Technology
Built for decisions
you can defend.
A structured campaign state, rules you can read and models routed per task. Every suggestion comes with its evidence and can be reverted.
- 06Audit logEvery suggestion, approval and change
- 05Publish pipelineApproved changes only · rollback · 7-day revert
- 04ReasoningYour rules + a Friday AI review
- 03Campaign stateTyped record of spend, results and settings
- 02GateOSRate windows · breaker · MCP → Graph fallback
- 01SourcesMeta · Google Ads · TikTok · GA4
Campaign state
Every campaign becomes a structured record.
Friday AI never reasons over a pile of exports. Each sync turns spend, results and settings into fields on one typed record per campaign.
{"campaign": "CP-1121","platform": "meta", "status": "ACTIVE","budget": {"daily_idr": 5000000,"pacing": 1.32,"cpa_14d_idr": 84000},"rule": { "id": "TARGET-CPA", "max_cpa_idr": 52000 },"brand_kit": "BK-07@v3","sources": ["insights#14d", "rule#TARGET-CPA", "kit#BK-07"]}
Typed, not guessed
Spend, results, budgets and dates become typed fields, so Friday AI compares numbers, not sentences.
Linked to sources
Every suggestion links back to the numbers and the rule that triggered it.
Versioned
Approvals are tied to the exact campaign version reviewed, so every decision can be traced.
Model routing
The right model for each task.
Claude Haiku, Sonnet and Opus, OpenAI gpt-image and DeepSeek. Each task is routed by type and plan, and your rules always have the final say.
| Step | Task | Engine | Plans | Output |
|---|---|---|---|---|
| 01 | Ad copy & chat | Claude Sonnet | All planstext | text |
| 02 | Quick drafts & labels | Claude Haiku | All planstext | text |
| 03 | Budget & policy rules | Rule engine (deterministic) | All plansdecision | decision |
| 04 | Deep campaign review | Claude Opus | Business+text | text |
| 05 | Banner images | OpenAI gpt-image | All plansimage | image |
Example routing: models are set per task and plan, and can change.
DeterministicModel
Checks & measurement
No suggestion reaches you without passing your rules.
Every draft is checked twice, by your rules and by a Friday AI review, before it reaches the approval queue. After you approve, the result is measured a week later.
W1: 88.1%, W2: 90.4%, W3: 92.2%, W4: 91%, W5: 94.8%, W6: 95.6%, W7: 93.9%, W8: 97.1%, W9: 97.8%, W10: 98.4%
Suggestion gate · CP-1121Passed
- Rules checked
- 12 / 12
- Friday AI review
- Agrees
- Budget change vs. limit
- +18%
- Policy issues found
- 0
Measured again in 7 daysOpen in queue →
Data & privacy
Your ad data stays in your workspace.
Built for agencies and brands that run on client accounts, budgets and customer lists.
- Access
Read-only AI
Friday AI reads Meta through a fixed read-only list, scoped to your workspace. It never writes.
- Crypto
Encrypted secrets
OAuth tokens and keys are encrypted with AES-256-GCM and rotated. They never appear in API responses.
- Audit
Every decision logged
Each suggestion, approval and integration change is logged per workspace and kept 365 days.
Reliability
Built for the limits of every ad platform.
GateOSrate windows · breaker · drift checks
1×Each create sent once
2Reads: MCP, then Graph
365Days of audit log
3Ad platforms, one gateway
MetaGoogleTikTok
Illustrative bars, not live data.