Can AI Do a Heat Pump Heat Loss Calculation? What MCS Expects

AI can help around a heat loss calculation, but it should not be the calculation. MCS expects a room-by-room calculation from a competent designer using accepted software and methods, kept on file. Use AI to check survey inputs, explain the results in plain English and draft customer wording, then run the numbers in approved software yourself.

The reason is practical as well as regulatory. A heat loss figure sets the size of the heat pump, the flow temperature and the size of every radiator. If the number is wrong, the customer lives with a system that struggles on the coldest day or one that is oversized and cycles. A chat model can sound confident while using the wrong U-value, so it makes a poor sole author of that number.

Follow me on Instagram@sagnikteaches

The free calculators that rank for this search return one whole-house number. That number is not what an MCS designer is held to: the obligation is a room-by-room calculation tied to survey evidence, a design that follows from it, and a file that survives an audit. For software that runs the calculation itself, see heat pump installer software for surveys, heat loss and quotes.

Connect on LinkedInSagnik Bhattacharya

What MCS expects from the designer, in outline

MCS is the certification scheme for small-scale renewable installations, and its heat pump design standard is MIS 3005-D. The version on the MCS site when this was checked was Issue 2.0, dated 5 December 2025. The standard changes, so read the current issue yourself; what follows is a summary of the shape of the obligation, not a substitute for it.

Subscribe on YouTube@sagnikteaches

• A heat loss calculation for the building. MCS's own installer guidance says that under MIS 3005-D the heat pump must be selected on a heat load calculation to BS EN 12831-1:2017, which works room by room, not on a rule-of-thumb estimate from a table or a square-metre multiplier. The whole-dwelling result sets the heat pump capacity; the room results set the emitters.
• Design based on that calculation. The heat pump selected has to meet the calculated heat loss, and the designer works out flow temperature and emitter sizes from it. Inputs such as the external design temperature follow the method and data the standard points to, so check them against the current issue rather than accepting a figure a calculator or chat model supplies.
• A performance estimate at the quoting stage. MCS 031 (031 a for wet systems, 031 b for air-to-air) is the heat pump system performance estimate: an annual energy-use figure for a typical year, produced with MCS's own spreadsheet tool and given to the customer with the quote. MCS states plainly that this tool is not a design tool for calculating heat load or sizing a heat pump, so it never stands in for the calculation above.
• Records. The installer's certification body can audit the file. The calculation, inputs and design have to be there, with evidence for each input.
• Approved tools. MCS links its own Heat Load Calculator from its installer resources, free for certified installers, and many commercial design packages follow the standard method. The calculation should come from one of those, not from a general chat tool.

Whether a particular AI tool could ever be an accepted calculation method is a question for MCS and your certification body, not for a vendor's marketing page. Until you have that in writing, treat AI as support, not as the calculation.

What goes wrong when a chat model does the calculation

It invents inputs

Ask for the heat loss of a three-bedroom semi and a model will happily pick U-values, an air change rate and room temperatures. They look plausible and are not from your survey. Because the inputs are hidden inside a narrative, the error is hard to see.

It adds up unreliably

Summing eight room figures and a ventilation loss is a task where a language model can drop a term or round inconsistently, and it will not flag that it did. A spreadsheet or the calculator does this exactly.

It does not know the current standard

Design temperatures, method details and clauses change between issues. A model trained earlier may quote the old ones. That is a compliance risk, which is the opposite of what the file is meant to prove.

It cannot own the result

A heat loss calculation is signed off by a person. An output that nobody can trace to inputs and a method is not something you can defend to a customer or an auditor.

A worked house: figures and what they decide

This is an illustration with invented numbers for a three-bedroom semi-detached house from the 1930s. The figures are not from a real survey, and the room temperatures and U-values are examples, not the ones the standard requires for any job.

