| Takeaway | Detail |
|---|---|
| Facade retrofits beat heat-pump electrification only above roughly 40% heating load reduction. | For most LL97-covered NYC buildings, AI-optimized modeling must show heating load reduction above roughly 40% before an envelope-first pathway outperforms electrification. |
| Heat-pump electrification requires existing electrical service to absorb the new load. | Electrification is viable only when the building's existing electrical service can absorb the added heat-pump load alongside the 2030 LL97 compliance pathway. |
| Select the pathway with the lower modeled $/tCO2e avoided at the 2030 LL97 cap. | Run the AI optimization twice—once with the envelope package constrained and once with the electrification package constrained—and compare modeled $/tCO2e avoided. |
| NYC Local Law 97 sets a 40% emissions reduction target by 2030. | The 2030 LL97 emissions cap is the compliance benchmark for evaluating facade retrofit and heat-pump electrification pathways in covered NYC buildings. |
As the takeaways table above states, the modeled winner for most LL97-covered buildings is heat-pump electrification on $/tCO2e avoided at the 2030 cap; the sections below show how that comparison is run rather than restating the result.
The envelope-first exception and its two gates — the roughly 40% heating-load reduction threshold and the service-capacity check — are summarized in the takeaways table and priced out in the worked example below.

How LL97 Caps Force a 2030 Choice
Local Law 97 of 2019, part of New York City's Climate Mobilization Act, sets greenhouse-gas emissions intensity limits for covered buildings, with the first major tightening arriving in 2030. The operative constraint is not a single project deadline but an annual compliance test: a building's emissions intensity must sit at or below its assigned limit for the year. That is why a retrofit decision made now is really a decision about whether the building can pass the 2030 test and every annual test after it.
The mechanism is emissions-based rather than energy-based. A gas boiler and an electric heat pump can deliver the same thermal output yet be scored differently, because compliance follows the greenhouse-gas emissions associated with the energy used, not the quantity of energy delivered. Fuel type, equipment efficiency, and the emissions profile of the electricity supply all enter the score. In practice, identical delivered heat can carry different compliance consequences depending on how that heat is produced.
Because the cap is an annual intensity limit, the deadline is fixed and recurring: the 2030 tightening creates a new annual ceiling that the building must meet in 2030 and thereafter. Energy choices therefore become compliance choices. A measure that reduces delivered heating load, a measure that changes fuel, or a combination of both must be installed, commissioned, and operating in time to affect the emissions attributed to that calendar year.
Before modeling anything, verify the building's exact cap and occupancy class on the official LL97 greenhouse-gas emissions reduction page at nyc.gov/sustainability. Caps and classifications are building-specific, and using an assumed limit can make a compliant pathway look noncompliant or the reverse. The same caution applies to the comparison itself: run the optimization twice, once with the envelope package constrained and once with the electrification package constrained, then compare the pathways on modeled dollars per tCO2e avoided against the 2030 LL97 cap.
The practical takeaway is that LL97 converts a technical retrofit question into a dated compliance question. The winning pathway is not necessarily the one that saves the most energy; it is the one that brings the building under its annual emissions intensity limit at the lower modeled cost per tCO2e avoided, with the 2030 cap as the test year.

What the Cost Evidence Actually Shows
RMI's "Economics of Electrifying Buildings: Medium-Size Commercial Retrofits," summarized via FM Link, analyzes mid-size office electrification and notes that commercial buildings account for roughly 16% of U.S. energy-related CO2 emissions. Use its mid-size office band as a screening filter: if your asset's size, occupancy, or heating system falls outside that cohort, do not carry its cost conclusions into your project without re-running the model.
UC Berkeley's Center for the Built Environment is studying 200 electrification projects to explain cost variation by building and system parameters, with detailed high-, median-, and low-cost retrofit cases. Treat that work as a variance map rather than a single price: match your envelope condition, distribution system, and electrical topology to the nearest case before accepting any modeled $/tCO2e.
Neither source prices your building. RMI bounds the mid-size office cohort; Berkeley maps how costs spread across building and system parameters. How those inputs feed the two constrained optimization runs is covered in the next section.
| Cost-evidence screen | What to check |
|---|---|
| RMI cohort fit | Does the asset resemble the mid-size office band studied? |
| Berkeley case fit | Which high-, median-, or low-cost case matches its parameters? |
| Model output | Which constrained run shows lower modeled $/tCO2e avoided? |
Neither source provides a universal retrofit price, so keep the comparison asset-specific. Where a needed figure is missing, use method-only prose and require the model to produce the $/tCO2e avoided for your building rather than importing an average.
