Ramp Accounting Agent — Enablement & Daily Use (Admin guide)
Overview
Available on Ramp Plus. See Ramp Plus overview for plan details.
Feature overview
This feature is turned on by default for all Ramp Plus accounts.
- AI-Coding — Ramp's AI automatically codes all relevant accounting fields using transaction context and history for entries you typically populate when syncing to your ERP system.
- Suggested Actions (Beta) — recommends which transactions to mark ready or manually review.
- Auto-mark ready (Beta) — marks low-risk, fully completed expenses as ready.
AI coding
Ramp’s Accounting Agent is always running to automate your busywork.
Settings
To adjust the settings of AI coding navigate to the Accounting tab → Settings → Automation Settings.
You can then choose whether you want AI coding to auto-code:
- All fields (default)
- Some fields by selection
- No Fields
You can also toggle on whether to auto-mark ready these transactions.
For every transaction, the agent evaluates GL account, department, class, and location and steps in under these conditions:
When the agent will code a field
- Blank field + high confidence → AI takes the first pass and fills it. Example: Starbucks $8 “client coffee” with GL blank → AI sets Meals & Entertainment (Client).
When the agent will not override
- Employee value present → AI does not change it. Example: Employee chose Travel: Airfare for Uber → AI leaves it.
- Admin value present → AI does not change it. Example: Admin set Client Gifts for Staples → AI respects admin value.
- Rule value present→ AI does not change it. Example: Apple $2.99, rule Computer Equipment→ keeps rule
Review faster with suggested actions and smart groups
Suggested actions
You’ll see a Suggested action column. Each transaction row shows a suggested-action badge (e.g., “Waiting for Cardholder” “Waiting for Approval,” “Mark ready”).
- Hover a badge to see why it’s suggested (receipt present, all fields coded, approved, etc.).
- Click a badge to jump straight to the fields that need attention.
Suggested actions that Ramp will provide:
| State | Meaning |
|---|---|
WAITING_CARDHOLDER | The transaction is waiting on input or action from the cardholder. |
WAITING_APPROVAL | The transaction is waiting for approval. |
REVIEW_MATERIALITY | The transaction amount is high enough to be considered potentially risky and should be reviewed. |
REVIEW_GL_ACCOUNT | The current GL account coding on the transaction appears suspicious and should be reviewed. |
REVIEW_EXPENSES | The transaction appears suspicious overall, due to coding or other factors, and should be reviewed. |
REVIEW_AMORTIZATION | The transaction has been amortized from a previous period and requires review. |
SUGGEST_MARK_READY | Ramp believes the transaction is properly coded and not suspicious, and recommends marking it as ready. |
Smart groups
Toggle ONSmart Groups at the top of the Card transactions table (optional).
- When Smart Groups are ON: Transactions are grouped under a header row that highlights the suggested action for that set (with an item count).
- Ready to sync
- Complete, high confidence, auto-marked ready (by AI, a rule, or a user).
- Complete with medium/low confidence and a recommendation to mark ready.
- Needs review
- Complete, high confidence but requires review (e.g., materiality).
- Complete with medium/low confidence.
- Policy or workflow holds
- Awaiting approvals, missing submission policy, or out of policy.
- Ready to sync
When to use which view:
- Use the Suggested action column (Smart Groups OFF) for quick, line-by-line review.
- Use Smart Groups (ON) to batch similar work under a clear header and reduce context switching.
Improve accuracy with AI-coding feedback
Use the feedback prompt whenever you change AI coding.
When an admin overrides AI-applied coding, Ramp prompts for feedback. Add a brief note so the agent learns from your correction.
Coach reviewers to be specific.
Ask reviewers to include short, concrete guidance—e.g., “client meal, not team” or “software → user support.” Specific notes improve future suggestions.
Know the impact.
Corrections and feedback train the model, reducing future edits and increasing auto-coding accuracy over time.
- Tip: Keep the note to one sentence. Name the distinction (e.g., client vs. internal; software vs. support). No need to restate details already visible on the transaction.
