Phones call handling

Forecasting phone demand by hour and weekday

A spreadsheet method for GP surgeries to forecast calls by half hour and weekday from 12 weeks of data, then adjust for bank holidays, flu and winter.

Ava research teamPublished 7 min read

On this page
  1. Why forecast at all
  2. Step 1: collect and clean twelve weeks of data
  3. Step 2: build the weekday-by-half-hour grid
  4. Step 3: turn the grid into an index
  5. Step 4: adjust for known events
  6. Step 5: convert the forecast into staffing
  7. Step 6: check the forecast every week
  8. Forecast online requests the same way
  9. Common forecasting mistakes
  10. Frequently asked questions
  11. References

Key takeaways

  • Twelve normal weeks of call records are enough to build a useful weekday-by-half-hour forecast.
  • Remove bank holiday weeks and outages first; NHS England notes call volumes shift with working days, holidays and weekends[1].
  • Turn the grid into a weekday index so you can scale it for known events, such as the day after a bank holiday.
  • Plan Christmas 2026 now: Christmas Day is a Friday and the Boxing Day substitute holiday a Monday, a four-day closure for many practices[2].
  • Check the forecast against actual calls every week and adjust; a forecast that is never checked quietly drifts.

Why forecast at all

It is common to staff the phones on habit: the same number of people every morning, more on Mondays, fewer in the afternoon. That works until demand moves, and demand has moved a long way. GP appointments per 1,000 patients in England rose from 389 in January 2018 to 460 in January 2026[3]. NHS England's own analysis of an example practice found an average-sized practice can receive over 100 calls in the first hour of opening on a Monday[4].

GP appointments per 1,000 patients in JanuaryEngland, from NHS appointments data
GP appointments per 1,000 patients in JanuaryColumn chart. January 2018: 389 per 1,000; January 2026: 460 per 1,000.0 per 1,000200 per 1,000400 per 1,000600 per 1,000January 2018: 389 per 1,000389 per 1,000January 2018January 2026: 460 per 1,000460 per 1,000January 2026GP appointments per 1,000 patients in JanuaryColumn chart. January 2018: 389 per 1,000; January 2026: 460 per 1,000.0 per 1,000200 per 1,000400 per 1,000600 per 1,000January 2018: 389 per 1,000389 per 1,000January 2018January 2026: 460 per 1,000460 per 1,000January 2026

Source: Wessex Local Medical Committees (from NHS appointments data), 2026[3]

Show the data
ItemValue
January 2018389 per 1,000
January 2026460 per 1,000

A simple forecast does not need special software. It needs twelve weeks of call records, a spreadsheet and an hour of setup. The payoff is that rota decisions are based on when patients actually ring, and that you can see trouble coming before a bank holiday or the first cold week of winter, rather than finding out from the queue.

Step 1: collect and clean twelve weeks of data

Export raw call records from your telephony system for the last twelve complete weeks. You need, at minimum, the date and time each call joined the queue. Outcome and wait time are useful for later analysis but not essential for the forecast.

  • Remove any week that contains a bank holiday. Those weeks are useful later, for modelling holiday effects, but they distort a normal-week baseline.
  • Remove days with a phone or clinical system outage, a building closure or a training afternoon when the phones were diverted.
  • Decide whether to count calls or unique callers. For staffing, calls are usually the right measure, because each redial still has to be handled or dropped by the system.
  • Count only calls in opening hours. Out-of-hours calls that hear the closed message are a separate measure.
  • If you have fewer than eight usable weeks, wait and collect more before relying on the forecast.

Step 2: build the weekday-by-half-hour grid

In the spreadsheet, add two columns to the call records: weekday and half-hour slot. Then build a pivot table with weekdays across the top, half hours down the side and the count of calls in each cell. Divide each cell by the number of weeks of that weekday you kept. The result is the average number of calls offered in each half hour of a normal week.

Example forecast grid: average calls offered for a fictional 12,000-patient practice (selected half hours).
Half hourMonTueWedThuFri
08:009670646266
08:307455504852
09:005240383639
10:003428262528
14:002622212024
16:002219181820

Example: in the fictional grid above, Monday 8.00 to 8.30 averages 96 calls, about 1.5 times the Thursday figure at the same time. The 8.00 to 9.00 hour on a Monday carries more calls than the whole afternoon from 14.00 to 16.00 on the same day. It is a common shape, and it is why one fixed rota for every morning rarely works.

Step 3: turn the grid into an index

Raw counts are hard to adjust. An index is easier. Pick a reference day, usually the quietest normal weekday, and divide each day's total calls by that day's total. For example, Monday might come out at around 1.4 and Friday at around 1.05 against a Thursday of 1.0. Do the same for half hours within a day: divide each half hour by the day's average half hour.

The index lets you answer questions quickly. If you expect next Tuesday to behave like a Monday because Monday is a bank holiday, multiply the Tuesday baseline by your Monday index. If a local outbreak pushes demand up, you can scale the whole week by a factor and see which half hours exceed your staffing.

