What runs in your head — risk, coach quality, the next leveraged move — runs in the system. 7 venues. 6,000 kids. 150 coaches. Continuously.
You manually combine 5 reports every Monday. Delta Intelligence delivers the same brief at 6am — plus the actions, the evidence, the next experiment — before you've made coffee.
Each stream watches a different lever of the business. The shared decision layer connects them — a coach issue informs a child-at-risk score; a schedule change informs an acquisition spend.
Which kid is about to leave, and why?
Right kids, right classes, right times?
Thriving, struggling, or red-flagged — beyond just NPS?
Where do enrolments actually come from?
We don't ship 11 agents at once — that's the trap. Phase 1 is five concrete deliverables that free John's Monday, halt the biggest revenue leak, and lay the foundation for the spin-out IP.
John's 5-report manual ritual replaced by an AI-generated weekly briefing. What changed, what's trending, what needs attention — delivered at 6am.
Rules-based scoring (missed classes, sibling cancelled, NPS drop, coach change, drive time). Each Orange/Red flag fires a per-family investigation: tailored intervention with a named coach.
Each centre manager gets a Monday action list, generated per venue: the 5 most leveraged moves this week — in language a 19yo team leader can act on, not a 40-page report.
UTM tagging + Meta Conversions API + CallRail dynamic numbers. Closes the "anonymous visitor → phone call → enrolment" gap. Phase 2 layers Data Cloud multi-touch resolution.
Extracts John's tacit knowledge from his Loom recordings + decision history. Over time becomes "what would John do here?" — the defensible IP at the heart of Delta's spin-out.
Honest read: 11 agent builds is too ambitious for one phase. T1, T3+A1, T4, T5, A5 is the realistic Phase 1 list. Everything else moves to Phase 2+.
Once Phase 1 is live and the team trusts the system, we layer in pattern discovery, coach development, schedule optimisation, and the cross-business referral synergy.
Weekly per-venue. Real search over schedule reshuffles that capture waitlisted kids. Respects coach availability, room capacity, parent disruption. Top 3 ranked proposals.
Investigates each detractor for root cause — coach? facility? schedule? competitor? Weekly synthesis across all venues: "what's actually going wrong this week."
Cross-business. Delta Gym observations (motor delay markers, attendance patterns, NPS verbatims) → candidate referrals to Progress Pathways. Safeguarded handoff to a named human.
Pattern discovery done right: agent proposes, Python statistical code validates (significance tests, effect size, multiple-hypothesis correction), agent only explains the patterns that pass. No sycophancy.
Per-coach development plans. Themes verbatim NPS → benchmarks vs peers → identifies 1-2 specific behaviours → micro-coaching plan → measures post-intervention. Monthly 1:1 prep doc.
Consolidation note: A1 / A4 share the per-case investigator engine — same architecture, different triggers and tools. A3 Chief of Staff is the meta-layer that consumes outputs from everything else.
Connects directly to Tribe and the existing Salesforce stack. No separate login. No CSV exports. Pulls live data, runs investigations, proposes actions, and creates Tasks in Tribe — exactly where your team already works.
Phase 1 lives in Tribe. Custom Objects (AI_Recommendation, AI_Risk_Score), Flows for triggers, Tasks for actions. The team works where they already work — no separate login, no new tool.
Decoupled brain via Apex callout. Tribe queues the work; external Python service processes; results write back as records. Avoids Apex's 120s / 100-callout limits without leaving the ecosystem.
Phase 2: lift & shift. Clean REST API between Tribe and the brain means the brain is already standalone. When Delta outgrows Tribe — or productises as Performance Delta — Tribe becomes one consumer, not the host. No rebuild.
Two integration users (AI_ReadOnly + AI_Writer). Prompt caching mandatory. Haiku/Sonnet/Opus model routing per task complexity.
6,000 kids' PII leaving Salesforce to external Python and Anthropic needs compliance review. State privacy laws vary. AI_Writer integration user has zero read access to core child objects.
NPS at 25% from 4 responses isn't a red flag. The Rihanna Beckley problem. Bayesian sequential testing distinguishes Confident Red Flag from Needs More Data — not arbitrary timers.
$2/conversation at parent volume = $6–10K/mo. Self-service FAQ escalation logic and conversation budget caps are mandatory before turning anything on.
Without controls, $50–68K/mo is possible. Hard contract cap of $8K/mo with auto-shutoff. Hard loop limits (max 50 per investigation, 200/day always-on). Human gate on irreversible actions.
Each step of each stream is handled by a specialist agent — purpose-built, with its own tools, connected to live infrastructure. Humans set objectives and gate the irreversible. Agents do the heavy lifting.
What you build at Delta becomes the product for the next 50 venues. Because Phase 1 was Tribe-native but portable from day one, Phase 4 isn't a rebuild — it's a packaging exercise. US PE rollups acquiring $50–150M gym chains need this on day one. The first chain on the network owns the benchmark data for the entire vertical — and Delta gets there first.
Performance Delta exposes its own intelligence — churn scores, class optimisation, coach performance, cash projections — as MCP tools that external agents call directly. PE board decks, internal AI agents, ops dashboards all read from Performance Delta. Switching cost: everything else breaks if you do.
Every chain on Performance Delta makes every model smarter for every other chain. Churn prediction trained on hundreds of thousands of children, not 6,000. New customers get years of accumulated intelligence on day one. Competitors can't catch up without the install base.
Build the agent logic once. Render natively wherever the team already is — WhatsApp, Slack, Teams, email, parent app. A 19yo team leader is on WhatsApp. A regional manager is on Teams. Same engine, every surface. Removes the #1 enterprise sales objection.
Cross every coach's pay grade against revenue per child-hour they retain. "Coach A costs 20% more, generates 45% more retained revenue." For PE running 50+ venues, that's workforce optimisation at scale — a P&L tool. Different audience, different price point, different deal size.
Autonomously onboards newly-acquired venues by mapping Mindbody / Jackrabbit / others to Tribe schema. 60-day integration vs 2 years of parallel systems — the killer PE-buyer feature.
Per-PE-customer: continuously benchmarks each venue against the rollup network + anonymised cross-customer data. "Here's what your top quartile does differently." Recurring revenue per tenant.
Sold to PE acquirers, not operators. Autonomously conducts operational diligence on prospective targets. Different audience, different price point — 7-figure deals.
The judgment, the data, the 30+ years of Delta heuristics — no one else has what it takes to make this work. Let's build it.