Plan for weather in two separate tracks: the operational one, where you sequence weather-sensitive work and control cold and heat exposure, and the contractual one, where entitlement to time depends on notice and a contemporaneous record. AI is useful for keeping that record complete and for re-sequencing when a week is lost; it does not create entitlement.
Key takeaways
A Canadian build loses days to weather every year, and the difference between a well-run job and a badly-run one is not that one had better weather. It is that one planned weather-sensitive work into the season it belonged in, kept a daily record, and gave notice when a delay occurred.
Split the problem. Operationally, weather changes what work is possible, how fast crews work and what protection they need. Contractually, weather may or may not entitle you to an extension of time, depending on what your contract says and what you did about it at the time. AI helps considerably with the first and only indirectly with the second.
Cold is the constraint most Canadian schedules are shaped around. The Canadian Centre for Occupational Health and Safety sets out control measures for cold environments, including the combined effect of air temperature and wind speed expressed as the wind chill index, and a work/warm-up schedule developed by the Saskatchewan Occupational Health and Safety Division that identifies how many breaks are recommended in cold conditions. CCOHS also cites ACGIH guidance that a work-warming regimen is needed at around -7°C or below, with heated warming shelters such as tents, cabins or rest rooms available nearby and used at regular intervals. Engineering controls it describes include radiant heaters, heated shelters, wind shields and insulated equipment handles; administrative ones include pacing work to avoid excessive sweating, scheduling heavy work for warmer periods and rotating workers. CCOHS’s own wind chill table gives the schedule-relevant number behind that guidance: in the −28 to −39 band it rates frostbite risk on exposed skin as high within 10–30 minutes of exposure, and once wind chill reaches −40 to −47 that window shrinks to 5–10 minutes — the same range where a work/warm-up regimen is called for, and a concrete trigger for pulling exterior crews rather than a judgment call remade fresh every cold morning.
In British Columbia, WorkSafeBC frames the same issue through the hierarchy of controls — elimination or substitution first, then engineering controls, then administrative controls, with personal protective equipment least effective and never used alone — and its cold stress guidance lists work rotation, training on cold stress symptoms and adjusting work pace among the administrative measures to consider.
Heat is the mirror problem and is under-planned on Canadian sites because the season is short. CCOHS documents the health effects of hot environments and the first aid they require. In practice the scheduling consequence is the same shape as cold: move the heaviest work to the cooler part of the day, and treat the mitigation as part of the plan rather than as a reaction.
What AI actually does with weather
Useful: capturing observed conditions in the daily record every day without anyone remembering; summarising a month of records into a delay narrative; re-sequencing the schedule when a week of exterior work is lost; drafting the notice for a person to review and send.
Not useful: forecasting your project. A general-purpose model has no live weather feed unless you give it one, and a forecast is not a plan.
Never: deciding whether conditions are safe to work in. That is a supervisor’s judgment, made on site, against the applicable occupational health and safety requirements.
Whether a weather day earns you time depends on your contract. Standard-form Canadian contracts are published by the Canadian Construction Documents Committee, whose catalogue includes CCDC 2 (Stipulated Price Contract), CCDC 16 (Guide to Changes in the Contract) and CCDC 20 (Guide to Construction Contract Administration). Read what your actual contract says about delay, notice periods and extensions of time before you need it, because nearly all of them require prompt written notice and none of them accept a claim assembled from memory afterwards.
Where weather delay turns into a dispute about who pays, it usually arrives as a change question — which is the territory covered in change order disputes with a contractor. The practical lesson from that territory is consistent: the party with the contemporaneous record wins the argument about what happened.
Construction management platforms already have the field for this. Procore’s daily log includes, among its entry types, “Observed Weather Condition Entries”, “Delay Entries” and “Manpower Entries”, and supports attaching photographs to entries so they populate the photos tool. The weather entries are observed and recorded by a person, not captured automatically, which is the point: what matters evidentially is what the site experienced, not what a regional forecast said.
The automation worth building is the one that makes the record complete. A prompt at the end of each shift, a dictated sentence about conditions and what it stopped, and a monthly summary assembled from those entries. A model can turn thirty short entries into a coherent chronology in seconds; it cannot invent the thirty entries.
A siding contractor in Thunder Bay loses eleven days over a February to a combination of wind and temperature. The general contractor asks why the elevation is behind.
Because each foreman dictated two lines at the end of every shift — conditions, what was attempted, what stopped, who was sent home — the contractor produces a dated chronology the same afternoon, cross-referenced to the manpower entries showing crews mobilised and stood down. The claim for time is then a matter of applying the contract to an agreed set of facts, rather than a dispute about whether the days happened.
The following season the same contractor sequences the north elevation into the autumn window and prices a winter heating and hoarding allowance into any work that has to run through January. That is planning; the log is only what made the case for it.
Not reliably, and be sceptical of any tool that claims a number. What is defensible is using your own historical daily logs to see which work types actually stopped and in what conditions, then sequencing next year around your own record.
Most experienced Canadian builders carry contingency for the season rather than for individual events, and state the assumption in the tender. What matters is that the assumption is written down somewhere both parties saw before signing.
It is corroboration, not a record. What carries weight is a contemporaneous site log saying what the conditions were at that location, what work was attempted, and what specifically stopped — ideally with photographs and the crew count for that day.
A person on site, applying the occupational health and safety requirements of the province and the employer’s own controls. No scheduling tool should be configured to make that call, and none should be presented to a crew as having made it.
A 30-minute call is enough to tell you whether AI pays for itself here.