Step 4: adjust for known events

NHS England's statistical notes on NHS 111 point out that call volumes are affected by the number of days in a month, public holidays and weekend days[1]. The same is true of practices, and some of the biggest effects can be planned months ahead.

  • Day after a bank holiday: treat it as a Monday or heavier. In 2026, Christmas Day falls on Friday 25 December and the Boxing Day substitute bank holiday on Monday 28 December, creating a four-day closure for many practices in England and Wales[2]. Tuesday 29 December is likely to be one of the busiest days of the year.
  • Flu season: the 2026 to 2027 adult flu programme in England started on 1 October, with children and pregnant women from 1 September, and most eligible people should be vaccinated by the end of November[5]. Expect booking calls to rise in September and October unless invitations go out by text with an online booking link.
  • Start of school terms: childhood infections often rise in the weeks after children return, which shows up as more same-day calls from parents.
  • Practice changes: a new online form, a change to appointment release times or a local practice closing can all shift demand. Note the date so you can see the effect in later weeks.

Step 5: convert the forecast into staffing

A forecast in calls becomes useful when it is turned into people. Multiply calls in each half hour by your average call length to get the minutes of talk time needed, then divide by the minutes one person can realistically spend on calls in a half hour, allowing for wrap-up time and breaks.

Example: 96 calls in a half hour at an average of 3 minutes each is 288 minutes of talk. If one receptionist can handle about 22 minutes of talk per half hour once wrap-up is included, that is about 13 people-half-hours, so roughly 13 people for that slot, or fewer people plus call-back and a strong online route. Few practices can staff that, which is why the release rule and online options matter as much as the rota.

Our guide to how many receptionists you need on the phones explains the workload maths in more detail, including how to allow for call-backs and lunchtime cover.

Step 6: check the forecast every week

  1. Compare forecast with actual

    Each week, put the actual calls offered beside the forecast for each weekday. Note the difference as a percentage.

  2. Look for patterns in the error

    If Mondays are always higher than forecast, the baseline needs updating. If one week is off, look for a reason before changing anything.

  3. Roll the baseline forward

    Every month, drop the oldest four weeks and add the newest, so the forecast reflects current demand.

  4. Keep the event log

    Record bank holidays, outages, local news and practice changes beside each week. Next year, this log is what makes the seasonal forecast accurate.

Forecast online requests the same way

Online requests have their own daily shape. Many arrive in the evening and early morning, when the phones are closed, and land in the list at 8am alongside the phone peak. Build a second grid from your online consultation system's export, using the same weekday and half-hour layout, but forecast when requests become workable rather than when they were submitted.

Put the two forecasts side by side. A practice that staffs the phones well at 8am but leaves the overnight online requests for the same people to read will still struggle. It often works better to have one or two people dedicated to processing online requests from 8am, separate from the phone team, so neither queue starves the other.

Common forecasting mistakes

  • Using a monthly total. It hides the weekday and half-hour pattern that matters for staffing.
  • Including bank holiday weeks in the baseline, which makes normal Mondays look quieter than they are.
  • Forecasting only calls answered. If abandonment was high, answered calls understate demand.
  • Ignoring online requests. If online requests rise, phone demand may fall, and the staff needed to process requests moves to a different queue.
  • Building a perfect model and never using it. A rough forecast that shapes the rota is worth more than an accurate one in a drawer.

The forecast sits on top of good weekly reporting; see reading telephony reports for that routine. To see whether the forecast helped, compare abandoned call rates before and after you changed the rota.

Frequently asked questions

How many weeks of call data do we need for a forecast?

Twelve normal weeks is a good starting point. Eight is a minimum; fewer than that and one odd week can skew the result.

Should we forecast unique callers or calls?

For staffing, forecast calls, because each one still arrives at the phone system. Track unique callers separately to understand how many patients are trying to get through.

Can the PCN forecast for all its practices?

Yes, if the practices share a telephony platform. A PCN-level forecast helps plan shared overflow, but each practice still needs its own grid because list sizes and patterns differ.

How far ahead should we forecast?

Use the weekly grid for the next four to six weeks, and a seasonal view, built from last year's event log, for winter and the Christmas period.

References

  1. NHS England (2026). Statistical Note: Integrated Urgent Care Aggregate Data Collection (IUC ADC) July 2026 (opens in a new tab) ↑

  2. GOV.UK (2026). UK bank holidays (opens in a new tab) ↑

  3. Wessex Local Medical Committees (from NHS appointments data) (2026). Appointment Overview January 2026 (national) (opens in a new tab) ↑

  4. NHS England and DHSC (2023). Delivery plan for recovering access to primary care (opens in a new tab) ↑

  5. GOV.UK (UKHSA, DHSC and NHS England) (2026). National flu immunisation programme 2026 to 2027 letter (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo