Skip to main content

Decision Intelligence (DIE)

Why this matters for your business​

Operators don't have a data problem. They have a decision problem. Most stores have more dashboards than time to read them; the hard part is figuring out which signal to act on this week. The biggest opportunity might be hiding in a dashboard nobody opened. The biggest problem might be obscured by 50 metrics that all moved a little.

Decision Intelligence (DIE) is the layer that does this for you. Every hour, it ingests every signal across the platform — sales, attribution, predicted LTV, anomaly status, journey performance, ad ROAS, customer feedback, inventory — scores each signal for trust (is the data quality good?) and bias (is something unusual skewing it?), then ranks the actions that would have the biggest impact this week. Each recommendation comes with explainability ("this is the reasoning") and a one-click apply to the right engine (launch a campaign, adjust a journey, rebalance ad spend).

The result: operators stop looking for what to do and start deciding what to do. The most leveraged minute of the week becomes the one spent reading DIE's output.

What this typically unlocks​

OutcomeResult
Time to "what to do this week"30 seconds vs. hours
Recommendations actioned60-75% typical
Revenue impact per actioned rec$3-30K typical
Decisions that compound (acted, then forgotten)near 0 — DIE remembers
Founder/operator strategic confidence+ measurably

What you actually get​

Three layers:

LayerDescription
Signal aggregationIngests every metric from every engine; one source of truth
Trust + bias scoringMarks low-quality data; flags biased samples
Ranked recommendationsTop 5-10 actions per week with explainability + apply button

Recommendation example​

Recommendation #1 (Impact: high; Confidence: 0.92):

Issue: Cart-recovery email's open rate dropped 28% over last
3 weeks. Mostly concentrated in Apple Mail Privacy Protection
opens (which stopped reporting accurately).

Underlying cause: Shift in your audience composition — 14%
more iOS visitors quarter-over-quarter from the Meta
campaign launched 6 weeks ago.

Recommended action:
- Switch cart-recovery primary KPI from open rate to
  click + conversion rate (these aren't affected by APP)
- Adjust threshold for "successful campaign" accordingly

Estimated impact:
- More accurate measurement of cart-recovery effectiveness
- Avoid mis-throttling the campaign (currently throttled
  due to "low" open rate signal)

[Apply to cart-recovery journey →]   [Dismiss]

You can apply with one click, save for review, or dismiss with optional reason. DIE learns from your responses (rec acceptance rate, dismissal reasons) and improves over time.

Trust + bias scoring​

Every recommendation comes with:

  • Trust score (0-1): Is the underlying data reliable? Low scores happen when sample size is small, signal noisy, or recent data quality issues.
  • Bias indicator: Is the signal being skewed by something unusual? (e.g. one $50K order in a 100-customer sample inflates AOV)
  • Counter-evidence: What signal against this recommendation also exists? Avoids over-confidence.

How it works​

Signal sources​

DIE reads from:

  • Sales engine (Customer 360, segments, journeys, campaigns, attribution, anomalies, predictive LTV)
  • Orchestration (ad spend, ROAS, audience overlap)
  • Content (blog performance, ad copy CTR, SEO)
  • Communications (channel engagement, fatigue posture, deliverability)
  • Ops (Shopify orders, refunds, inventory, support volume)

Recommendation cadence​

  • Weekly digest (Monday 7am): top 5-10 actions
  • In-app (always available): live ranked list
  • Real-time anomaly-driven (P1 only): immediate
  • Quarterly review (every 90 days): strategic posture shifts

Real merchant scenarios​

Scenario A — Founder discovers cart-recovery should change​

Setup. Founder reading Monday DIE digest. Top recommendation: "Cart-recovery email open rate dropped 28%. Likely cause: Apple Mail Privacy Protection. Switch primary KPI."

Action: Clicked apply. Cart-recovery journey now uses click + conversion as primary metric. Throttling rules adjusted.

Result over 30 days: Cart-recovery still recovering carts at historical rate — confirmed not a real performance drop. Avoided incorrectly killing a working journey.

Scenario B — Mid-market catches over-fatigue early​

Setup. DIE flagged: "37% of customer base at fatigue ≥ 0.6. WhatsApp campaigns scheduled this week will reach customers already over the cap — net audience: 12,400 (was 28,000)."

Action: Marketing manager reduced WhatsApp campaign ambitions; instead activated a tiered audience approach (top half by engagement only).

Result: WhatsApp ROI went up 38% on smaller send volume. Without DIE's flag, would have over-sent and degraded list further.

Scenario C — Agency uses DIE to brief weekly client calls​

Setup. Agency lead opens each client's DIE before the weekly call.

Result: Calls go from "what should we do?" debate to "DIE recommends X — let's discuss whether to do it." Calls 30 min shorter; decision quality higher.

Scenario D — Catching a bias​

Setup. DIE recommendation: "Predicted LTV cohort #3 underperforming — consider lower retargeting budget."

Bias flag: "Note — 60% of this cohort's recent purchases came from one promotional moment 21 days ago. Cohort behavior may revert to historical norms over next 30 days."

Action: Founder waited 30 days. Cohort recovered. Without the bias flag, would have prematurely cut budget.

Best practices​

✅ Read DIE before opening individual dashboards. It's the pre-digest; dashboards are the deep-dive.

✅ Apply or dismiss every recommendation. Don't ignore; the engine learns from your feedback.

✅ Trust the bias indicators. Most over-confidence comes from missing the counter-evidence.

✅ Set Slack integration for high-confidence recs. P1+ on Slack is more actionable than email.

❌ Don't auto-apply DIE recommendations. They're recommendations, not decisions. Human judgement adds the context DIE doesn't have (brand voice, organizational priorities).

❌ Don't ignore low-trust recommendations entirely. They still surface real signals; just verify before acting.

Plan tiers​

CapabilityFreeStarterProAgencyEnterprise
DIE in-app + digest——✓✓✓
Trust + bias scoring——✓✓✓
Explainability——✓✓✓
One-click apply——✓✓✓
Multi-shop DIE roll-up———✓✓
Custom DIE rules————✓
Slack integration——✓✓✓

See also​