Two paths to “zero-touch”
Get items toReadywithAuto-mark ready**; move them to the ERP withAuto-sync.**
Used together, these create a fast lane from swipe → ERP for in-policy, fully completed transactions.
How to auto-mark items “Ready” (two options)
- Advanced rules (if–then logic) Use rules to mark transactions ready when they meet exact conditions.
- Example: IF Merchant = Zoom AND receipt present AND all required fields coded THEN Mark ready.
- Ramp AI Allow Ramp’s AI to identify eligible transactions based on signals like receipt/memo present, required fields complete, in-policy, and high-confidence coding.
- Example: Small recurring SaaS charge with receipt + fully coded → AI marks Ready .
Note: Control lives here: Decide what can be auto-marked ready via rules and/or AI. Use your high-spend flag and policy settings to keep judgment calls out of the ready lane.
How auto-sync works
- Global switch: When Auto-sync is ON, all transactions already marked Ready will sync to your ERP nightly (at midnight) .
- Important: Auto-sync does not decide what’s safe. It simply moves your Ready queue. The control is on the Auto-mark ready side.
Safe rollout pattern (recommended)
- Start by auto-marking repeatable, low-risk spend using advanced rules (e.g., small recurring vendors).
- Layer in AI-based auto-marking for similar low-risk patterns once signals are reliable (receipts/memos present, fields complete).
- Turn Auto-sync ON once you trust that items reaching Ready are truly post-worthy. Keep Smart Groups on to monitor “Needs review” and “Policy holds.”
Frequently asked questions
Can I override Ramp’s suggestions?
Yes.
How does this affect the employee experience?
If your business requires employees to code transactions, those codings are not affected by the accounting agent.
What fields are coded by the accounting agent?
All fields that have historically been frequently used to code transactions will be coded.
How does this work with conditional filtering?
The accounting agent will respect conditional filtering rules by not coding any fields that are dependent on the input of another field.
How can I tell if something is AI coded?
AI coded fields will have a Ramp intelligence icon:
Can I filter to only see AI codings?
We are currently working on building this functionality, but it is not available today.
Can I turn this off for specific fields?
To adjust the settings of AI coding navigate to the Accounting tab → Settings → Automation Settings.
You can then choose whether you want AI coding to auto-code:
- All fields (default)
- Some fields by selection
- No Fields
Will AI change a rule’s coding?
No. AI will not override rule-based codings.
What happens with AI coding when a transaction clears but the cardholder hasn't uploaded a receipt or memo? Does the AI update its coding if a receipt or memo is added later with new information? Or does the AI wait to code these transactions?
The AI codes transactions immediately upon clearing. However, if new information (such as a memo) is added later, the AI will reconsider and update to the best coding suggestion.
If a transaction is already marked "ready," it will not be updated with new information.
How do I know which items are safe to sync?
Start with items that are complete, high confidence, and already auto-marked ready. Then expand using your high-spend flag and policy checks.
Can I turn Smart Groups off?
Yes. It’s optional and off by default. Enable it when you’re ready.
What if an employee coded a field incorrectly?
AI won’t override employee entries. Fix it and move on—your edits help improve future suggestions.
Can I set my own AI confidence threshold?
Not at this time.
What do the icons mean on transactions?
A person icon = coded by an employee. A bolt icon = coded by a rule. Low confidence appears in yellow with on-hover details showing who coded it and why.
Will the agent retroactively change past codings after it learns?
No. It won’t go back and adjust prior entries. Admin corrections and feedback inform future codings.
Can I review the feedback submitted to the agent?
Not today.
Does AI coding work for Ramp Checking Account open-loop transfers (external deposits and withdrawals)?
Ramp suggests a GL account (Category) for Ramp Checking Account open-loop transfers (such as external deposits and withdrawals) in the Accounting > Banking table. Ramp Banking AI coding suggests the GL Category field only. If your business has defaults set up, the defaults will override AI coding and be applied automatically. See Ramp Banking Accounting for details.
Is Accounting Agent a Ramp Plus feature?
Yes—all agents are gated behind Ramp Plus, and Accounting Agent is priced for Plus.