Author: Nadali. Prepared with AI assistance.
A smaller electricity bill after introducing AI does not establish that a hotel became more energy-efficient. Tariffs, guest numbers, weather or laundry operations may have changed. A useful first task for an assistant is to assemble a checked description of what was measured before the change.
This guide is for small-hotel owners and operations managers. Its output is a baseline-period record and a list of gaps for the responsible specialist. It is an editorial workflow, not an energy audit, certification or control-system trial. AI receives no access to equipment.
1. Define the measurement boundary
Start with a sentence: “We are comparing electricity for the whole building at the main meter, including the kitchen and laundry.” Add the meter and building identifiers and a list of included areas. If a tenant restaurant shares the supply, make that explicit.
Do not subtract invented kitchen consumption simply because you want a guest-room metric. Without separate measurement or an agreed allocation method, it remains part of the total. Mark a new series when the boundary changes instead of silently joining it to the old one.
ENERGY STAR’s energy-management guidelines recommend accounting for purchased and on-site-generated energy sources alongside building information. In our working copy, list electricity, gas and other available sources separately. If the hotel generates electricity, do not label grid imports as its entire consumption. A specialist approves the reconciliation method.
2. Assemble traceable records
Our suggested fields for each record are:
- Period start and end, including its number of days.
- Source: a bill, meter export or another approved document.
- Energy type, quantity and unit.
- Actual or estimated reading, missing value or correction.
- Cost recorded separately from physical consumption.
- Occupied room-nights for the same period.
- Floor area within the boundary and operational changes.
Here, one room-night means one occupied room for one night, regardless of its number of guests. This is the example’s working definition, not a count of people or reservations. Specify how complimentary stays and unavailable rooms are treated.
Do not automatically label a 28-day bill as a calendar month. AI can flag overlapping dates, incompatible units and absent documents. A person checks every total against the original. Do not add a building’s main meter to its own submeters: that counts the same energy twice.
Use fictional data for a practice run. For real work, use only approved aggregates without guest names, payment details or unnecessary contractual information.
3. Show several measures together
ENERGY STAR explains EUI as energy use relative to floor area; the annual metric uses annual consumption and gross building floor area. Do not label monthly electricity per square metre as full annual EUI.
For an internal review, you can separately show electricity over the period, per day and per occupied room-night. Labels must retain the energy type, boundary and period. None of these ratios accounts for every operating condition by itself.
Floor area, occupancy and weather answer different questions. A reduction in one metric should not hide an increase in another. A US median from an ENERGY STAR table is not automatically a suitable target for a Ukrainian hotel.
4. Example: less overall, more per room-night
All numbers are fictional. Both periods contain 30 days and use the same meter boundary:
- Period A: 12,000 kWh and 600 occupied room-nights.
- Period B: 10,800 kWh and 450 occupied room-nights.
Total consumption fell by 10%. Daily consumption moved from 400 to 360 kWh. However, consumption per occupied room-night rose from 20 to 24 kWh, an increase of 20%.
That establishes neither success nor failure of a new process. We do not know, for example, whether the pool was operating, what temperatures prevailed or how much laundry work was done. Do not attribute the difference to AI or use it to judge an individual employee. At this stage, the correct output is three checked metrics and a list of missing context.
5. Do not ask the model to invent a weather adjustment
ENERGY STAR’s technical reference explains weather normalization through the relationship between energy and temperature. Its described method needs monthly data and treats energy sources separately. It is not an arbitrary deduction for a “cold winter”.
AI can help list dates, weather-data sources and operating changes. It must not invent coefficients or call a raw ratio “normalized savings”. Alongside weather information, retain a log of building closures, renovation, new equipment and service changes.
A competent specialist checks the normalization method, adequacy of the history and comparability of periods. If the evidence is insufficient, keep the conclusion provisional. Do not replace missing values with zeros.
6. Give AI a bounded task and hand over the result
An editorial prompt example: “Using the approved table, prepare a baseline-period record. Show meter boundaries, periods, units, sources and missing data. Check overlapping dates and possible double counting. Separate physical consumption, cost, floor area and room-nights. Show the calculation behind each total. Do not invent weather adjustments, savings or causes. Do not change equipment settings.”
Attach the checked table, correction log, explanations of non-comparable periods and the person responsible for the next review. Verify arithmetic in a spreadsheet independently of the assistant’s answer. Our guide to reviewing AI chart commentary helps prevent stronger claims appearing when the report is written.
Takeaway: establish a reproducible comparison baseline first. This workflow needs neither BMS access nor a new service purchase; it needs usable records and someone who understands the accounting boundary. Savings and the causal impact of a particular intervention require separate assessment. Nadali has not tested this workflow in a hotel or measured its results. Sources checked on 3 November 2026.
