Copilot
How to use it
- Ask the copilot a question — "why did Meta drop?", "should I scale TikTok?". (Orchestration → Copilot)
- Get a structured answer with supporting data and a recommended action.
- Act on it — or ask a follow-up.
⏱ 5-minute decisions · 💳 Pro · 🎯 Answers grounded in your real data
Why this matters for your business
Most ad data is available but not legible. You can stare at a Meta dashboard for an hour and still not know whether ROAS is dropping because of audience saturation, creative fatigue, or a competitor outbidding you. The dashboards show numbers; they don't explain them. The investigation that should take 5 minutes becomes a 2-day Slack thread, and the decision often gets made on intuition because the data wasn't legible enough to ground it.
Copilot is the conversational interface that translates "I have a question" into "here's the answer with the reasoning." Built on top of unified attribution + Customer 360 + ad platform data, it can answer questions across the whole stack: "why did ROAS drop?", "what's the best budget for next month?", "is our creative tired?", "should we expand to a new platform?" The answers come with the supporting data and a recommended action.
The unlock isn't just speed — it's making strategic ad decisions legible to non-specialists. A founder without an ad-ops background can ask the copilot the same questions they'd ask a $200K/year ad strategist and get answers that are 80% as good. That's the leverage.
What this typically unlocks
| Outcome | Result |
|---|---|
| Time to answer "why did X happen?" | 30 seconds vs. hours |
| Strategic ad decisions made by non-specialists | possible with copilot |
| Hours/week on dashboard analysis | −6h typical |
| Confidence in answer to "what should we do?" | high — grounded in data |
| Onboarding new team member to ad strategy | days vs. weeks |
What you actually get
A chat interface scoped to your store's data:
| Capability | Description |
|---|---|
| Diagnostic Q | "Why did ROAS drop on Meta this week?" |
| Recommendation Q | "Where should I shift $5K next week?" |
| What-if Q | "If I 2× the TikTok budget, what happens?" |
| Comparison Q | "How does our acquisition CAC compare to industry?" |
| Forecast Q | "How much revenue will Q3 ads generate at current pace?" |
| Threading | Conversation context preserved within a session |
| Source citations | Every answer cites the data points it used |
| Action handoff | Recommendation → click-to-execute (handoff to budget optimizer / campaign plan) |
How it works
Read-only by default
Copilot answers and recommends but doesn't do. Recommendations hand off to the budget optimizer or campaign-plan editor, where human approval is required before action.
This is intentional. The trust contract for an AI is much easier when "the AI talks; the human acts." Action-taking AI without explicit handoff has a much higher accidental-blast- radius.
What it knows
- Every campaign on every connected ad platform (live + historical)
- Every audience definition + saturation status
- Every creative + its performance across platforms
- Customer 360 + predictive LTV (for audience valuation)
- Anomaly detector signals (for "what changed?" questions)
- Industry benchmarks (where available, with disclaimers)
What it does NOT know
- Your competitive positioning beyond what's visible in ad data
- Your organizational priorities (must explicitly tell it)
- Things outside the platform (your CRM, your in-store stuff unless connected)
It will say "I don't have data on that" rather than guess.
Real merchant scenarios
Scenario A — Diagnose ROAS drop in 30 seconds
Question: "Why did Meta ROAS drop this week vs. last week?"
Answer (paraphrased):
Meta ROAS dropped from 4.6× to 3.4× (−26%). The drop is concentrated in the lookalike-1%-US audience: CPM up 22% (auction got more competitive), CTR down 14% (creative may be tiring — has been live 5 weeks), and conversion rate stable. The other 3 Meta campaigns are at-baseline.
Recommended actions:
- Refresh creative on the lookalike-1%-US campaign (top fatigue signal)
- Or: throttle that campaign by 30% and shift to the Google search campaign (which is over-target)
Cited data: Meta Ads ROAS, audience auction insights, CTR trends 2026-05-03 → 2026-05-10.
Time: 4 seconds. Action confidence: high — backed by specific numbers.
Scenario B — Founder asks for next month's budget plan
Question: "What budget should I run next month? Total $25K. Goals: keep ROAS above 4×, drive new customer acquisition."
