In short: Building performance simulation is often assumed to trade speed for accuracy but the research shows that is a false choice at early design stage. A simplified model on a validated EnergyPlus engine matched a detailed hand-built model within 4.8% on heating and 14.0% on cooling, inside the 20% screening margin, with automation costing about 2% (Picco and Marengo, 2019).
Every practitioner who has been offered a “fast, simple” simulation tool has had the same thought *“What is this simplicity costing me?”. * It is the right instinct. In building performance simulation, speed is usually purchased with one of two currencies: a degraded calculation method, or hidden assumptions where the user cannot inspect them. Either can quietly turn a result from evidence into decoration.
So the question deserves a precise answer rather than a reassurance. How much accuracy does simplification actually cost and is what remains good enough to make design decisions with? The research behind the FREDS platform answered it the only credible way: by building both versions of the same building and measuring the gap (Picco and Marengo, 2019).
Building performance simulation: defining "accurate enough"
Accuracy is meaningless without a benchmark and a threshold. The benchmark in this research is the fully detailed dynamic model: a building described in EnergyPlus surface by surface, zone by zone, system by system, by an expert modeller, the gold standard, and the reason such models take weeks and significant budgets to produce.
The threshold is a 20% deviation margin at the early design stage. That number is not arbitrary. Published studies of detailed, even calibrated, simulation models routinely report total deviations from measured reality in a similar range, uncertainty in occupancy, weather, construction quality and operation guarantees it. Demanding laboratory precision from a concept-stage screening model would be demanding something the detailed model itself cannot deliver against reality. What matters at an early stage is decision validity : would the screening result and the detailed result lead you to the same design choice? Inside a 20% band, for comparative early-stage decisions, they do.
The experiment
The test case was deliberately non-trivial: a medium-sized private clinic in Bergamo, Italy. A healthcare building with demanding internal conditions, complex occupancy and real mechanical systems. Three models of the same building were compared:
- The detailed model : fully described in EnergyPlus by hand. Weeks of expert work.
- The simplified model : built manually in EnergyPlus using the simplified description method developed in the research. Geometry reduced to its thermally significant essentials, zoning rationalised, inputs cut to the set that drives the physics.
- The final tool model : the same simplified approach, but generated automatically by the platform from a limited set of structured inputs. Minutes of anyone’s work.
Comparing 1 to 3 measures the total cost of simplification plus automation. Comparing 2 to 3 isolates the cost of automation alone.
The results
Annual heating demand: detailed model 791.075 kWh ; final tool model 753,134 kWh , a deviation of 4.8% . Annual cooling demand: 128,102 kWh against 110,076 kWh , a deviation of 14.0% . Both sit inside the 20% margin; the heating result, the dominant load in this climate, lands remarkably close to the gold standard.
Just as telling is the comparison between the hand-built simplified model and the automatically generated one: differences within roughly 2% even in the worst case, traceable to small geometric conventions (how roof and ground-floor dimensions are defined for automatic generation) rather than to the physics. The automation or as we say the part that converts weeks of expert time into minutes is essentially free in accuracy terms. The simplification method carries a modest, quantified, acceptable cost.
Why simplification works better than intuition suggests
The instinct that “more detail = more accuracy” assumes all detail matters equally. It doesn’t. A building performance simulation is dominated by a relatively small set of drivers: climate, geometry at the level of orientation and proportion, envelope thermal quality, internal gains, ventilation, system efficiencies. Much of what makes detailed models slow, precise internal partition geometry, furniture-level zoning, exhaustive surface description contributes effort faster than it contributes signal, especially at a stage when those details are guessed anyway.
There is a second, less comfortable reason. A detailed model demands hundreds of inputs; at an early design stage, most are unknown, so the modeller fills them with assumptions. Detail that outruns knowledge does not add accuracy, it adds false precision , and it buries the assumptions where no one can audit them. A disciplined simplified model, built on the inputs that are actually known and explicit about its defaults, can be more honest about uncertainty than a detailed model padded with guesswork. (How much those buried assumptions can move results is a story of its own see our analysis of occupancy databases, where standard inputs shifted outcomes by orders of magnitude.)
Where each tool belongs
None of this retires detailed simulation, and a platform built by simulation researchers will not pretend it does. The honest division of labour:
Screening simulation (minutes, limited inputs, ±20% band): concept and feasibility stages. Comparing massing, orientation, glazing and envelope options; testing whether a target is realistic; deciding which design deserves detailed work. Its value is iteration and analysis at the time of design.
Detailed simulation (weeks, full information, expert-driven): technical design forward. Compliance, plant sizing, control strategies, optimization of a frozen design, calibration against measured data. Its value is precision applied to a design that has earned it.
The failure mode of the industry has not been using detailed tools, it has been using only detailed tools, which in practice means the earliest, most influential decisions get no simulation at all. A 4.8% deviation available in minutes beats a 0% deviation that arrives after the decision.
What to take from this
Speed and accuracy in building performance simulations are not, within sensible limits, in conflict.They were only made to look that way by tools designed for a single stage of the process. The evidence stacks up the other way: a validated engine, a simplification method tested against the gold standard, and a verified result well within the threshold the discipline accepts for design-stage screening, on a truly complex building.
The next time a fast tool asks for your trust, ask it for this: the engine it runs, the benchmark it was tested against, and the numbers. Fast is easy to claim.
FREDS is the published version. Run a validated dynamic simulation on your own building in minutes start free or talk to our consultancy team when the project needs the detailed treatment.
Related reading
Frequently Asked Questions
How accurate is simplified building performance simulation?
What is an acceptable accuracy margin for early-stage simulation?
Does faster simulation always mean less accurate results?
When should I use detailed simulation instead of screening?
References
Written by Marco Picco, PhD co-founder of FREDS4Buildings and a building-physics lecturer at the University of Lancashire, whose work on building energy simulation has been published in Energy & Buildings , Building and Environment and the IBPSA Building Simulation conference proceedings. About Marco Picco
Reference: Picco, M. and Marengo, M. (2019). A fast response performance simulation screening tool in support of early stage building design. Proceedings of the 16th IBPSA Building Simulation Conference, Rome. View paper
