Entropiex

Live Today vs. Roadmap

We are not claiming that the full eleven-agent operating system is running in production today. One agent (AI Receptionist) is live and has 90 days of documented outcomes. The remaining ten agents, the Office Manager, the Shadow Mode engine, and the multi-tenant data plane are currently being built on the 12-week v1 timeline committed in the PRD. This thesis is underwritten by (a) the proven receptionist revenue recovery, (b) the methodology (Shadow Mode → Convergence → Auto-Pilot) that the platform enforces, and (c) the Harvard/INSEAD 2026 RCT that validates the AI-mapping problem Entropiex solves at acquisition.

The methodology is the moat. Entropiex OS deploys through three phases: Shadow Mode, in which AI runs in parallel alongside human decisions without executing; Convergence, in which per-workflow accuracy is tracked over a rolling window until statistical significance is reached; and Auto-Pilot, in which a specific workflow switches to AI-executed only after accuracy hits ≥95% and the business owner explicitly approves it. A competitor can copy the architecture. They cannot replicate six-to-eighteen months of accumulated convergence data tuned to a specific business's customers, geography, and job mix.

The Arbitrage

Buy at 1.5–2.5x SDE. Deploy Entropiex OS. Recover waste and capture revenue across two compounding horizons — 90 days (AI Receptionist + Rules Layer + operational waste audit = 25–45% capacity recovery) and 6–18 months (per-workflow Auto-Pilot activates as convergence is earned). Exit at 3.5–6x SDE to PE rollup buyers who pay premiums for systemized, technology-enabled operations.

Unlike the search-fund model — which relies on conventional management improvements over 6–10 year holds — Entropiex deploys a repeatable AI operating methodology with a clearly stacked value-creation mechanism:

• Days 1–90: AI Receptionist + deterministic Rules Layer + operational waste audit + owner-time recovery → 25–45% capacity recovered.
• Months 6–18: Per-workflow Auto-Pilot activates on the highest-volume workflows as each clears its convergence threshold → incremental +10–20%.
• Months 18–36: Accumulated Shadow / Auto-Pilot convergence data and autoresearch improvements compound → non-transferable moat widens; exit premium justified.

Timeline to Moat Proof

The thesis separates what is provable today from what compounds over the hold.

I.The Entropiex OS Platform

Architecture

Entropiex OS is a deployment methodology and multi-agent architecture that adapts to the specific waste profile of each acquired business.

Hub-and-Spoke Orchestration. All agents coordinate through a central Office Manager. Every event, routing decision, and confidence-threshold enforcement flows through one orchestrator. At 10+ agents, a mesh topology creates 90+ communication paths — impossible to audit. Hub-and-spoke means every decision chain is traceable to a single control point, and adding a new agent never introduces wiring complexity.

Eleven Specialized Agents

All agents share a single customer-context object. Information captured during an inbound call propagates to every downstream agent. Multiple agents execute in parallel; the customer experiences one coherent interaction.

Confidence-Gated Autonomy. Every agent action produces a confidence score from 0.0 to 1.0. Actions ≥0.95 auto-execute only once the originating workflow has reached its convergence threshold and the owner has approved Auto-Pilot for that workflow. Actions 0.70–0.94 are flagged for human review. Actions below 0.70 are rejected and escalated. AI earns operational control incrementally; humans remain the backstop until confidence is statistically proven.

Deterministic Core + AI Layer. The platform functions reliably without AI. Job creation, invoicing, scheduling — every core business operation runs deterministically regardless of agent availability. If every AI agent went dark, the business would still operate.

Circuit Breakers. Each agent runs an independent circuit breaker. Three consecutive failures open the circuit — the Office Manager stops routing to that agent and redirects work to human review until recovery. A single agent failure never cascades into system-wide degradation.

Shadow Mode → Convergence → Auto-Pilot

This is the autonomy methodology — and the mechanism that makes Entropiex OS structurally defensible.

Shadow Mode. At deployment, AI agents run in parallel without executing. Humans operate normally. AI processes the same inputs and logs its proposed decision alongside the human's actual decision — every proposed action creates a side-by-side record: AI proposal, human actual, confidence score, delta.

Convergence (per-workflow thresholds):

A standard residential quoting workflow might hit the threshold in 20–30 weeks at pilot-like volume. A complex commercial scope may remain in Shadow Mode longer. Auto-Pilot activates per-workflow, not all-or-nothing.

