Ada is BOT's on-premise optimization intelligence. Every morning she reads the weather and the ERCOT 4CP risk, plans the building day, drives the Atenea BAS — and then checks her own work against an IPMVP Option C baseline, interval by interval. Not a dashboard. A loop that closes.
Commercial buildings burn roughly 40% of U.S. electricity, and most of the software sold to fix that shares one structural flaw: it grades its own homework once a year, if at all. The standard arrangement is an optimization platform on subscription — and, separately, a measurement-and-verification study that costs $10,000–25,000 and runs annually. Twelve months is a long time in a building. Schedules revert after holidays. Overrides stick. An economizer fails part-open in April; nobody prices it until next spring's report.
We came at this problem as building owners, not as a software company. Our portfolio — 4 buildings, roughly 421,000 SF, owner-operated — has held operating expenses below CPI for 19+ years. When you pay the utility bill yourself, a savings claim you cannot defend is worth exactly nothing. So we built the loop we wished we could buy, ran it on our own buildings first, and only then started licensing it. How the measures stack is covered in our guide to HVAC optimization for commercial buildings.
Ada is the head of the loop — the daily optimization brain. Here is her working day, the same one she runs on our buildings and on licensed ones.
Weather forecast, building telemetry, and the day's ERCOT 4CP risk — the four summer peaks that price a building's transmission for the following year.
Start times, setpoints, pre-cool depth, demand budget, curtailment windows — one operating strategy per building, set fresh each morning rather than copied from last Tuesday.
The plan lands at the equipment through the Atenea BAS: schedules, pre-cooling, staged starts, curtailment. It runs on edge microcontrollers with zero licensing fees.
Actual load is scored against a weather-normalized IPMVP Option C baseline at 15-minute resolution, R² > 0.96. Drift is flagged the interval it opens.
Two things about that last step matter more than the AI. First, the cadence: most M&V in this industry is a once-a-year study; ours runs automatically every 15 minutes. Second, the honesty: Ada's plan and its verified outcome live in the same record. If a strategy underperforms, the baseline says so within the day — not in next year's report.
And to be clear about what Ada is not: an unattended black box. She is human-on-the-loop — she plans and drives, engineers review her work, and every action she takes lands in the verified M&V record. She also runs on-premise, so portfolio data stays under owner control instead of in someone else's cloud. If the 4CP mechanics are new to you, we wrote them up plainly in ERCOT 4CP, explained.
The savings from AI energy optimization are real, but in the conventional stack they leak straight back out through licensing and consulting fees. We built both layers ourselves — the BAS and the verification — so they don't.
| Layer | Typical commercial stack | The BOT loop |
|---|---|---|
| BAS licensing | Proprietary building automation with recurring licensing — landlords commonly pay $50–200K per year. | Atenea BAS on edge microcontrollers, zero licensing — we built it ourselves for under $50 per controller. |
| M&V | A $10–25K engineering study, once a year. Savings drift for months before anyone measures it. | Continuous AI M&V, every 15 minutes, IPMVP Option C, R² > 0.96 — no study to commission, ever. |
| Optimization | Static schedules and setpoints, tuned at commissioning and left to decay. | Ada re-plans every morning from weather and ERCOT 4CP risk, human-on-the-loop. |
| Data residency | Vendor cloud; your interval data becomes their asset. | On-premise. The building's data stays in the building, under owner control. |
| Skin in the game | The consultant's fee clears whether the savings persist or not. | Proven first on our own 4 buildings, ~421,000 SF — we ran it on our money before asking for yours. |
Choosing a platform more broadly? Our EMIS buyer's guide covers the whole category.
Every figure below comes out of the M&V record or from utility bills. Qualifiers are stated because they matter.
The one worth dwelling on is 11200 Richmond: we acquired it in September 2025, put it on the loop, and roughly eight months later it is running 22.4% below its verified baseline year-to-date. The demand story compounds it: peak demand at one building fell 36%, from 372 kW to 239 kW, which cuts the distribution charge every single month. (Why demand is its own game: our demand-charge guide.)
The long arc shows the same shape. 800 Wilcrest — our flagship — cut annual energy roughly 50% between 2015 and 2025 (time-normalized, not weather-normalized), then dropped a further ~25% year-over-year after our own BAS beta went in around March 2025. And the verification engine itself runs well beyond our walls: the RMS platform serves ~600 K-12 schools plus Harris County, and across 15 buildings it carries 44 validated M&V reports. For owners who need the carbon side documented, the same IPMVP-grade record feeds ESG reporting without a second data pipeline.
That's the whole ask. One bill starts a free RMS trial — we'll show you what's hiding in your interval data, and what a verified daily loop would hold onto. No capital required to start; conservation begins with a schedule.