already in the loop.

The total AI fit score is the sum (max 15). Anything ≥ 11 is a strong candidate. Anything ≤ 7 is probably the wrong tool.

The trap to avoid: “creative” workflows where the bar for output quality is subjective. AI can draft them, but the review takes longer than writing from scratch. Skip these for your first three automations.

Step 5 — Estimate build time and rank

For each strong candidate, estimate build time honestly:

• Trivial (< 1 day) — single API call, no integrations, no UI
• Small (2–5 days) — one integration, simple UI, basic error handling
• Medium (1–3 weeks) — multiple integrations, custom logic, dashboards
• Large (1+ months) — full system, multiple users, training, rollout

Then sort the spreadsheet by (weekly_hours_burned × AI_fit_score) / build_time_in_days. The top of that list is your build queue, in order.

This gives you a ranked, defensible, objective prioritization. No politics. No “who shouted the loudest in the meeting.” Just the math.

What the output actually looks like

Here’s a snippet from a real audit we ran with a 30-person ops team last quarter (anonymized):

Workflow	Hours/week	AI fit	Build days	Score
Lead qualification + routing	8	14	3	37.3
Weekly pipeline report assembly	14	12	5	33.6
Vendor PDF → Contract record	6	13	4	19.5
Meeting notes → SF Tasks	5	11	2	27.5
Renewal reminder workflow	4	9	3	12.0
Case routing	7	13	6	15.2

We built the top three in the first month. The team got back 28 hours per week — most of a full headcount — for about 12 days of build effort.

What this is not

This audit is intentionally not:

• A vendor evaluation. Don’t pick tools yet. The audit tells you what to build, not who builds it.
• A consensus exercise. You don’t need stakeholder workshops. You need 90 minutes of honesty about where the time goes.
• A predictive model. Build estimates are rough. The point is relative ordering, not absolute precision.

Build it. Measure the actual time saved after 30 days. Use those numbers to fund the next round.

When to bring in help

If you do this audit and the top of your list is straightforward — “draft renewal emails,” “extract data from PDFs,” “route support cases by topic” — you can almost certainly build the first 3 yourself with Claude Code, an MCP server, and a few weekends.

If the top of your list is more involved — “orchestrate a multi-system approval workflow,” “build an internal Q&A bot grounded in our knowledge base,” “automate the entire QA pipeline for our Salesforce releases” — that’s where we come in. Send us the spreadsheet, and we’ll come back with a one-page proposal: what we’d build, in what order, and what it costs.

Book a free strategy call →. Bring your audit. We’ll walk through it together and tell you which three things you should build first — whether you hire us or not.