← deliverables · download

Business Plan — Project Brody

Owner: CEO (Claude Desktop). Reviewed by: CTO (Claude Code). Version: v1 — June 24, 2026. Companion file: business-plan-zh.md (Chinese). Reads alongside tech-assessment.md, market-research.md, decisions.md, CLAUDE.md.

0. Note on alignment with the CTO's assessment

The CTO's tech-assessment.md is correct that standalone Guangdong energy arbitrage loses money (−25,299 CNY over 120 days with the ML forecast; payback never; daily spread ~0.19 CNY/kWh is too thin to cover round-trip losses + degradation).

This plan does not contradict that — it changes the business model. Project Brody is not an arbitrage fund. We sell a managed transition service into Guangdong's new 15-minute spot market, capture demand-charge savings, and aggregate customers into a Virtual Power Plant (VPP). Arbitrage is, at most, late-stage upside — never the foundation. Where the CTO recommends Visayas for arbitrage economics, this plan recommends Guangdong for the transition→VPP wedge, and treats Visayas / other China provinces as later optionality. This disagreement is flagged for the founder to adjudicate (see §13 and Pending Decisions).

1. The Thesis

Guangdong is forcing its commercial & industrial (C&I) electricity users out of stable flat / time-of-use tariffs and into a market priced on 15-minute spot intervals. Most of these customers have no in-house capability to understand, hedge, or optimize against this volatility — and the retailers (售电公司) meant to shield them mostly cannot do it either.

Project Brody's wedge is to be the capability layer for this forced transition: software + managed service that lets a C&I customer move into the spot market without getting hurt. We land customers during the migration, take over the optimization of their consumption and on-site assets, then aggregate their flexible load and storage into a VPP that participates directly in energy, ancillary-service and demand-response markets.

The endgame is the VPP; the entry point is the pain of transition.

Why this beats a trading-algorithm pitch: - Regulatory urgency sells. Customers must act on a deadline, not a "nice-to-have." - Revenue starts before the hardest tech is proven. Transition advisory + bill optimization earn money without a world-beating price model. - The moat is the customer book + aggregated controllable capacity + market standing — not a better algorithm. These compound and create switching costs.

2. Market Opportunity

China's reform is abolishing the regulated retail tariff for C&I users in spot-active provinces and pushing real-time price signals to the end user. Guangdong's spot market has run officially since end-2023 and clears on 15-minute locational prices.

The migration is large and mostly not yet done — which is the opportunity:

Guangdong is also the densest C&I region on earth (Pearl River Delta), and the CTO already holds 2 years of real Guangdong spot data + 29 real C&I load profiles.

3. The Problem We Solve

Moving from a predictable tariff to 15-minute spot pricing creates problems a factory has never had to manage: - Bill uncertainty — monthly cost now swings with wholesale volatility (spikes; negative midday prices). - Contract decisions — how much to lock in medium/long-term vs leave on spot, at what ratio, with deviation penalties (偏差考核) if they get it wrong. - Operational blind spots — which loads to shift, when, and what it saves; demand charges (32 CNY/kVA/month) remain a large separate cost. - No talent — the "power-trading + storage-operations" composite team does not exist in a typical C&I business, and is scarce even among retailers.

Brody removes this burden: the customer keeps running their business; we manage their market exposure and optimize their consumption and assets.

4. Solution & Product — a Staged Journey

Stages 2–3 are where the existing codebase (BESS optimizer, backtester, market adapters, data pipeline, forecaster v1) becomes the engine. Stage 1 is a new, customer-facing product that does not yet exist in the repo (see §11) — and it is lighter to build than the full agentic trader.

5. Critical Regulatory Map (what actually gates the model)

5.1 Operating a C&I BESS — low barrier

• No electricity business license needed. Other than pumped hydro, new-type storage is not yet under electricity-business-license management.
• Required instead: local filing (属地备案) + grid interconnection (并网) + construction/design/supervision qualifications.
• EMC (合同能源管理) is the supported model: we invest/operate, share saved electricity cost with the owner.
• Red line: 转供电 (reselling power with markup) is prohibited. Commercial structure must be "share the savings," not "buy low / resell high."