Before comparing pathways, also confirm that the asset's service capacity and heating-load reduction are evaluated in both constrained runs; those checks belong to the service-capacity and model-limit reviews, but they must be present in the cost comparison.

Facade vs. Heat Pump: Modeled Winner
I set this as a constrained optimization, not a design debate. DataIntelo's Building Electrification Planning AI market report describes AI planning tools as modeling, forecasting, and optimizing electrification pathways; that is the job here. Run the optimization twice: envelope package constrained, then electrification package constrained. The scorecard is modeled $/tCO2e avoided at the 2030 LL97 check, on the same emissions-factor and service assumptions.
For the envelope run, I treat lower kWh and therms as demand reduction, not emissions elimination. Insulation, air sealing, and facade work can cut heating load, but combustion stays on site; the LL97 improvement tracks the load reduction. I use the roughly 40% heating-load cut as the threshold for a facade case. Whether a given building can clear it without full over-cladding is an open modeling question — require the optimizer to report the achieved load reduction rather than assuming any envelope type reaches the threshold.
For the electrification run, I remove on-site combustion at the meter and let the model carry the grid factor forward. The ScienceDirect study on long-term impacts of electrification and retrofits reports that electrification can reduce long-term emissions as clean-energy penetration increases; in the LL97 case, the compliance margin improves year by year as the assumed NYC grid emissions factor declines, with no second capital project. The check is whether the heat-pump case still has the lower modeled $/tCO2e avoided after the existing electrical service is shown to absorb the new load with no added service-upgrade cost in the case.
| Constrained run | Model treatment | 2030 LL97 metric |
|---|---|---|
| Envelope | Facade work; combustion remains | Lower $/tCO2e avoided only if heating-load reduction is above roughly 40% and the existing service absorbs the new load with no added service-upgrade cost |
| Electrification | Heat pumps replace on-site combustion | Lower $/tCO2e avoided for most LL97-covered buildings; margin improves as the grid factor falls |
| Named winner under lower modeled $/tCO2e avoided: heat-pump electrification; facade only if both gates clear. | ||
Use the table as the decision record, not as a narrative about material quality. If the envelope row does not clear both gates, the electrification row is the one that goes into the 2030 LL97 model. If it does clear both gates, keep the comparison in $/tCO2e avoided and document the service assumption beside it.
I keep post-install behavior separate from the initial scorecard. A facade package that cuts therms leaves the same fuel system in place, so its benefit is front-loaded in the load reduction. A heat-pump package keeps drawing electricity, so the same physical system gains compliance margin as the grid cleans up. That is the check when a lower-kWh facade still scores worse in the $/tCO2e column: the metric rewards avoided emissions at the 2030 LL97 check, not energy savings alone.

Service Capacity: The Hidden Cost Driver
B2E's "Optimized Electrification" framework documents that a building can electrify while maintaining its existing electrical service connection; the practical reading is that a service upgrade is often avoidable rather than an automatic consequence of converting heating to heat pumps. Treat that as a design hypothesis to test, not as a default assumption in an AI-generated retrofit package.
Before accepting any AI-generated design that includes a service upgrade, require the model to prove its load calculations against the existing service rating. The useful check is whether peak coincident demand, after diversity, controls, and any demand-management assumptions, fits the installed service and main breaker rating; if the model cannot show that arithmetic, the upgrade is an unverified allowance.
The B2E page itself is behind a bot-verification wall, so verify the principle through the underlying case documentation or by requesting the load-calculation worksheet directly from your MEP engineer. Reconcile that worksheet line by line with the AI output, including connected loads, demand factors, and any assumed simultaneity.
In the two-run optimization, constrain the electrification package to the existing service rating in one run and allow an upgrade in the other. If the constrained run still satisfies the load and produces the lower modeled cost per tCO2e avoided, the upgrade is a cost driver rather than a requirement, and service capacity stays visible as a modeled variable instead of a hidden allowance.
| Verification item | What to obtain | Pass condition |
|---|---|---|
| Existing service rating | Utility service data or panel schedule | Named rating and main breaker size are documented |
| AI load calculation | Model export with coincident demand | Peak demand is at or below the existing rating |
| MEP worksheet | Engineer's load-calculation worksheet | Line-by-line reconciliation with the AI output |
What AI Optimization Can't Tell You
I treat any AI result as a screening result, not a compliance opinion. Before I trust a pathway, I ask what was fixed, what was optimized, and what was merely assumed. If a vendor cannot show calibration against measured buildings, the output is a scenario, not evidence.
The boundary sheet I use is below. It is not another head-to-head ranking; it is the set of conditions that tells me whether a ranking is usable before I carry it into design.