That total, 1.55 + 1.05 + 0.95 + 0.55 + 0.85 + 0.70 + 0.50 + 0.65, is 6.80 kW. It decides the heat pump: the selected unit has to deliver at least that at the design conditions, judged on its output at those conditions, not on the figure on its label. It also decides the emitters. Say the lounge radiator is rated 2.0 kW at the traditional 50 degree temperature difference. At a heat pump flow temperature, the mean water temperature is much lower, so the radiator's output drops to a fraction of its rating, perhaps a third (the exact factor comes from the manufacturer's data). A 0.65 kW output against a 1.55 kW need means the lounge needs a much larger emitter or a lower heat loss. That is the kind of finding the room-by-room calculation exists to produce, and it is also the kind of finding that changes the customer's quote.

Where AI genuinely helps

Auditing the survey inputs before you calculate

Paste your survey notes and ask the assistant to list what is missing or inconsistent. It reads fast and does not get bored, which suits this kind of checking.

Below are my survey notes for a heat loss calculation. Do NOT calculate anything and do NOT supply any missing values.
Task: list (1) rooms with no target temperature, (2) walls, floors, roofs, windows or doors with no construction or U-value source recorded, (3) rooms where dimensions look inconsistent with each other, (4) anything on the notes that suggests an unheated space, a party wall or a ventilation system I have not recorded, (5) questions to ask the customer on a revisit.

Survey notes:
[paste notes]

An illustrative output: "Bedroom 3: no target temperature recorded. Rear extension: wall construction not stated. Loft room: roof insulation depth mentioned as 'some' with no figure. Kitchen: extractor fan noted, no air change information. Revisit questions: what is above the garage, and is the porch heated?" Check each flag against the notes and fix the ones that are real.

Sense-checking the output of the calculator

After the software has produced the room figures, ask the assistant to flag results that look odd relative to each other, given the notes. Something like a bathroom with a loss higher than the lounge, or a small north bedroom with a loss that is low next to a larger one with similar walls. It cannot say the figure is wrong; it can send you back to the inputs. Give it the figures and the notes, tell it to question, not recalculate.

Explaining the result in plain English

I will give you a heat loss summary for a customer. Write a short explanation (about 200 words) in plain English at a reading age of about 14. Explain what the whole-house figure means, why room figures matter for radiator sizes, and what the designer will do next. Do not mention running costs, savings, payback or grants. Do not add any numbers that I have not given you. Use only these figures:
Whole dwelling: 6.80 kW. Lounge 1.55 kW, dining room 1.05 kW...
[paste figures]

Sample output (illustrative): "Your home loses about 6.8 kilowatts of heat on the coldest design day. We use this figure to choose a heat pump that can match it. The same calculation is done room by room, which tells us how big each radiator needs to be. Some of your current radiators will be too small for a heat pump's lower water temperature, so we may recommend larger ones in the lounge and dining room." What you would fix: nothing in this draft, but if the assistant adds "this will cut your bills" you delete it.

Drafting the cover note and the follow-up questions

The customer pack needs a covering letter, a list of what you have assumed, and questions for any gap. These are the writing jobs where AI saves the most time. Explaining heat pump options in a quote without savings promises gives a template for the commercial side.

A record of what AI did and did not do

Because the calculation file has to stand up to audit, keep a one-line note on each job: which software produced the heat loss, which version, who entered the inputs, who reviewed them, and what AI was used for (for example "survey notes input check; customer explanation draft"). If anyone asks, the file shows that the calculation came from the accepted tool and a named person. Keep the AI prompts and outputs only if they affect what the customer was told.

Two shortcuts that undo the calculation

Reusing a house's figures for its neighbour

Semis on one street look alike, and a model will cheerfully say they should lose similar amounts. They do not: one has had loft insulation, another a rear extension. Every job gets its own calculation from its own survey.

Letting wording drift from the design

An AI-written explanation says "a 7 kW heat pump will heat your home" when the designed unit is 8 kW or the calculation shows 6.8 kW. Read every number in the draft against the calculation before it goes out, and check the wording does not promise comfort levels the design does not state.