5.2 Behind-the-meter storage cannot directly sell to the grid

• Pure 用户侧储能 (inside the customer meter) is designed to serve the customer's own load; in principle no reverse power flow, no power-sales settlement. Its revenue is avoided cost (buy at spot troughs, discharge to offset high-priced consumption + cut demand charge) — not selling energy.
• In a spot province, that avoided cost is spot-driven (you capture the spread on the buy side), but it is a savings, not a sale.
• To earn spot sales revenue, the asset must change market identity: either become independent storage (独立储能) (own grid interconnection; discharge as a generator, charge as a user; 报量报价 in spot) or join a VPP.

5.3 VPP requirements (the path that monetizes the book) — Guangdong rules, 2025-06-27

5.4 Direct-control on existing C&I BESS — the practical trap

• Technically possible only if the PCS/EMS exposes control (Modbus TCP / IEC-104 / MQTT). Add a smart gateway / edge controller upstream. If the battery runs on a closed manufacturer cloud EMS, we need the manufacturer to open the protocol.
• The bigger blocker is the existing manufacturer/integrator contract: EMC ownership (the integrator, not the customer, controls the battery), closed cloud EMS, warranty clauses (third-party control may void warranty), O&M contracts, and possible existing aggregation into the integrator's own VPP/DR (control and revenue already taken).
• Therefore: customer-owned + open-protocol batteries = easy to take over; manufacturer-locked = hard (partner/buy the integrator or wait for contract renewal); greenfield deployment is cleanest (we specify the EMS from day one).
• Fallback: where hard real-time direct control isn't attainable, the day-ahead-response unit participates in spot without direct dispatch control — looser coupling, works on existing/semi-controllable assets. Direct-control type → our own / fully controllable assets.

6. Revenue Model & Pricing

Progression: high-margin software (1) → asset-linked value (2) → scaled market-facing platform (3), each deepening switching costs. Indicative pricing to be set in a financial model (see CTO Questions / Pending Decisions).

7. First Customers & Go-to-Market

Prioritize by acuteness of pain × ease of landing × value of flexibility.

A. Large flexible-load C&I forced into direct participation (primary beachhead). 10kV+ / ≥5 GWh/yr sites with shiftable load: cold-chain / cold storage, ice plants, water treatment & pumping, electroplating / printing-dyeing (印染) parks, plastics injection, building materials, data centers (IDC), EV charging / battery-swap operators.

B. Industrial parks (园区) — the aggregation accelerant. One park = one sales motion → many tenants and a natural VPP node with co-located load + storage at the same clearing node. Compresses CAC and seeds Stage-3 capacity simultaneously. Highest-leverage entry point.

C. Existing C&I storage (工商储) owners. Large installed base; owners explicitly short on dispatch/trading capability and shifting from arbitrage to "asset operation." Map directly onto Stage 2 + the current codebase (subject to the §5.4 lock-in checks).

D. Small/medium 售电公司 — a B2B channel. ~265 small independents have customers + license but lack capability. White-label forecasting / risk / trading-as-a-service = fastest reach (one sale = a book). See §8 for the competitive nuance.

Recommended beachhead: PRD industrial parks + flexible-load large C&I (A+B) in Dongguan / Foshan / Shenzhen, with a parallel B2B pilot with one or two small retailers (D). The repo already holds real load profiles for Foshan, Guangzhou, Huizhou, Jiangmen, Shenzhen and Zhongshan companies — these are the target archetypes and a head start for product + targeting.

8. Competitive Landscape — and Can 售电公司 Trade?

Can retailers trade themselves? Legally yes; in practice most cannot. ~313 Guangdong retailers (~85% small independents); the industry's own diagnosis: insufficient spot price-forecasting ability, no multi-scenario strategy, no storage dispatch, exposed to deviation penalties. The published advice to them is to "build or buy" a forecasting / support system. That gap is our opening.

