The crypto treasury policy playbook: approvals, limits and allow-lists
A practical playbook for designing crypto and stablecoin treasury controls: vault structure, approval quorums, spending limits, allow-lists, velocity rules, separation of duties and audit evidence.
Key takeaways
- Start with vault structure: separate vaults by entity, purpose and temperature (hot, warm, cold).
- Write policies as ordered rules: block what must never happen, then route by amount and destination, then auto-approve the routine.
- Separate who initiates, approves, administers and audits, and require a quorum to change the rules.
- Test rule changes against past transfers before they go live, and keep a verifiable audit trail.
Digital assets move in minutes and can’t be recalled. That makes the controls around a crypto treasury more important than in traditional finance, not less. The good news: the building blocks are familiar to any finance team. This playbook walks through them in the order we recommend setting them up.
1. Design your vault structure first
Policies attach to vaults, so structure comes first. Separate vaults along three lines:
- Entity. One vault (or set of vaults) per legal entity keeps reporting clean and makes segregation easy to prove.
- Purpose. Operating funds, payroll, customer deposits, reserves and trading collateral each have different risk profiles.
- Temperature.
- Hot vaults auto-sign routine transfers under tight limits.
- Warm vaults need human approval for most transfers.
- Cold vaults need an extra offline approval and hold most of the balance.
A useful rule of thumb: keep in hot and warm vaults only what you expect to move in the next few days, and sweep the rest to cold.
2. Write policies as ordered rules
A good policy reads top to bottom, and the first matching rule wins. Structure it in three layers.
Layer 1: hard blocks
Things that must never happen, whatever the amount or initiator:
- transfers to destinations not on the allow-list (for warm and cold vaults);
- transfers flagged as high risk by transaction screening;
- transfers to sanctioned addresses.
Layer 2: routing by amount and destination
| Amount | Destination | Requirement |
|---|---|---|
| Under $10,000 | Allow-listed | Auto-approve (API key or initiator) |
| $10,000 to $100,000 | Allow-listed | 1 approver from Finance |
| Over $100,000 | Allow-listed | CFO plus 1 approver |
| Any amount | New destination | 2 approvers and a cooling-off period |
These numbers are illustrative; set yours from your actual payment patterns.
Layer 3: the routine
Once the risky cases are handled, let routine flows run automatically: sweeps from deposit addresses to treasury, gas top-ups and scheduled payouts to known vendors. Automation inside policy is safer than manual work, because people get tired and scripts with whole keys get stolen.
3. Add velocity limits
Per-transaction limits are not enough on their own; an attacker can send many small transfers. Add velocity limits:
- per vault, per day (for example, $500,000 out of the operating vault);
- per API key, per hour (for example, $50,000 from the payouts key);
- per initiator, per rolling window.
When a limit is hit, transfers should pause for approval rather than fail silently.
4. Separate duties
Map people to roles so no one can complete a sensitive action alone:
| Role | Can | Cannot |
|---|---|---|
| Initiator | Request transfers | Approve their own requests |
| Approver | Approve or reject | Change policies |
| Admin | Propose user and policy changes | Apply them without a quorum |
| Auditor | Read everything | Initiate, approve or administer |
| API signer | Request within its scope | Exceed its policy limits |
Then require an admin quorum for changes to the rules themselves: adding users, editing policies and changing the allow-list. A policy engine that one admin can quietly edit is only a suggestion.
5. Make approvals readable
Approvers should see what will actually happen: the asset, the amount, the destination’s name from the address book, the policy that matched and, for smart-contract calls, a decoded, plain-language description. “Approve 0x…” is how phishing works. Readable approvals are a control in their own right.
6. Test before you change
Before applying a rule change, dry-run it against recent history. How many of last month’s transfers would now be blocked, routed to approval or auto-approved? Dry-runs prevent the two classic failures: a rule so loose it changes nothing, and a rule so tight that operations stop on Monday morning.
7. Keep audit evidence that proves itself
Auditors will want to know who requested, who approved, which policy applied and what was signed, for every transfer and every configuration change. Keep a tamper-evident, hash-chained audit log and export statements with fiat values at transfer time. Give auditors a read-only role instead of screenshots.
8. Plan for recovery and departures
- Departures. Removing a person should take effect immediately, without moving funds. With MPC, no individual holds a whole key, so nothing needs to be rotated on-chain.
- Lost devices. Recovery should be time-locked, require your admin quorum and be cancellable by any admin, so a fake recovery request can’t succeed quietly.
A starter policy, end to end
- Block destinations not on the allow-list (warm and cold vaults).
- Block transfers with high screening risk and alert Compliance.
- New allow-list entries need 2 admins and a 24-hour delay.
- Over $100,000: CFO plus 1 Finance approver.
- $10,000 to $100,000: 1 Finance approver.
- Under $10,000 from the payouts API key to allow-listed vendors: auto-approve, capped at $50,000 per hour.
- Sweeps from deposit addresses to treasury: auto-approve.
How Merkle Vault enforces this
Merkle Vault’s policy engine runs inside the signing service, next to our key share, so share B refuses to co-sign anything that fails your rules, no matter where the request came from. Combined with 2-of-3 MPC keys where your share is always required, every control in this playbook is enforced by cryptography as well as process. See the policy engine or book a demo to model your own rules.
Frequently asked questions
How many approvers should a crypto transfer need?
It depends on size and risk. A common pattern is automatic approval for small transfers to allow-listed destinations, one approver for mid-size transfers, and two or more, including a senior finance officer, above a set threshold.
What is an allow-list?
An allow-list is a set of pre-approved destination addresses or counterparties. Adding to it requires its own approval, and transfers to anything not on it can be blocked or sent for extra review.
What is a velocity limit?
A cap on the total value that can move within a period, such as $500,000 per day from a vault or $50,000 per hour from an API key. It limits the damage from any single compromised credential.