Intelligence Engine
Every Monday: what moved, what’s broken, what to fix first.
Five always-on lanes watching your business. One inbox. No dashboards to check.
The five lanes
Each lane has a real threshold and an action you can take in one click.
Lane 1
topic_trendsTopic Trends
What callers are asking about across the last 7 days.
Triggers when
A topic shows up 2+ times in the call signal payload.
Threshold
2+ calls in 7 days, top 5 topics ranked
Example card: "Metal roofing trending up. 4 callers this past week asked about metal roofing installs."
Primary action: Teach the agent or add a service page
Lane 2
gbp_healthGBP Health
Phone calls, website clicks, direction requests, and impressions from your Google Business Profile.
Triggers when
A metric drops 25% or more week-over-week.
Threshold
25%+ drop, prior week ≥ 10 events
Example card: "GBP phone calls dropped 65%. 12 calls last 7 days vs 34 the prior 7."
Primary action: Open Google Business Profile
Lane 3
pagespeed_regressionSite Speed Watch
Your home page’s page speed score across deploys.
Triggers when
Latest score drops by 8 or more points.
Threshold
8+ point drop, home page only
Example card: "Site speed dropped 14 points. Latest deploy scores 72 vs 86 on the prior version."
Primary action: Audit Lighthouse or roll back
Lane 4
agent_gapsAgent Gaps
Calls where your AI receptionist hedged or said "I don’t know," grouped by topic.
Triggers when
2+ calls in a week with the same topic the agent couldn’t answer.
Threshold
2+ gap events on same topic in 7 days
Example card: "Agent didn’t know Tesla wall connector. 3 callers asked about it. The agent hedged each time."
Primary action: Teach the agent (one click adds to the knowledge base)
Lane 5
competitor_watchCompetitor Watch
Tracked competitors’ rank across your grid scan points week-over-week.
Triggers when
A competitor averages 3+ rank gain across grid points.
Threshold
3+ avg rank gain, week-over-week
Example card: "Diamond HVAC moved up 4 spots. Averaged a 4-spot rank gain across 32 grid points."
Primary action: Run a fresh grid scan or audit their GBP
One inbox. Not seven dashboards.
Cards land in a single “Noticed” section at the top of your dashboard. Each card has the context you need to decide and a one-click action. Dismiss with a reason and the engine learns from the signal.
- Sorted by confidence, then recency
- Per-card evidence (the actual calls or scans the card was built from)
- Optimistic dismiss with 4 standard reasons
- Mark done or take the typed action — both close the card
Noticed
Customer demand · 70% confidence
EV charger installs trending
4 callers this past week asked about EV charger installations.
→ Add an EV Chargers page or teach the agent.
Google profile · 85% confidence
GBP phone calls dropped 65%
12 calls last 7 days vs 34 the prior 7.
→ Post a fresh photo and What’s New post.
Agent knowledge gap · 90% confidence
Agent didn’t know Tesla wall connector
3 callers asked. The agent hedged each time.
→ Teach the agent about Tesla wall connector pricing.
“The Monday email is the only marketing report I actually read. Three things, one click each, done by 10am.”
— Taurus Home Remodeling, one of the seven brands we run on AutomationFire
How it stays out of your way
Most analytics tools are too loud. The four design choices that keep this one quiet.
Quiet by default
Most days, most projects see zero cards. The engine ships a soft empty state, not a wall of low-confidence noise.
Deterministic deduplication
Each card has a stable dedup_key. When numbers update, the new card supersedes the old one. Audit trail intact.
Optimistic dismiss with reason capture
You dismiss a card and pick why (Not useful, Wrong, Already did, Not now). Those reasons feed back into prompt tuning.
Cost-bounded
Worst-case ~20 LLM calls per project per day across all 5 lanes. Most projects run at 0–5 because lanes skip cleanly when their data source is empty.
Other engines feeding signal
See your insights in 5 minutes.
Connect your business. The engine starts watching tomorrow. The first cards land Monday.