🕑

Estimated setup time:** 15–20 minutes. If you use the default taxonomy and tags without changes, 10 minutes.

Step 1 — Activate the template

1. Open the Phacet template gallery
2. Select "Banking Data Categorization"
3. Click Create a new project from the template

Your project is now pre-configured with all eight processing stages, column definitions, and agent instructions.

Step 2 — Review the expense taxonomy

The template ships with 15 expense categories tuned for startup and SME spend profiles. Review these against your actual transaction mix before running your first batch.

1. Open the expense categorization configuration (Stage 3)
2. Review the default categories: SaaS, Cloud & Infrastructure, AI & ML Tools, Marketing & Sales, Transport, Meals & Entertainment, Accommodation, Payroll & Benefits, Taxes & Social Charges, Banking Fees, Office Supplies, Professional Services, Revenue/Income, Internal Transfer, To validate
3. If your company has expense types not covered, add them to the category list with detection criteria in the agent instructions
4. If you use different category names internally (e.g., "Software Licenses" instead of "SaaS"), rename to match your conventions.

Step 3 — Review the analytical tags

The template ships with eight cost center tags. These should map to your actual department or team structure.

1. Open the analytical tagging configuration (Stage 8)
2. Review the default tags: Sales & Marketing, Design, Marketing, Product Marketing, Product, Engineering, AI Engineering, G&A
3. Add, rename, or remove tags to match your organization's cost center structure
4. Update the tagging guidelines in the agent instructions if you change the tag list.

Step 4 — Add company-specific vendor rules

The template recognizes common vendors (OpenAI, Slack, AWS, Uber, etc.) out of the box. If your company uses vendors that wouldn't be recognized from a typical bank description, add them now.

1. Open the vendor normalization configuration (Step 2) and expense categorization configuration (Step 3)
2. Add rules for any company-specific vendors. For example: if your bank shows TECHCO*MONITORING and that's your Datadog reseller, add a rule mapping that pattern to the vendor "Datadog" and category "Cloud & Infrastructure"
3. If you have vendors that appear under multiple bank description patterns, list the known variations so the agent can collapse them to a single canonical name

Step 5 — Run a test batch

1. Upload your 30–50 test transactions into the project (CSV upload or manual entry)
2. Let the agent process the full batch
3. Review results against this checklist:
   • ✅ Descriptions cleaned to readable labels?
   • ✅ Vendor names normalized consistently (same vendor, same name across rows)?
   • ✅ Expense categories assigned correctly for your known transactions?
   • ✅ Flow direction (Inbound/Outbound) correct?
   • ✅ Opex/Capex split makes sense for your business?
   • ✅ Recurring vs. one-off matches your expectations?
   • ✅ Amounts extracted correctly?
   • ✅ Analytical tags aligned to the right departments?
   • ✅ Payroll transactions masked as CONFIDENTIAL - EXCLUDE?

If categorization accuracy is low

• Add missing vendor rules
• clarify ambiguous category boundaries in the agent instructions
• add detection criteria for categories that aren't matching correctly.

If too many items land in "To validate": the agent defaults to this category when it can't classify with confidence.

• Add vendor-specific rules (Step 4) for the items that should have been classified automatically.

Step 6 — Go live

1. Establish your processing routine: monthly batch after bank export, weekly for higher-volume accounts, or automated via integration (see the Connection Guide)
2. Assign team members to review the Internal Control Process column
3. Set a validation cadence, after each batch, one team member marks records as Ok, NOk, or To verify
4. Export validated results to your accounting system, reporting tool, or budget tracker

Troubleshooting

💻

Start building your workflow directly in Phacet