Auto-Pilot. When a workflow crosses its threshold, the system flags it for activation — but does not activate automatically. The business owner must explicitly approve. Once active, continuous monitoring applies: accuracy dropping below the workflow-specific revert threshold (85% for Quoting, 80% for Scheduling, 90% for Invoicing) triggers automatic reversion to human-review mode with an owner alert.

The compounding moat. Six-to-eighteen months of workflow-specific convergence data is not transferable. A competitor deploying the same architecture tomorrow faces a minimum per-workflow convergence window on every workflow they want to operate autonomously. The gap widens with every acquisition and every job processed.

The Autoresearch Loop (v2)

The Autoresearch Loop is distinct from Shadow Mode. Shadow Mode is the autonomy qualification process — it determines whether AI has earned the right to execute a workflow independently. The Autoresearch Loop is the continuous optimization engine — it generates and tests hypotheses to improve how workflows perform, whether human-executed or AI-executed.

The system generates its own hypotheses: Should the receptionist ask for the customer's email in the first 30 seconds or after qualifying the job? Should quotes include three pricing options or one? Should the follow-up text go out at hour 4 or day 2? The system designs experiments, runs them within statistically valid cohorts, and measures outcomes against KPIs from the waste audit.

Pilot data point (AI Receptionist only): On the Bay Area pilot's receptionist workflow, documented iterative prompt + scripting optimization moved quote close rates from 31% to 44% over ~90 days — not from a single breakthrough, but from a series of retained improvements. Full Autoresearch Loop automation ships in v2.

Deployment Timeline (Per Acquired Business)

II.Target Verticals

Eight verticals under active evaluation, ranked by operational waste severity, moat defensibility, and acquisition pipeline depth.

Sequencing

• Phase 1: HVAC and Electrical — deepest operator domain expertise, highest pipeline volume, most developed PE exit market. Capstone Partners tracked 149 HVAC M&A transactions in 2025, a 12.9% YoY increase.
• Phase 2: Insurance and CPA/Accounting — highest exit multiples, strong rollup demand, slightly longer convergence timelines due to regulatory complexity.
• Phase 3: Expand into behavioral health, mortgage, legal, and medical as platform matures and vertical-specific agent configurations are proven.

III.Acquisition Criteria & Playbook

Deal Criteria

Note on hold period (Tier 1): Extended from the earlier 18-month floor to 24–36 months. Per-workflow Auto-Pilot activation takes Month 7–15. PE buyers pay the premium for multiple workflows on Auto-Pilot with at least 3–6 months of post-activation performance data. A 24-month floor lets the fastest-converging workflows (Quoting, Follow-Up) contribute at least 9–12 months of compounding post-activation data before exit.

Technology Threshold Requirements

Every target must clear these bars before scoring:

• Active business phone (landline or VoIP) — required for AI Receptionist deployment
• Repeat or schedulable service work — non-recurring project work doesn't compound AI improvement
• Revenue ≥ $300K AND ≥15 jobs/week documented over trailing 12 months — below these floors, multi-workflow convergence cannot complete inside the 24-month Tier-1 hold
• ≥800 total jobs projected over 24 months across 3+ repeatable workflows
• Licensable trade or regulated practice — creates durable moat AI-native entrants cannot shortcut
• Preferred: Manual processes, basic or no FSM tools (higher AI leverage)
• Preferred: Owner-operator with no succession plan (motivated seller, better pricing)

Sourcing

1. Off-market direct outreach. CSLB license database mining, Diamond Certified research, owner demographics analysis. Personal letters to owners showing retirement signals (license expiration, sole ownership, 20+ year tenure). No broker, no auction, no competing bids.
2. Broker relationships. BizBuySell and service-business specialists. Higher priced but faster-moving deals.
3. Referral network. CPAs, attorneys, and financial advisors who know which clients are thinking about transition before a listing exists.

Underwriting Discipline

• Three years of tax returns minimum — never trust broker-presented SDE
• Monthly P&L mandatory — annual figures hide dangerous volatility
• RMO/license holder commitment in writing before closing
• Full employee disclosure including medical and key-person dependencies
• Fleet and equipment age audit — hidden capex kills returns
• Lease review with 12+ months remaining or renewal secured pre-close
• Independent verification of every material claim

The Entropiex Score™

A proprietary 100-point scoring framework:

Pillar A — Business Quality (50 points): Revenue quality and concentration, profitability and cash-flow trends, operational health, market position and reputation.

