One runtime. Fifty client shops.
Agencies face problems the rest of commerce doesn't - multi-tenant isolation, approval workflows for paid-spend mutations, per-client RBAC, per-day metering, white-label surfaces. Sumeru was built with the agency reality in mind: fast portfolio rollout, slow per-shop blast radius, an audit log that satisfies your enterprise client's procurement team.
Three pain points unique to agencies.
One bad rule, fifty clients.
A typo in a paid-spend automation can hit every shop in your portfolio. Most platforms have no per-shop dry-run gate, no portfolio-wide kill switch, no rollback.
Who approved that bid change?
When ops touches a $50k campaign, the senior PM needs to know. No approval queue means everything ships immediately or nothing ships at all.
Junior ops shouldn't touch paid-spend.
Most platforms have one role per user. Agencies need: scoped read for one client, scoped write for another, paid-spend disabled across the board for juniors.
Four capabilities most relevant to agencies.
Approval queues
When `requiresApproval=true`, fires land as `pending_approval`. A user with `automations:approve` reviews + approves before the worker dispatches. /app/automations/approvals UI.
Per-shop RBAC
145 permissions × per-shop scope. Junior ops gets `read` on all shops, `write` only on assigned. Scoped role inheritance from agency-level templates.
Per-day metering
Metering surfaces dead permissions. /app/rbac-admin shows which permissions are unused per client - agency narrows roles to actual usage, not theoretical.
Campaign Orchestration
Multi-platform plan execution across Google + Meta + TikTok. Daily budget reallocation per client, plan-level approval gates for high-blast actions.
Three agency mechanics.
Capability, not customer outcomes - we have none to report yet.
Every client shop, one approval queue
Spend-affecting and high-blast actions across the entire book land in one reviewer queue, scoped per rule and per shop, with the reviewer recorded on every approval. All 13 action handlers ship snapshot and undo, so an approval that turns out to be wrong is reversible rather than merely logged.
Bad rules surface before money moves
Any rule that can change paid spend serves a seven-day dry-run first, and the window is enforced rather than advisory. During it you see every mutation the rule would have made, on which shop and at what value - so an off-by-one threshold shows up as a list to reject instead of a client conversation to have.
Three platforms, one plan, hourly
Google, Meta and TikTok run from one plan with budget reallocated hourly against attribution rollups rather than each platform's self-reported conversions. The agency reviews the plan; the runtime executes it. Spend changes still serve the dry-run.
What's included.
Demo-led pricing. Per-shop billing or pooled - your choice.
| Capability | Included |
|---|---|
| Multi-shop management | ✓ unlimited shops · per-shop dashboards |
| Approval queues for paid-spend mutations | ✓ agency-tier default |
| Per-shop RBAC + role inheritance | ✓ 145 permissions × shop-scoped |
| Per-day metering surfaces dead permissions | ✓ /app/rbac-admin |
| Custom-weighted attribution | ✓ per-client model definitions |
| Campaign plans (max) | 100 concurrent |
| AI Copilot · pooled or per-shop | 1,000 msg/mo · pooled |
| White-label · client-facing surfaces | ✓ enterprise-tier add-on |
The integrations agencies depend on.
See multi-shop RBAC + approval queues live.
We walk through 5 client shops in your portfolio, show how RBAC inherits, fire a sample approval-required automation in dry-run, and demonstrate the audit trail - end to end in 30 minutes.