AEE CMVP common mistakes
AEE CMVP · Understand the traps and how to avoid them.
Treating the baseline as simply 'last year's utility bills' and adjusting only for weather, rather than adjusting for all relevant routine and non-routine changes (occupancy, schedule, throughput, equipment additions).
Candidates anchor on the familiar weather-normalization workflow from utility billing analysis and underweight the broader concept that the baseline must be adjusted for any measurable, verifiable change in conditions affecting energy use.
What to do instead
Before computing anything, enumerate every independent variable that drives load and classify each change as routine (expected, modeled) or non-routine (one-time, handled by special adjustment). Build the adjustment plan from that inventory, not from the bill history alone.
Applying a non-routine adjustment to a change that was actually routine (or vice versa), for example retroactively adjusting the baseline for a production increase that was forecastable, or failing to adjust for a one-time plant shutdown.
The routine/non-routine distinction feels like semantics under exam time pressure, so candidates assign adjustments by which is easier to compute rather than by the definition.
What to do instead
Use the test: was the change expected and quantifiable at the time the M&V plan was written? If yes, it belongs in the routine adjustment model. If it is a one-time or unanticipated change, it requires an explicit special non-routine adjustment documented in the report.
Choosing whole-facility (Option C) measurement when savings are a small fraction of total facility consumption, or choosing retrofit isolation when interactive effects dominate.
Candidates pick the option that sounds most rigorous or most familiar instead of matching the option to the savings-to-noise ratio and the presence of interactive effects.
What to do instead
Screen options against two criteria first: expected savings as a percentage of the measured quantity (small percentages push you toward isolation or submetering) and the significance of interactions between measures and the rest of the facility (large interactions push you toward whole-facility or careful isolation with interaction terms).
Selecting a verification approach without stating the required measurement boundary, the measurement period, and the acceptable uncertainty, so the approach cannot actually be executed or audited.
Candidates treat approach selection as a label (A/B/C/D) rather than as a specification that implies concrete instrumentation, duration, and precision commitments.
What to do instead
For every proposed approach, write the boundary (which meters, which loads), the measurement period and frequency, and the target precision with its basis. If those three cannot be stated, the approach is not yet selected.
Using spot or short-term measurements to establish a constant-load baseline without demonstrating that the load is genuinely constant, or without a valid sampling plan for variable loads.
Short-term metering is cheap and fast, so candidates generalize from a few days of data to a full year without testing for seasonality, schedule variation, or duty-cycle effects.
What to do instead
Justify constant-load assumptions with data (e.g., logging across representative operating modes and seasons) and use a documented sampling plan with stated confidence and precision for any extrapolation from short-term to annual values.
Ignoring interactive effects and double-counting when isolating multiple measures in the same system, such as crediting both a lighting retrofit and its reduced cooling load to the same savings.
Isolation is intuitively additive, and candidates forget that measures sharing a system interact, so individually measured savings do not sum to the combined savings.
What to do instead
Explicitly identify interactions during planning, decide whether to measure them or estimate them, and either measure the combined effect or apply documented interaction adjustments so total reported savings are not double-counted.
Using whole-facility regression with too few data points, ignoring autocorrelation and seasonality, or extrapolating outside the range of observed independent variables without flagging it.
Candidates focus on getting a high R-squared and treat that as sufficient evidence of a valid model, overlooking data sufficiency, residual diagnostics, and range-of-validity limits.
What to do instead
Check minimum data sufficiency for the chosen option, examine residuals for autocorrelation and pattern, report model uncertainty, and state the range of independent variables over which the model is valid. Flag any post-retrofit conditions outside that range.
Writing the M&V plan after the retrofit is installed, or writing it without specifying baseline data requirements, meter locations, responsibilities, and the savings calculation method.
M&V planning is treated as paperwork to complete before reporting rather than as the document that must exist before baseline data collection begins.
What to do instead
Draft the plan before construction, and require it to state the IPMVP option, boundary, baseline period and data sources, measurement equipment and calibration, adjustment method, reporting frequency, and named responsibilities for each party.
Reporting savings without separating them from avoided cost, without stating the baseline and adjustment period used, and without disclosing uncertainty, so the number cannot be reproduced.
Stakeholders want a single dollar figure, and candidates compress the reporting into that figure, dropping the assumptions and uncertainty that make it defensible.
What to do instead
Report energy savings, then cost savings using stated rates, then the baseline and reporting periods, all adjustments applied, and the uncertainty at a stated confidence level. Include enough detail that an independent reviewer could recompute the figure.
Specifying meters without matching accuracy class, calibration interval, and data-logging frequency to the magnitude of the expected savings and the variability of the load.
Candidates default to whatever meter class is common or cheapest rather than deriving the requirement from the measurement and verification plan's precision target.
What to do instead
Work backward from the required precision: estimate the load and expected savings, determine the allowable meter error as a fraction of savings, then select accuracy class, calibration schedule, and logging interval to meet it. Document the rationale.
Treating a regression model as causal and using it to claim measure savings when the model only captures correlation, or omitting a known driver (e.g., occupancy) that is correlated with the retrofit timing.
High fit statistics create false confidence, and candidates skip the step of checking whether the model's independent variables actually explain the load or merely track time.
What to do instead
Validate that independent variables are physically plausible drivers, check for confounding with the retrofit period, and prefer models whose coefficients have engineering meaning. Where confounding is possible, use an isolation approach or additional metering.