Answer (paraphrased):
Suggested allocation: $25K, distributed as:
Platform Spend Why Meta acquisition (lookalike + interest) $11K Highest historical ROAS for new-customer goal at this scale Google search (branded + category) $8K Steady ROAS 4.2× last 90d, good headroom TikTok (Spark Ads + brand) $4K Lower ROAS 2.8× but high incremental value (uncovered audience) Reserve for tests $2K Small new-channel test — recommend Pinterest catalog Projected outcome: ~620 new customers, CAC $40, ROAS 4.2× at 90d (using predicted LTV).
Risk: Meta lookalike approaching saturation; if CPM rises
15%, recommend shifting $2K to TikTok.
Cited data: 90d ROAS by platform, predicted LTV cohorts, current audience overlap.
Hand off this plan to the campaign-plan editor?
Founder clicks "Yes." Plan opens in editor for review; they can tweak then approve.
Scenario C — What-if simulation
Question: "What if I double TikTok spend from $5K to $10K?"
Answer (paraphrased):
Doubling TikTok from $5K to $10K (taking the extra $5K from the reserve / unspent buffer):
Projected outcomes (range):
- Customers: +85 to +140 (current TikTok audience can absorb this scale; CPM expected to rise ~12%)
- Incremental CAC: $36-58 (vs. current $35; modest worsening)
- Total ROAS contribution: +$8K to +$14K revenue
Risk factors:
- TikTok creative library is thin (3 videos); doubling spend may accelerate creative fatigue
- Lookalike audience saturation likely in 2-3 weeks at doubled spend
Recommendation: try $7.5K first (50% increase) for 2 weeks, evaluate, then decide on further scale.
Scenario D — Comparison with industry
Question: "How does our acquisition CAC compare to other beauty brands our size?"
Answer (paraphrased):
Your current acquisition CAC: $38 (last 90d). Industry benchmark for beauty DTC at $5-15M revenue: $32-$52 (interquartile range).
You're slightly better than median ($42). Specifically:
- Your Meta CAC ($35) is below industry median ($45) — your creative or audience is winning
- Your Google CAC ($28) is significantly below industry ($38) — likely your branded search is strong
- Your TikTok CAC ($32) is below industry ($40) — early-mover advantage probably helping
Caveat: industry benchmarks are aggregate; your individual mix of products / audience / season is different. Use as directional, not absolute.
Scenario E — Onboarding a new marketing hire
Setup. Brand hires a new marketing manager. Doesn't know the brand's ad history.
First-day questions to copilot:
- "Summarize our ad performance last 90 days"
- "Which campaigns drove the most revenue?"
- "What audiences are we currently targeting?"
- "What's our typical creative cadence?"
Each answered in seconds with citations. Marketing manager ramped to "informed enough to start contributing" in ~1 hour vs. ~2 days of dashboard archaeology.
Best practices
✅ Ask specific questions. "Why did ROAS drop?" gets a better answer than "what's wrong?"
✅ Use threads for related questions. Follow-ups stay in context: "and what about the previous month?"
✅ Verify recommendations against your domain knowledge. Copilot is grounded in data but doesn't know your context (seasonal, competitive, brand).
✅ Use the action handoff. When the recommendation is right, hand off to the optimizer/plan editor — don't manually re-implement.
❌ Don't treat copilot as a replacement for strategic thinking. It augments human decisions; it doesn't replace them.
❌ Don't ask copilot questions about external data it doesn't have. It'll say "I don't know" instead of hallucinating, but you'll waste time on dead-end Q's.
❌ Don't action recommendations without reading the reasoning. The recommendation can be right; the underlying data context might still need a human eye.
Plan tiers
| Capability | Free | Starter | Pro | Agency | Enterprise |
|---|---|---|---|---|---|
| Diagnostic + recommendation Q | — | — | ✓ | ✓ | ✓ |
| What-if simulations | — | — | ✓ | ✓ | ✓ |
| Industry benchmarks | — | — | ✓ | ✓ | ✓ |
| Conversation threading | — | — | ✓ | ✓ | ✓ |
| Action handoff to optimizer | — | — | ✓ | ✓ | ✓ |
| Multi-brand context (agency) | — | — | — | ✓ | ✓ |
| Custom data sources | — | — | — | — | ✓ |
| Conversation export / audit | — | — | — | ✓ | ✓ |
See also
- Orchestration overview
- Campaign plans — copilot hands off to here
- Budget optimizer — copilot hands off to here
- Anomaly detection — feeds copilot's "what changed?" answers