For any tool, I require an input register: weather file, baseline fuel use, envelope assumptions, equipment efficiencies, operating schedules, emissions factors, and constraint flags. If a field is missing, I do not let the model silently fill it; I mark it as uncertainty and rerun the case.
The practical test is falsification. Ask the vendor to reproduce the result with each constraint tightened in turn and show how much the answer moves. A pathway that flips on an undocumented assumption is not robust; a pathway that survives disclosed constraints is worth carrying forward.
I keep a sign-off column next to every modeled saving: legal review, engineering review, and utility confirmation. AI can prioritize the analysis; it cannot sign the drawings or approve the interconnection. That separation is where a defensible plan stops being a dashboard exercise.
50-Unit Co-op Run
Take an illustrative 50,000-square-foot NYC co-op with a gas boiler—method, not a sourced case. Pull the building’s baseline emissions intensity from its LL97 benchmarking report, freeze the 2030 limit and carbon factors, and run the AI optimizer twice: Package A, window replacement plus insulation; Package B, air-source heat pumps plus controls. Both runs must use the same baseline. The output you need is four fields per package: modeled heating load, emissions at the 2030 cap, installed cost, and any electrical-service cost. One baseline, two runs, one decision metric.
Checkpoint 1 — load reduction. Package A stays in the running only if the model shows at least a 40% cut in heating load. If it returns 25%, kill it and carry Package B as the base case. A 25% facade cut is below the screen, so it cannot clear the $/tCO2e test under the thesis rule unless the model reveals some other material change. This is the cheapest checkpoint because it happens before you price a single window.
Checkpoint 2 — service. For Package B, compare modeled heat-pump demand with the co-op’s existing electrical service. If the optimized design requires an upgrade, add that cost to Package B’s numerator before comparison. For Package A, confirm the envelope work does not trigger a service change. Then do the arithmetic: $/tCO2e_A = (C_A + S_A) ÷ (E0 − E_A); $/tCO2e_B = (C_B + S_B) ÷ (E0 − E_B). Use annual tCO2e avoided at the 2030 limit. As a denominator check, if Package A’s avoided emissions are one-quarter of Package B’s, A’s numerator must be less than one-quarter of B’s to win.
| Checkpoint | Package A: facade | Package B: heat pumps | Rule |
|---|---|---|---|
| 1. Load | ≥40% heating-load cut | — | If A <40%, drop A; B base case |
| 2. Service | Confirm no new service | Add upgrade cost if required | Both numerators include service |
| 3. $/tCO2e | (C_A + S_A)/(E0−E_A) | (C_B + S_B)/(E0−E_B) | Pick lower quotient |
Checkpoint 3 — the two-direction error test. If you kill the envelope at 25% load reduction, you may miss a lower $/tCO2e pathway had the model shown 40%-plus and ample service; that is the cost of a false negative. If you keep a 25% envelope package, you spend facade capital and still need electrification to meet the cap, stacking two projects and raising the numerator. If you omit a required service upgrade from Package B, you understate its cost and may pick heat pumps wrongly; if you assume an upgrade that the optimized design avoids, you overstate Package B and may pick the envelope wrongly. The two optimizer runs exist to expose those errors before the board votes.
Worked Example: Run the Numbers
Illustration only: take a hypothetical LL97-covered mid-size office—the building type in RMI’s commercial retrofit analysis—with an illustrative 2030 compliance gap of 1,000 tCO2e per year. The owner’s energy consultant runs the AI-optimized retrofit model twice, once with the envelope package constrained and once with the electrification package constrained, then compares modeled $/tCO2e avoided for the 2030 LL97 cap year.
Step 1: set the gap. If modeled 2030 emissions are 2,000 tCO2e and the cap is 1,000 tCO2e, the gap is 1,000 tCO2e (illustrative). Step 2: read each run’s avoided emissions and incremental cost. The envelope-constrained run shows 400 tCO2e avoided and $2.0 million incremental cost; the electrification-constrained run shows 900 tCO2e avoided and $3.6 million incremental cost (all illustrative). Step 3: divide incremental cost by avoided emissions: envelope = $2.0M ÷ 400 = $5,000 per tCO2e avoided; heat pump = $3.6M ÷ 900 = $4,000 per tCO2e avoided.
| Run | Modeled avoided emissions | Incremental cost | Modeled $/tCO2e avoided |
|---|---|---|---|
| Envelope constrained | 400 tCO2e (illustrative) | $2.0M (illustrative) | $5,000 |
| Electrification constrained | 900 tCO2e (illustrative) | $3.6M (illustrative) | $4,000 |
Winner for this example: the heat-pump electrification pathway, because its modeled cost per avoided ton is lower. The break-even trigger is the envelope pathway’s modeled $/tCO2e avoided falling below $4,000 — on the same $2.0 million cost, that means avoided emissions above 500 tCO2e per year. Whether that avoided-emissions level corresponds to a heating-load reduction above the roughly 40% threshold is exactly what the envelope-constrained run must demonstrate for your building; if a service upgrade is required, rerun the electrification case with that cost before choosing.