Retailers are three things to us at once:

Other competitors: genco-/grid-backed retailers & large energy groups (the minority building real trading teams — the serious long-term rivals); EMS / C&I storage integrators (strong hardware, weak market optimization + aggregation); global analogues (Fluence Mosaic, Habitat, Gridmatic — validate the model, not localized to Guangdong rules/data/relationships).

9. Moat

• Customer book acquired during a forced, deadline-driven transition (high intent, switching costs once we manage their position).
• Same-node controllable capacity — clustered storage + flexible load that is hard for others to assemble (the §5.3 node constraint is a barrier that protects whoever gets there first).
• Market standing — 售电 + VPP/aggregator registration; the real gate, not the algorithm.
• Localized data + compliance + EMS direct-control — proprietary metering/load data, a dynamic rule-tracking function, and a dispatch-grade control stack competitors can't quickly copy.

10. Capital Requirements & Funding Approach

• Stage 1 is capital-light (software + team + licensing) → suited to a seed round.
• Storage capex stays off our balance sheet where possible: customer-owned, or via EMC / financing-lease partners. Do not balance-sheet batteries alone.
• Use of seed funds: Stage-1 product build, the founding power-trading + storage-ops + ML team, and the 售电/VPP licensing path (build vs partner vs acquire — the acquire option needs deal capital).
• BESS hardware reference (CTO): ~$45k for 200kW/400kWh, ~$102k for 500kW/1MWh, ~$187k for 1MW/2MWh (China, 2026; LFP still falling). A full financial model (CAC, SaaS ARPU, savings-share %, storage IRR, VPP revenue per MW) is a near-term deliverable — see CTO Questions.

11. Where the Current Technology Fits

Implication: the engine for later stages is largely in place; the Stage-1 wedge product that earns first revenue is the main build gap.

12. Milestones

12-month 1. 0–3 mo: ship Stage-1 transition product MVP (bill simulation + exposure dashboard); sign 3–5 design-partner sites (parks / flexible C&I); validate forecaster vs naive baselines on real Guangdong data; prove demand-charge savings on real load profiles; scope the 售电 license path (build/partner/acquire). 2. 3–6 mo: convert pilots to paid SaaS; deploy first storage + controls under management (greenfield, open-protocol); begin demand-response participation (no 售电 qualification needed) to learn dispatch. 3. 6–12 mo: secure 售电 qualification (partner/acquire); aggregate first same-node cluster; pass the capability test; register a day-ahead-response VPP unit; first spot-market participation.

24-month 4. 12–24 mo: move to direct-control VPP on our own/controllable assets; expand to multiple parks; harden the agentic dispatch + risk layer; begin adapter-based expansion scoping for Shandong / Shanxi.

(These align with CTO timeline Phases A–D in tech-assessment.md, recast for the transition→VPP business.)

13. Key Risks & Mitigations

14. Proposed Decisions (pending CTO + founder ratification)

1. Primary market: Guangdong, on the transition→VPP model (not arbitrage). Visayas / other China provinces = later optionality.
2. Target customer: behind-the-meter C&I + industrial parks, owning the customer relationship and assets directly (not pure B2B-to-retailer).
3. Sequence: Transition/EMS SaaS → asset operation → VPP, with demand response (no license) before spot (license required).
4. Licensing: pursue 售电 qualification via partner/acquire, in parallel from Day 1.

CTO concurrence: YES on all four. Rationale:

1. Guangdong first is correct — not because arbitrage works (it doesn't), but because the transition→VPP wedge is a better business than pure trading. The CTO's earlier "lead with Visayas" recommendation was based on arbitrage economics; the CEO's model does not depend on arbitrage and is stronger. Demand-charge savings alone (6,400 CNY/month per 200 kW BESS) provide a floor that doesn't exist in Visayas.

