Every company that asks us "where should we start with AI?" gets the same answer: start with one workflow.
Not an "AI strategy." Not a "digital transformation initiative." One specific, recurring workflow with a clear owner and a measurable outcome.
The selection framework
We've deployed AI teammates across manufacturing, distribution, professional services, agencies, and group companies. The workflows that succeed fastest share four characteristics:
1. It repeats
The workflow happens on a regular cadence โ daily, weekly, or monthly. Quote preparation. Collections follow-up. Monthly reporting. Supplier tracking.
One-off projects are bad candidates. Recurring work is where AI compounds.
2. It has a clear owner
Someone in the organization owns the outcome of this workflow. They know when it's done well and when it's not. They can tell you what "good" looks like.
If nobody owns it, nobody can evaluate whether the AI is doing it right.
3. It touches existing systems
The data the workflow needs already exists somewhere โ ERP, CRM, email, spreadsheets, accounting system. The AI teammate connects to these systems rather than requiring new data entry.
If the workflow requires data that doesn't exist in any system, it's not ready for automation.
4. The outcome is measurable
You can measure the result in hours saved, turnaround time reduced, revenue recovered, or errors prevented. Not "improved efficiency" โ actual numbers.
Good first workflows by department
Sales:
- Quote drafting from RFQ emails (3 hours โ 15 minutes of review)
- Follow-up sequences on sent proposals (40% of follow-ups never happen today)
- Margin checking before quotes go out
- Collections follow-up on overdue invoices (the AR person's least favorite task)
- Weekly cash position report (half a day of manual assembly)
- Bank reconciliation exception flagging
- Supplier PO follow-up (30+ POs with no status tracking)
- Monthly operations report assembly (one week of data pulling)
- Order status compilation (10 interruptions per day)
- Campaign performance report assembly (2 days per month)
- Event follow-up email sequences (3 weeks to start today)
What to avoid
Don't start with:
- Creative work โ AI can draft, but original creative strategy isn't a good first workflow
- Complex decision-making โ Workflows that require nuanced judgment with no clear rules
- Processes nobody understands โ If the current process isn't documented (even informally), automate something simpler first
- The CEO's pet project โ Start where the pain is real and the owner is engaged, not where the enthusiasm is highest
The two-week test
Here's how we validate a workflow choice:
Week 1: Map the workflow. Document the trigger, data sources, steps, approval points, and delivery channel. Connect the AI teammate to the relevant systems.
Week 2: Run the AI teammate on real data with human review on every output. The owner evaluates each output and provides corrections.
After two weeks, you have real data: how many hours saved, what the turnaround time is, how many corrections were needed. If the numbers work, expand. If they don't, pick a different workflow.
No six-month pilot. No abstract ROI projections. Two weeks of real work with real numbers.
Want to identify your best first workflow? Book a 30-minute call and we'll map it together.