Researched by BEAST Library Curator | Verified by Harper | Quality: 7.6/10

BLUF: Your AI-HVAC or BaaS vendor's "20% savings" pitch is not a number until it is measured against a contractually-specified baseline using a named IPMVP option. Most 2026 vendor contracts still skip this step — which means the vendor grades its own homework. Here is the exact clause language, the ASHRAE Guideline 14 pass/fail thresholds, and the 90-day path to put it in your next RFP.

The Gap Nobody Names in the RFP

Here's what I'd do if this were my building: before signing anything with an AI-HVAC or building-OS vendor, I'd stop asking "how much will this save?" and start asking "which IPMVP option, and what CV(RMSE) does your baseline model hit?" Almost no vendor gets asked that second question, and it shows — most performance contracts I've reviewed name a savings percentage and a monitoring dashboard, but never name a measurement protocol. That is the gap this report closes.

The tools to fix this already exist and are free. The International Performance Measurement and Verification Protocol (IPMVP), maintained by the Efficiency Valuation Organization, defines the standard terms for quantifying efficiency-project savings and is the reference protocol utilities and government agencies already use for demand-side management programs. ASHRAE Guideline 14 (current version 2023) sets the pass/fail statistical bar for the baseline models underneath those savings claims. The problem isn't a missing standard — it's a missing clause. We've written before about the trust gap this creates: our look at the 2026 AI-HVAC verification gap found the same missing-protocol pattern at the chiller-plant level; this report is the standards-based fix.

Pick the Right IPMVP Option for an AI-Driven Retrofit

IPMVP defines four measurement options, and the right one depends on what the AI system is actually doing to your building:

For most 2026 AI-HVAC deployments — where the vendor is retrofitting control logic onto existing chillers, boilers, and AHUs — Option B for single-system pilots and Option C for whole-building rollouts is the pairing worth specifying by name in the contract, not left to the vendor's discretion.

The Statistical Bar: ASHRAE Guideline 14 Calibration Criteria

Naming an option isn't enough — the baseline model built under that option has to clear a numeric bar, or "savings" is just curve-fitting. ASHRAE Guideline 14 sets the goodness-of-fit thresholds a baseline model must hit before its savings estimate is considered valid:

Data interval / duration Max NMBE Max CV(RMSE) — energy Max CV(RMSE) — demand
Monthly data±5%15%—
Hourly data±10%25–30%—
<12 months post-retrofit data—20%30%
12–60 months post-retrofit data—25%35%
>60 months post-retrofit data—30%40%

NMBE = Normalized Mean Bias Error; CV(RMSE) = Coefficient of Variation of the Root Mean Squared Error, per ASHRAE Guideline 14-2014 calibration criteria.

Ask your vendor for the CV(RMSE) of their baseline regression before you sign. If they can't produce that number, they haven't built a defensible baseline — they've built a marketing number.

The Alternative Path: NMEC (If Your Utility Offers It)

If your building sits in a jurisdiction with a pay-for-performance utility program, Normalized Metered Energy Consumption (NMEC) is worth checking before you build a custom M&V plan from scratch. NMEC emerged in California from AB 802 and SB 350, which together authorized utilities to pay for efficiency based on actual whole-building meter data rather than deemed, prescriptive savings estimates. The mechanism is the same core idea as IPMVP Option C — a statistically-screened baseline model against whole-building interval data — but it comes with a pre-built utility incentive check attached. The catch: NMEC pre-screening rejects buildings whose energy use is too irregular to model accurately (a building with erratic, unpredictable occupancy may not qualify), so ask your utility program administrator for the pre-screening pass rate before you count on this path for a specific building.

Close the Non-Routine Adjustment Loophole

Even with an Option and a CV(RMSE) threshold specified, there is one clause vendors will still try to leave vague: non-routine adjustments. Per Guideline 14, the core savings equation is:

Savings = (Baseline Energy − Performance Period Energy) ± Routine Adjustments ± Non-Routine Adjustments

Non-routine adjustments account for genuine changes since baseline — a new tenant, an added floor, a changed operating schedule — and they are legitimate. But they are also the exact mechanism a vendor can lean on to explain away a savings shortfall ("the building's occupancy changed, so we adjusted the baseline upward"). The fix is not to ban non-routine adjustments — it's to require that every one be documented, dated, and jointly signed off before it's applied, not proposed retroactively at the annual reconciliation meeting. Put that sign-off requirement in the contract now; it costs nothing and it closes the loophole before the first dispute.

The Clause, Written Out

Here is contract language a facility GM can hand directly to procurement or legal for review:

"Vendor shall report all claimed energy savings under [IPMVP Option B / Option C], with the baseline regression model meeting ASHRAE Guideline 14-2023 calibration criteria (NMBE and CV(RMSE) thresholds per the applicable data interval and post-installation duration). Any non-routine adjustment to the baseline shall be documented with supporting evidence and jointly approved in writing by Owner and Vendor prior to application to the savings calculation. Vendor shall provide the baseline model's CV(RMSE) and NMBE values to Owner upon request and at each reporting period."

90-Day Action Plan

  1. Weeks 1–2: Pull your current AI-HVAC / BaaS contract and check whether it names an IPMVP option or any CV(RMSE) threshold. Most don't — that's your finding, not a failure on your part.
  2. Weeks 3–6: Ask your vendor for the baseline model's current CV(RMSE) and NMBE. If they can't produce it, that is itself the answer to whether their savings claim is verifiable.
  3. Weeks 7–10: Check with your utility account rep whether an NMEC pay-for-performance program exists in your territory — it may fund the M&V work you're about to require anyway.
  4. Weeks 11–13: Route the clause language above through procurement/legal for your next renewal or new vendor RFP.

None of this requires new instrumentation if your BMS already trends the relevant points — it requires naming a protocol in the contract that both sides already have the data to satisfy.

For more practitioner-grade breakdowns like this one, browse the rest of our Library.

Have a question about this topic? Ask our CRE AI Agent →