2. Behind-the-meter C&I is correct — we own the relationship, the data, and the asset control. B2B-to-retailer would give us reach but not the controllable capacity that compounds into a VPP.

3. Sequence is correct — the demand-charge savings in Stage 1-2 require no forecasting edge and no license. Revenue starts before the hardest tech is proven, exactly as the CEO argues.

4. Partner/acquire for licensing is correct — the 售电 qualification gates VPP energy trading. A retail license + customer book is worth more than the same capital spent on software.

These four decisions are now confirmed and can be moved to decisions.md.

CTO Questions — ANSWERED

1. Forecaster vs baselines

Ran the comparison on 29 out-of-sample validation days (Guangdong DA price, last 30 days of data). Results:

The v1 ML model loses to simple persistence. The best forecast of tomorrow's Guangdong DA price is "same as the last known price." This is not an ML failure — it's a market signal: Guangdong 2024-2025 prices are so flat that persistences wins. The ML model with 72 features is overfitting to noise.

This actually supports the CEO's thesis: if naive methods already produce usable forecasts, then Stage-1 transition tools don't need a world-beating model. The ML edge will come from: (a) adding weather/system-load features not yet in v1, (b) switching to Visayas or a higher-volatility regime where persistence fails, (c) predicting not just mean but the full distribution for risk management.

Backtest profit impact: Since persistence beats ML on MAE, the ML-forecast backtest (−25,299 CNY) would be slightly worse than a persistence-forecast backtest. Both lose money because the underlying spread is too thin. The demand-charge savings (Q3) are independent of forecast quality.

CTO recommendation: Ship v1 forecaster as-is for Stage 1 (bill simulation). The MAE is adequate for exposure dashboards and what-if scenarios. For Stage 2-3 where profit depends on forecast edge, prioritize: weather features, multi-market training, and probabilistic calibration.

2. Stage-1 product reuse

What's reusable from the current codebase:

What's net-new:

MVP timeline: 4-6 weeks for a working Stage-1 product that can be demoed to design partners. This assumes one full-stack developer. The core computation (bill simulation, what-if) is already working in Python — the UI is the main new work.

CTO recommendation: Build the Stage-1 MVP in Python/Streamlit first (fastest path to demo). The backend computation reuses brody/ directly. Target: 4 weeks to a demo with Jiangmen/Foshan load profiles showing bill simulation + demand charge savings + contract what-if.

3. Demand-charge savings (real load profiles)

The optimizer's demand-charge term is active and fixed (32 CNY/kVA/month, amortized as 32/30 per day in the objective).

Measured on real C&I load profiles with a 200 kW BESS doing pure peak shaving:

Key finding: A 200 kW BESS saves exactly 6,400 CNY/month in demand charges for EVERY site (200 kW × 32 CNY/kVA). This is independent of spot arbitrage — it's pure peak clipping. Annual: 76,800 CNY/year.

Economics by site size: - Small (Foshan-class, ~200 kW avg): BESS pays back on demand charge alone in 4.3 years. Perfect beachhead. - Medium (Jiangmen-class, ~500 kW avg): Demand charge savings + load shifting = viable. Larger BESS (500 kW) captures more. - Large (Shenzhen9, ~5 MW): 200 kW BESS is too small. Need 1-2 MW to make a dent. But the absolute savings opportunity is huge.

CTO recommendation: Lead Stage-1 sales with demand-charge savings as the primary quantifiable benefit. It's guaranteed (peak shaving always works), requires no forecasting, and the math is simple enough for a factory owner to understand. Quote 6,400 CNY/month/kW-of-BESS as the floor.

4. Direct-control feasibility

Current state: The optimizer runs hourly, not sub-hourly. No dispatch loop, no setpoint tracking.

What's needed for 15-min / 5-min dispatch:

Total effort: 8-12 weeks for a dispatch-grade control stack. This is Stage-3 work and should not block Stage 1-2.

