Manaz Afzal
Public sector · Housing repairs

Maintenance demand, forecast a week ahead, with an alert before the surge

A dashboard and alert service for teams that dispatch repairs. Built on 3,531,417 housing-maintenance requests to New York City's 311 line since 2022 and daily temperature. The forecast cuts weekly planning error from 30% to 21%, and for heating faults from 48% to 25%.

3,531,417requests, 2022 to 2026, 10 categories
21%forecast error 7 days ahead, against 30% for same-day-last-week
25%error on heating faults, against 48%; temperature does most of the work
79alerts a year from the three categories worth alerting on, about 1.5 a week

Executive summary

A week-ahead maintenance forecast cuts planning error from 30% to 21%, and for heating faults from 48% to 25%, on 3.5 million real repair requests.

21%forecast error seven days ahead, against 30% for same-day-last-week
25%error on heating faults, against 48%; temperature does most of the work
1.5alerts a week from the three categories worth alerting on

Situation

Housing repairs teams book contractors a week out from last week’s numbers, and learn about a cold snap from the phones.

Approach

A seven-day forecast per category from request counts and the temperature forecast, with an upper range, tested over a full year on NYC housing-maintenance requests.

Result

Heating faults are where the forecast earns its keep. For steady categories a four-week average is as good, so each category runs whichever method won last quarter.

The business problem

A public or community housing provider’s repairs team books contractors and call-centre staff a week ahead from last week’s numbers. Cold snaps and storms change that overnight: housing providers have reported calls more than doubling in a hard winter week, with heating, plumbing and leak jobs leading. The team usually learns about the surge from the phones.

I built a seven-day forecast per repair category from request history and the temperature forecast, with an upper range and an alert for the categories that surge, and measured it against last week’s count.

What was built

A nightly job pulls requests by category from the maintenance system and the temperature forecast, refreshes a 7-day forecast per category with an upper range, and publishes a one-page dashboard. When a forecast or the day's running count crosses the range, it emails or posts to Teams for the operations lead with the numbers and the categories involved.

Maintenance systemTechnologyOne, Civica,or the asset systemNightly refreshCounts by category;temperature forecastForecast7 days ahead per categorywith an upper rangeDashboardOne page, Looker Studioor Power BIAlertEmail or Teams whena range is crossed
Nightly workflow. The alert names the categories and the expected excess, so the lead can book contractors the same day.

Results

Heating faults follow the thermometer

02,0004,0006,000Dec 25Feb 26Apr 26Jun 26Aug 26Oct 26RequestsForecast, 7 days ahead8 Feb 2026: 6,118 requestsforecast 3,149; alert raised
Test year, 2025-11-16 to 2026-11-15. Each forecast was made seven days before the day it describes, using counts up to that point and the day’s temperature. Data: NYC Open Data 311 Service Requests (HPD complaints); Open-Meteo historical weather
02,0004,000-15°C-9°C-3°C3°C9°C15°C21°C27°C33°CRequests per dayHeating season threshold(about 15°C)Daily mean temperature, New York (Open-Meteo), 3-degree bands
Below about 15°C, every three degrees colder adds several hundred heating requests a day. This relationship is why a temperature forecast beats last week’s count.

Every category, side by side

Heat and hot water · 959/day4,390Error 25% · simple 48%Unsanitary condition · 387/day508Error 13% · simple 11%Plumbing · 236/day449Error 20% · simple 22%Paint and plaster · 192/day254Error 16% · simple 16%Doors and windows · 151/day240Error 18% · simple 16%Water leak · 137/day234Error 29% · simple 26%General · 126/day180Error 27% · simple 15%Electrical · 101/day158Error 21% · simple 20%Flooring and stairs · 87/day113Error 17% · simple 16%Appliance · 73/day94Error 19% · simple 16%
Small multiples, each on its own scale. Solid: actual weekly average per day. Dashed: the forecast. Errors are the test-year weighted absolute percentage error.
Best simple ruleModelHeat and hot water25%Unsanitary condition13%Plumbing20%Paint and plaster16%Doors and windows18%Water leak29%General27%Electrical21%Flooring and stairs17%Appliance19%
The model wins clearly for heating and modestly for plumbing; for steady categories a 4-week same-weekday average is as good or better. In use, each category runs whichever method was more accurate over the previous quarter.
CategoryPer daySame day last week4-week averageModelDays inside rangeAlerts
Heat and hot water95948.4%47.9%25.1%97.3%10
Unsanitary condition38712.4%10.9%12.8%86.6%49
Plumbing23622.4%22.5%20.1%90.4%35
Paint and plaster19217.8%16.1%16.0%84.7%56
Doors and windows15117.3%16.1%17.7%81.4%68
Water leak13728.7%26.0%28.7%90.7%34
General12618.0%15.0%26.8%58.1%153
Electrical10122.3%19.5%21.2%84.7%56
Flooring and stairs8718.0%16.1%16.7%93.2%25
Appliance7319.2%16.0%18.8%89.6%38

Error is weighted absolute percentage error over the test year: total absolute miss divided by total requests. "Days inside range" should be about 97.5% if the upper range is well set.