What the job file should show before the quote goes out

1. Calculation produced by accepted software or the MCS calculator, on the current standard.
2. Every input traceable to the survey or a recorded source.
3. Room-by-room results stored with the whole-dwelling figure.
4. Heat pump selected on its output at the design conditions, not its label.
5. Emitters checked at the design flow temperature.
6. Pre-sale information given to the customer as the standard requires.
7. AI use, if any, limited to checking, explaining and drafting, and noted on the file.
8. Every figure in customer wording matched to the calculation.

The same mindset applies to the load calculations used in other heating and cooling work: why Manual J software still matters makes the parallel case for the load calculation used in many replacement quotes, and what AI estimating software measures and misses covers the estimating side.

How much one wrong input moves the answer

A quick sum shows why inputs deserve an audit. Ventilation loss is roughly the air volume, times air changes per hour, times 0.33 watt-hours per cubic metre per degree, times the temperature difference. For an illustrative house of 250 cubic metres with a 21 degree difference between inside and outside:

• At 0.5 air changes an hour: 250 x 0.5 x 0.33 x 21 = about 866 W.
• At 0.2 air changes an hour: 250 x 0.2 x 0.33 x 21 = about 346 W.

That is a swing of over 500 W from one assumption, nearly 8% of the 6.8 kW total in the worked house. The correct figure depends on the method the standard prescribes and the evidence you hold for the building, such as a measured air permeability test, so do not copy these values. The lesson is that a chat model that silently picks 0.5 or 0.2 has moved your total by enough to tip a borderline selection into the next model size. Make the assistant list every assumption and then check each against the survey.

Turning rough survey notes into inputs you can audit

The input audit works best when notes are written to be read by someone who was not there. Here is a before and after, illustrative.

Before: "Back bedroom cold, old windows, wall solid I think, loft some insulation, extension at back 2010ish."

After: "Bedroom 3 (rear, first floor): 3.1 m x 2.8 m, ceiling 2.4 m. One external wall (rear), solid brick, measured thickness 225 mm, no insulation seen. One window, double glazed, uPVC, installed about 2005 per frame label, 1.2 m x 1.1 m. Ceiling below loft: insulation about 100 mm mineral wool, measured at hatch. Floor: above heated room. Target 18 C. Rear extension (ground floor): built approximately 2010, cavity wall, insulation not confirmed, ask customer for building control certificate."

Run the audit prompt on the first version and it will list a dozen gaps. Run it on the second and it will list two or three real ones. The point is that the survey template, not the assistant, produces quality. Many firms tidy their templates after a month of reading the assistant's gap lists, because the same gaps keep appearing: loft depth, window age, unheated spaces, party walls. Add those as required fields.

Answering "my neighbour got a smaller one"

Customers compare. A semi next door with a 5 kW heat pump does not mean a 5 kW unit suits this house, and a clear, calm explanation avoids a lost job. Give the assistant the facts and ask for a draft you will edit.

Draft a friendly reply (under 150 words) to a customer who says a neighbour's similar house was fitted with a smaller heat pump. Use only these facts: our calculation for their house gives a whole-dwelling heat loss of 6.80 kW; the neighbour's house may have different insulation, extensions, windows or ventilation; we calculate each house room by room from its own survey. Do not criticise the neighbour's installer, do not promise bills or savings, and do not mention any figure I have not given you. Offer to walk through the room-by-room results by phone.

Illustrative output: "Thanks for mentioning this. Two houses that look alike can lose heat very differently: loft insulation, an extension, window age and how airtight the house is all change the number. We calculate your home room by room from our own survey, and the result is 6.80 kW. I would be glad to go through the room figures with you by phone so you can see where each part comes from." What you would fix: check it does not imply the neighbour's system is wrong, and that the figure matches the file.

What to do if you want AI closer to the calculation

Some commercial design packages include AI features, such as reading floor plans to pre-populate rooms or suggesting construction types from a photo. If you use one, treat the pre-filled values as survey suggestions, and apply the same input audit to them. The question for the vendor is whether the package follows the standard method, which version of the standard, and whether it is accepted by your certification body for MCS work. Ask for that in writing, and keep it on file. If the answer is vague, keep the calculation in a tool you can defend.