CTO recommendation: Start with day-ahead-response VPP unit (no real-time dispatch requirement, per §5.3 of the business plan). Defer direct-control to Stage 3. The day-ahead pathway needs only the 96-point forecast curve, not real-time telemetry.

5. PCS integration

Protocol landscape (Chinese C&I BESS market):

Chinese C&I BESS brands by openness:

Integration approach: For our own deployments, specify Sinexcel or Sungrow PCS with Modbus TCP from day one. The pymodbus Python library gives us read/write access to charge/discharge setpoints, SOC, and alarms in ~200 lines of code.

For existing locked BESS: Day-ahead-response VPP unit doesn't need direct control. We can still optimize and send schedules to the customer (or their EMS). The capability test (§5.3) can be passed with customer cooperation even without direct control.

CTO recommendation: Build a Modbus TCP integration module now (1 week). It's the standard interface and works with the majority of Chinese PCS. Avoid cloud-locked brands. For the Stage-1 product, direct control isn't needed — the bill simulation and what-if tools work with meter data alone.

6. VPP aggregation

What we can reuse:

Missing and effort:

Total: 6-8 weeks for a working VPP aggregation prototype.

CTO recommendation: VPP aggregation is Stage-3 work. Don't build it now. But architect the multi-site capability from the start — the optimizer's LP formulation should accept a list of batteries (not hardcode a single one). This is a small refactor now that saves a rewrite later.

7. Capacity math — thresholds

Using our 29 real C&I load profiles:

Concrete example — Dongguan node: - Take 5 Jiangmen-class factories at one node: 5 × 500 kW avg = 2,500 kW load - 15% flexible = 375 kW shiftable - Add 1 MW BESS: total controllable = 1,375 kW - Clears 1 MW spot threshold easily - For 5 MW DR: need 15-20 factories or 3-5 industrial parks at same node

Feasibility by city (from our data):

The VPP play works best in Shenzhen and Dongguan — highest factory density, many sites at the same 220kV node. One industrial park in Dongguan with 8-10 tenants easily clears both thresholds.

CTO recommendation: Target industrial parks in Dongguan / Shenzhen as VPP aggregation nodes (as the CEO already proposes in §7). The math works: 1 park = 5-10 factories = 2-5 MW controllable load = clears both spot and DR thresholds.

References

• Guangdong DRC — Southern (Guangdong-first) spot market implementation plan: http://drc.gd.gov.cn/ywtz/content/post_4273305.html
• Guangdong DRC / Energy Bureau — 2025 electricity market trading notice: http://drc.gd.gov.cn/gdsnyj/gkmlpt/content/4/4582/post_4582746.html
• Guangdong Power Exchange — VPP rules (广东交易〔2025〕152号, 2025-06-27): https://www.pvmeng.com/2025/06/27/41015/ ; full解读: https://www.gdsolar.org.cn/detail/419.html
• VPP operation management rules (试行) full text: http://editan.oss-cn-shanghai.aliyuncs.com/other/20260123/1769145326596750.pdf
• NDRC — independent storage: discharge as generator, charge as user: https://www.ideesz.com/zhdt/hyzx1/202509/t20250904_1013.html
• Guangdong independent-storage energy-market trading rules (试行): https://www.antgv.com/upload/202403/06/202403061411086404.pdf
• RMI China — 2025 power-market reform & price-system trends: https://rmi.org.cn/wp-content/uploads/2025/05/final-0508-2025电力市场化改革与电价体系洞察.pdf
• 2026 售电公司的核心竞争力在哪里？: https://www.nationalee.com/newsinfo/8852353.html
• 2026 工商储：套利时代终结，"能源资产运营"才是王道: https://www.ne21.com/news/show-227950.html
• S&P Global — China unified power market by 2030: https://www.spglobal.com/commodityinsights/en/ci/research-analysis/is-china-on-track-for-a-national-unified-power-market-by-2030.html