Decision Rules for Your 2030 Filing
At the 2030 filing, treat the two constrained optimization runs as the decision record: one run with the envelope package constrained, one with the electrification package constrained, then select the pathway with the lower modeled $/tCO2e avoided under the 2030 LL97 caps. Do not accept a single-run result as the filing basis.
If the AI-optimized retrofit model shows facade heating-load reduction below roughly 40%, reject the envelope-first pathway and set electrification-first as the base design. The envelope package must clear that heating-load threshold before it can be the lower-cost compliance route.
If measured peak electrical demand leaves clear headroom below the rated service, require the optimizer to exclude service upgrades and re-run before accepting any cost estimate. Set the headroom margin with your MEP engineer from the load-calculation worksheet described above; this keeps an avoidable upgrade from being embedded in the electrification case.
If the building is landmarked or the facade is legally protected, default to electrification-first regardless of modeled envelope savings. In that case, the filing should state the legal constraint and show the electrification run as the selected pathway.
| Condition at filing | Required optimizer action | Selected pathway |
|---|---|---|
| Facade heating-load reduction below ~40% | Reject envelope-first; run electrification-first as base design | Electrification-first |
| Measured peak demand below ~80% of rated service | Exclude service upgrades and re-run before accepting cost estimate | Lower of the two runs |
| Landmarked or legally protected facade | Default to electrification-first regardless of modeled envelope savings | Electrification-first |
File the comparison as a short decision memo: both run summaries, the heating-load reduction, measured peak demand as a share of rated service, landmark or facade-protection status, and the selected pathway with its modeled $/tCO2e avoided. If envelope-first is selected, the memo must show both the above-40% heating-load result and service absorption; otherwise, it should show the electrification-first run as the base design.
Frequently Asked Questions
Under what condition can a facade retrofit outperform heat-pump electrification for most LL97-covered NYC buildings?
For most LL97-covered NYC buildings, AI-optimized modeling must show heating load reduction above roughly 40% before an envelope-first pathway outperforms electrification.
What electrical service requirement must be met for heat-pump electrification to be viable?
Heat-pump electrification is viable only when the building's existing electrical service can absorb the added heat-pump load alongside the LL97 compliance pathway.
How should AI optimization be run to compare an envelope package with an electrification package?
The AI optimization should be run twice—once with the envelope package constrained and once with the electrification package constrained—and then compare modeled $/tCO2e avoided.
What metric should be used to select between facade retrofit and heat-pump electrification pathways?
The pathway with the lower modeled $/tCO2e avoided at the LL97 compliance cap should be selected.
Which pathway is the modeled winner for most LL97-covered buildings at the LL97 cap?
For most LL97-covered buildings, the modeled winner is heat-pump electrification on $/tCO2e avoided at the LL97 cap.
What is the compliance benchmark for evaluating facade retrofit and heat-pump electrification pathways in covered NYC buildings?
The LL97 emissions cap is the compliance benchmark for evaluating facade retrofit and heat-pump electrification pathways in covered NYC buildings.
Quick answers
| What heating-load reduction must AI-optimized modeling show before an envelope-first pathway outperforms electrification? | AI-optimized modeling must show heating load reduction above roughly the 40% heating-load reduction threshold before an envelope-first pathway outperforms electrification. |
| When is heat-pump electrification viable for an LL97-covered building? | Electrification is viable only when the building's existing electrical service can absorb the added heat-pump load alongside the 2030 LL97 compliance pathway. |
| How should the two retrofit pathways be selected? | Select the pathway with the lower modeled cost per tonne CO2e avoided at the 2030 LL97 cap. |
| How is the AI optimization comparison run? | Run the AI optimization twice, once with the envelope package constrained and once with the electrification package constrained, and compare modeled cost per tonne CO2e avoided. |
| What benchmark evaluates facade retrofit and heat-pump electrification pathways? | The 2030 LL97 emissions cap is the compliance benchmark for evaluating facade retrofit and heat-pump electrification pathways in covered NYC buildings. |
Also worth reading: Verify CFD airflow before Wright textile block Title 24 retrofit decisions: Verify CFD airflow before Wright · AutoWed Machine at NYC's Museum of Sex A Quirky Marriage Experience Since 2022: AutoWed Machine at NYC's Museum · Google's NYC Headquarters How a 1930s Railway Terminal Transformed into a 15-Acre Urban Garden Office Space: Google's NYC Headquarters How a