Alerts: useful in three categories, noise in the rest

General58% of days inside the range153Doors and windows81% of days inside the range68Paint and plaster85% of days inside the range56Electrical85% of days inside the range56Unsanitary condition87% of days inside the range49Appliance90% of days inside the range38Plumbing90% of days inside the range35Water leak91% of days inside the range34Flooring and stairs93% of days inside the range25Heat and hot water97% of days inside the range10
The upper range was set on the six months before the test year. Where demand drifted upward during the year (General, Doors and windows), the range was too low and alerts fired weekly.

Alerting on every category would have raised 524 alerts in a year, too many to act on. Limiting alerts to heat and hot water, plumbing and water leaks gives 79, about 1.5 a week. The largest fell in winter cold snaps, when extra contractors need booking:

DateCategoryRequestsForecastAlert threshold
Sun 8 Feb 2026Heat and hot water6,1183,1494,837
Sat 24 Jan 2026Heat and hot water5,6142,8954,513
Sun 25 Jan 2026Heat and hot water5,6012,9334,561
Sun 1 Feb 2026Heat and hot water4,6882,4313,914
Thu 2 Jul 2026Plumbing990213299
Fri 30 Jan 2026Heat and hot water5,2002,9234,550
01,0002,0003,00020222023202420252026Heat and hot waterPlumbingWater leak
Seasonal shape over five years: heating requests peak every January; plumbing and leaks are flatter with occasional spikes.

Data and method

Requests
NYC 311 Service Requests, agency HPD, 10 housing-maintenance categories, 2022-01-01 to 2026-11-15
Scale
3,531,417 requests, aggregated to daily counts
Weather
Open-Meteo archive, daily mean and minimum temperature for New York
Test
Last 365 days (2025-11-16 to 2026-11-15); calibration of the range on the 180 days before

One gradient-boosted regressor per category predicts the count seven days ahead from lagged counts (7 to 14, 21, 28, 35 and 364 days), rolling means known at forecast time, weekday, day of year, holidays and temperature. The upper range is the 97.5th percentile of scaled errors on the calibration window. Baselines: the same weekday last week, and the average of the same weekday over the last four weeks.

Assumptions and limitations

  • Observed temperature stands in for the forecast. A real 7-day forecast adds error, largest on the coldest days
  • This is New York, with mandated heating. An Australian rent roll will see heating faults only in Melbourne, Canberra and Hobart winters; leaks and storm damage matter more
  • Upper ranges drift when demand trends upward; they need recalibrating monthly, which the test year shows plainly
  • The model forecasts requests, not jobs: duplicates and multi-request jobs are not separated in this data

What a housing repairs team would do next

  • Week 1: export two years of maintenance jobs by category and date from the asset or repairs system
  • Week 2: fit per category, pick model or simple rule per category on the last quarter
  • Weeks 3 to 4: run the dashboard and alerts alongside current planning
  • After a month: keep alerts only where they led to a booking decision

Financial impact

About 1,750 contractor hours a year better placed for a provider with about 1,000 homes

Value is fewer surge days caught short and fewer idle on-call shifts, from a smaller weekly planning miss.

InputValueSource
Jobs a week50 a housing provider with about 1,000 homes
Planning error: last week30%Measured
Planning error: forecast21%Measured
Contractor hours per job7.5 including travel

Calculation

  1. Jobs mis-planned a week, before15
  2. Jobs mis-planned a week, after10.5
  3. Hours better placed a yearabout 1,750

Contractor hours better placed a yearabout 1,750 hours

The study measures a large city; a small portfolio is noisier and the error is re-measured on its own history before use.

Implementation and rollout

Five weeks from job export to a Monday dashboard and a same-day alert

  1. Week 1

    Connect

    Export two years of jobs by category from TechnologyOne, Civica or SAP, plus the BoM forecast.

    Operations lead, Manaz

  2. Week 2

    Backtest

    Forecast and alert thresholds tested on the last year against last-week planning.

    Manaz

  3. Weeks 3 to 5

    Run alongside

    Monday dashboard and Teams alerts beside the current spreadsheet.

    Dispatchers

  4. Then

    Tune

    Quarterly choice of method per category; alerts trimmed to the categories that matter.

    Operations lead, Manaz

Tools

TechnologyOne, Civica or SAP exports, BoM or Open-Meteo temperature forecast, Python forecasting with ranges, nightly, Teams or Outlook alerts and a one-page dashboard

Team

  • Operations or maintenance lead books contractors from the forecast
  • Dispatchers act on alerts and record surges
  • Manaz pipeline, forecast, alerts and quarterly review

The timeline is a plan for a typical engagement; it depends on data access and how many systems are involved.

Risks and mitigations

What could go wrong in production, and the control for each

RiskMitigation
Alert fatigue from steady categoriesAlerts limited to heating, plumbing and leaks, about 1.5 a week.
Small portfolios give noisy countsWeekly rather than daily forecasts below 50 jobs a week.
Resident data leaving the source systemDashboard holds counts only; no resident details.
Weather forecast wrong a week outForecast refreshed nightly; the alert fires on the running count too.

Key takeaways

Key takeaways

  • Temperature drives heating faults; below about 15°C each few degrees adds hundreds of requests a day in a large city.
  • For steady categories a four-week average is as good as a model.
  • Ranges matter as much as point forecasts for on-call planning.
  • Counts only: the service never needs resident details.