Pillar B — AI Leverage Potential (50 points): Lead-generation automation opportunity, operational efficiency gains, customer communication improvement, pricing and data-intelligence upside.

Counter-intuitive insight: low technology sophistication scores high on Pillar B. The most manual businesses offer the greatest AI transformation upside.

IV.Proof of Execution

Portfolio Company — Bay Area Electrical Contractor

A licensed C-10 electrical contractor in San Mateo County, California, is the live operating business where the AI Receptionist component of Entropiex OS has been developed and deployed. The documented 90-day outcome below is specifically attributable to the AI Receptionist workflow.

AI Receptionist — Before / After 90 Days

What This Proves vs. What It Does Not Prove

• Proves: The AI Receptionist handles real inbound calls for a licensed trade, captures after-hours leads, and compounds close-rate via iterative prompt tuning.
• Proves: A 90-day revenue step-up of the magnitude required to justify the thesis is realizable from a single workflow deployment.
• ◯Does not yet prove: End-to-end Auto-Pilot across Quoting / Scheduling / Invoicing / Follow-Up. That evidence accumulates on the 18–36 month hold.

The Platform — Building, Not Planned

Entropiex OS is a multi-vertical AI operating system under active build to v1 production in 12 weeks (PRD §10, SAD §9). PostgreSQL, Node.js/TypeScript API (Fastify + tRPC), Next.js dashboard, offline-first PWA. Eleven AI agents with hub-and-spoke orchestration, Shadow Mode autonomy engine, Stripe payments (card-present and card-not-present), and AI Receptionist integrated via a documented REST contract.

Stress-tested against real-world frictions (Day-1 mitigations in the SAD): A2P 10DLC SMS registration with transactional-email fallback (PRD §12, SAD §4.1), iOS Safari's missing Background Sync API (solved with foreground sync + manual queue, SAD §5.1.1), Stripe Terminal card-present complexity (scoped as isolated build phase in Week 7–8, SAD §9 Phase 4), and CCPA deletion across communications/recordings/photos/audit-log (runbook shipped v1, automation v2, PRD §11.5). These are the real-world complications that generic AI deployments discover at go-live. We solved them on paper before writing client-facing code.

Research validation: A Harvard/INSEAD RCT (515 firms, 2026) found the binding constraint on AI value isn't tools or capital — it's the ability to map AI systematically across business functions. Firms that did discovered 44% more use cases and generated 1.9x revenue. The Entropiex waste audit solves this at acquisition; Shadow Mode operationalizes continuous re-mapping. (Kim, Kim & Koning, INSEAD/HBS Working Paper No. 2026/20/STR)

V.The Market Opportunity

There are 36.2 million small businesses in the United States, employing 62.3 million people — 46% of the private workforce (SBA, 2026). Across our eight target verticals alone, approximately 1.8 million firms generate a combined $1.7 trillion in annual revenue. The estimated addressable market for AI services across these verticals is $30–74 billion annually.

By 2035, approximately 6 million SMBs will face ownership transitions representing $5 trillion in enterprise value (McKinsey, 2026). According to the Exit Planning Institute, 76% of business owners plan to exit within the next ten years. Fewer than 30% have a succession plan. Nearly 50% of exits are involuntary — triggered by death, disability, divorce, or economic stress.

AI adoption in production among SMBs stands at 8.8% (SBA, 2026), up from 6.3% just six months prior. Among those who have adopted AI: 91% report revenue increases, 86% see improved profit margins, and 58% save 20+ hours per month (Salesforce, 2025). Critically, 42% of small businesses report lacking the resources or expertise to deploy AI at all. That gap — between proven ROI and actual deployment — is the Entropiex opportunity.

These businesses trade at 1.5–2.5x SDE not because they lack value, but because buyers are pricing in what the seller hasn't built: no systems, no digital infrastructure, no growth trajectory, and key-person risk that exits with the owner. The discount reflects a correct assessment of the current operator's limitations — and an incorrect assumption that the gap is permanent. It is not. It is operational, solvable, and repeatable.