The same caution should apply to any free web calculator that ranks highly for this topic. Some are useful for a quick sense of scale in a first conversation, none are a replacement for a survey-based calculation, and they are not evidence an auditor will accept. A sensible use is as a sanity range: if your calculated figure is wildly different from a rough estimate, look at the inputs again before you trust either.

A realistic slip: the explanation that outran the design

Here is an illustration of how wording drifts. A small installer asks an assistant to write the customer summary from the calculator printout. The printout shows 6.80 kW for the dwelling and an 8 kW unit chosen after the designer checked the unit's output at the design temperature. The assistant writes "your 8 kW heat pump covers your 6.8 kW loss with room to spare". That sentence reads fine, but it implies deliberate oversizing and a comfort margin the design never claimed, and it invites the question of why the customer is paying for more than the calculation says. The designer catches it on the read-through and rewrites it to say the unit was selected because its output at the design conditions meets the calculated loss. The fix for next time was one line in the prompt: "describe the selected unit only in the words I give you." Wording that sounds helpful is exactly the kind an auditor or an unhappy customer will quote back later.

Keep a short library of approved sentences for the recurring points: what the whole-dwelling figure is, why emitters may change, what the designer checked, what happens if the customer alters the house later. Ask the assistant to assemble customer letters from those building blocks, and the risk of improvised claims drops sharply.

Where the designer's hour goes, and what AI adds to it

A competent designer can spend about an hour on the calculation for an ordinary house once the survey is in. The AI input audit adds around five minutes and tends to save a revisit. The customer explanation takes minutes to draft and ten to check. If you use a business plan such as ChatGPT Business ($20 per user per month billed annually, two seats minimum) the cost is small against a survey visit, but the saving only holds if the calculation itself stays in the approved tool. The point of the exercise is to spend AI's speed on the writing and checking, and to spend your time on the calculation and the design, which is the part MCS holds you to.

Heat loss and AI: what installers ask

Can I use ChatGPT to do the room-by-room heat loss?

Not as the calculation of record. A chat model can do arithmetic wrongly, invent a U-value or assume an air change rate without telling you. Run the calculation in approved software and use the assistant afterwards to explain the output or draft wording.

What does MCS expect from the designer?

In outline, a room-by-room heat loss calculation to the standard method, a design based on it, and an MCS 031 performance estimate given to the customer at quoting, all kept on file for the certification body to audit. Read the current MIS 3005-D and MCS 031 for the exact requirements.

Can AI help me check my inputs?

Yes. Give it your survey notes and ask it to list missing or inconsistent inputs, such as a room with no stated target temperature or a wall with no construction recorded. It should only flag gaps, not fill them in.

Can I let AI explain the result to customers?

Yes, if you give it the real figures and tell it not to promise running costs or savings. Read the draft against the calculation before sending, and keep the explanation consistent with the design documents.

Further reads

• Heating Engineer Software Compared: What AI Adds for a UK Firm — What AI adds to the job software a heating firm already runs.
• AI for HVAC Contractors: Getting Answers From Manuals and Spec Sheets — Get answers from manufacturer manuals and spec sheets with AI.
• Boiler Installation Checklist: AI Sorts the Photos, You Sign the Records — Sort install photos with AI while you sign the records.
• Boiler Installation Quote Template: Can AI Write the Quote and Cover Email? — A quote template and cover email for heating installs.
• How to Use AI in HVAC Design: Options, Duct Checks and Proposals — Where AI fits in system design, duct checks and proposals.
• Heating Engineer Website: What the AI Chatbot and Form Should Ask — What a heating firm's website form should ask before a survey.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: MCS standards library (MIS 3005-D:2025 version 2.0, issued 5 December 2025; MCS 031 a and b performance estimates); MCS installer guide to the updated MCS 031 standard and tool; MCS Heat Load Calculator page. Check the current issues before relying on any clause.