Business analysts build better agents than engineers — here's their secret

I've been watching something interesting happen in our agent consulting work. The best agents aren't coming from the engineers with the deepest technical knowledge. They're coming from business analysts who think in workflows, not code.

Here's what they do differently:

They map the human process first

Engineers start with "what can the API do?" Business analysts start with "how does Sarah actually do this task when she's having a good day?" They document the decision points, the edge cases, the places where humans pause and think.

Current process: Invoice approval
1. Check if vendor is in approved list
2. If new vendor → route to procurement
3. Verify PO number exists
4. Check budget remaining in cost center
5. If >$5K → require director approval
6. If approved → update accounting system
7. Send confirmation email with tracking number

Then they give the agent exactly that checklist. Not "handle invoice approval" — the actual steps with the actual decision tree.

They obsess over failure modes

Engineers think about happy path. Business analysts think about what happens when the database is down, when the approval person is on vacation, when someone submits a duplicate invoice.

They build explicit failure handling:

If vendor lookup fails:
- Log the failure
- Route to manual review queue
- Send notification to backup approver
- Set 24-hour follow-up reminder

They define "done" precisely

Engineers say "process the invoice." Business analysts say "invoice is processed when: accounting system shows 'approved' status, vendor receives confirmation email, and expense appears in this month's budget report."

The agent knows exactly when to stop working.

They think in handoffs, not automation

The biggest difference: business analysts don't try to automate everything. They design elegant handoffs between agent and human.

"The agent handles routine approvals under $1K. Anything else gets packaged into a clean summary and routed to the right human with all context attached."

They're not building AI replacements. They're building AI assistants that make humans more effective.

They measure business outcomes, not technical metrics

Engineers track API response times and error rates. Business analysts track "time from invoice submission to payment" and "percentage of invoices requiring human intervention."

The agent gets better at the things that actually matter.

The pattern that wins

Start with the current process. Map every decision point. Define failure modes. Specify handoff conditions. Test with real edge cases.

Your agent doesn't need to be smarter. It needs to be more systematic.

This is exactly the kind of operational discipline that turns a toy chatbot into a production system. The business analysts get it because they've been designing human processes for years. The patterns are the same — you just need to be more explicit about everything.