The window will close. As AI deployment becomes commoditized and more acquirers adopt technology-driven value creation, the arbitrage narrows. The advantage belongs to the firms that build the operating capability now, deploy it repeatedly, and accumulate proprietary performance data that compounds with every engagement.

VI.The Exit Thesis

Who Buys at the Back End

PE platform operators executing rollup strategies across fragmented verticals are the natural buyers. They pay premiums for systemized operations because systemization is what enables the rollup — they need businesses that can be integrated without the original owner.

The Multiple Expansion Math

Tier 1 Example (24–36 month hold)

• Entry: $2M revenue, $500K SDE, acquired at 1.6x SDE = $800K purchase price
• Post-Entropiex OS (24–36 months): 25–45% capacity recovered in the first 90 days; AI-captured leads increase revenue ~20%; per-workflow Auto-Pilot activates months 6–18 for another +10–20% throughput. New SDE: $750K–$900K
• Exit: Systemized operation, documented AI infrastructure, no key-person dependency. Exits at 3.5–4.5x SDE = $2.6M–$3.6M
• Return: $1.8M–$2.8M gross profit on $800K cash invested (2.3–3.5x cash-on-cash unlevered)

With SBA 7(a) financing (10–15% equity injection), an $800K acquisition requires approximately $100K–$120K in equity — pushing cash-on-cash returns to 15–23x on the upper end of the hold period.

Why PE Pays a Premium

1. Reduced integration risk. Systemized operations integrate into a platform without the original owner.
2. Documented performance data. The autoresearch loop produces a complete operational dashboard that PE diligence teams require.
3. Proven growth trajectory. 12–24 months of compounding improvement demonstrates an upward curve, not a plateau.
4. Transferable, owned AI infrastructure (internal to TMDP Capital). Shadow Mode convergence data, Auto-Pilot configurations, and accumulated workflow intelligence travel with the business at exit as documented operational history and intangible assets — not as a SaaS dependency the buyer must renew.

Exit Value Comes From the Businesses, Not From Software

Entropiex OS is purpose-built for TMDP Capital portfolio companies and is never offered as an external SaaS product. There is no self-serve onboarding, no white-label, no partner API, no external tenant. Exit returns come from selling the systemized portfolio businesses to PE rollup buyers at multiple-expanded EBITDA — not from licensing the OS.

QuickBooks Posture

QBO is treated as a lightweight bridge, not a strategic dependency. v1 ships no QBO integration (owner re-keys ~10–20 invoices/month); v2 ships a one-way QBO export for portfolio accountants. Long-term, Entropiex OS may replace QBO entirely with a native financial reporting layer that reads directly from the operational tables (invoices, payments, expenses, audit_log).

VII.Why Now

1. Demographic pressure. The average business owner is 57. The wave is cresting now. Early movers have their pick of the best assets at the lowest multiples before sophisticated acquirers crowd the market.
2. Technology readiness. Multi-agent AI systems capable of operating a service business end-to-end became viable in early 2026. The technology exists, the deployment methodology is specified, and the acquisition vehicle is the differentiator.
3. Market ignorance. The broker community, the SBA lending community, and the traditional search-fund ecosystem have not yet priced AI-driven value creation into their models. Deals are still underwritten on historical performance, not transformed potential — the informational asymmetry that generates outsized returns.
4. The vendor gap. No service-business operating platform offers native AI autonomy. Commodity software AI is bolt-on chat widgets and email drafting — not agents that run operations within a confidence-gated autonomy framework. That gap closes — and the firms that built it now will define the default infrastructure for transformed service businesses at exit.

VIII.The Team

Between us, we have deployed hundreds of millions of dollars in capital expenditure programs across Fortune 500 manufacturing and infrastructure operations. We have sold and delivered hundreds of millions in AI transformation engagements to C-suite buyers worldwide. We have architected and shipped production AI systems — software and hardware — at one of the world's largest cloud platforms.

We took all of it and applied it to a real business. We deployed the AI Receptionist component of Entropiex OS on a Bay Area C-10 electrical contractor and documented 90 days of measurable operational lift. We are now building the remaining platform to v1 production on a 12-week track and will deploy it into every business we acquire.

We are not theorists. We are not consultants. We built the AI Receptionist, proved it in production, and are now building — on a committed timeline, with a documented PRD and SAD — the multi-agent operating system that deploys exclusively into TMDP Capital portfolio businesses.