How to forecast with no history: a new centre, product or campaign
When there is no data to extrapolate: forecast from a driver, a contact rate, a shape and a handle time, staff the range not the point, and replace it fast.
Published
A forecast built on history is arithmetic. A forecast with no history is a set of assumptions written down so they can be corrected, and its job is not to be right but to be replaceable the moment actuals arrive. This guide is the method for the first six weeks of a new centre, product or campaign.
Start from drivers, not from contacts
Nobody knows the contact volume, but somebody knows the driver. Customers signed up, orders shipped, bills sent, users onboarded. Contacts are driver times a contact rate, and the contact rate is the assumption to argue about, because it is the one that exists elsewhere: a similar product in your own history, a comparable line, an industry benchmark with a range.
Write the driver forecast and the contact rate down separately. When the actuals arrive, you will want to know which one was wrong.
Add a handle time
Handle time for a new line is guessable from the nearest existing one plus an allowance for novelty: agents are slower on a new product, customers ask more questions, and the knowledge base is thin. Fifteen to thirty per cent above the comparable line for the first month is a fair start, falling as the AHT calculator shows where the time is going.
Pick a shape, not a number
The intraday and weekday shape is the part of a forecast that transfers best from elsewhere, because it is set by the customers’ lives, not the product. A consumer line will be double-peaked with a lunch dip; a business line single-peaked in the morning; a retail line builds to an evening peak. Borrow the shape from the nearest comparable line and use one of the planner’s presets if there is none.
Staff the range, not the point
With volume at plus or minus 30 per cent, the requirement is a range. Run the week planner at the low, middle and high figures and note the FTE at each. Then decide which end to staff and how to cover the gap to the other: overtime, borrowed agents, a later launch of the marketing that drives the volume. The planning decision is which risk you would rather carry, and it is much easier to make with three numbers than one.
Replace it fast
The first two weeks of actuals are worth more than everything above. Compare them with the forecast using the forecast accuracy calculator, find out whether the driver or the contact rate was wrong, and rebuild. By week six you have enough history for a real forecast, and the launch forecast should be retired, not adjusted.
Try the range
Run it at your low, middle and high volume and note the three FTE figures.
90% of rostered hours land on the curve.
- On the phones
- Shrinkage allowance
- Theoretical FTE
- 59.3
- Agent-hours per week
- 2,225.0 h
- Agent-hours per day
- 445.0 h
- Peak interval
- 53 at 11:00
- Average scheduled
- 37.1
- Contacts per day
- 3,000
Weekly scheduled agent-hours ÷ contracted hours.
5 open days.
37 on the phones for 187 contacts.
Across the open day.
15,000 a week.
Show the working
- 3000 contacts per day is the average open day: 15000 a week across 5 open days, spread over 8:00 to 20:00 in 30-minute intervals using the "Morning and afternoon peaks" shape.
- The day-of-week split is even, so every open day carries 3000 contacts.
- Each interval runs Erlang C for 80% within 20 s, adds agents until occupancy is at or below 85%, then divides by (1 − 30%) and rounds up.
- Mon's peak is 11:00: 187 contacts need 37 on the phones and 53 scheduled.
- Agent-hours per day = Σ scheduled × 30 ÷ 60 = 445.0 h; × 5 days = 2225.0 h a week.
- FTE = 2225.0 ÷ 37.5 contracted hours = 59.3. Shrinkage is already in the per-interval numbers, so contracted hours are the right divisor.
- At 90% roster efficiency, 59.3 ÷ 0.90 = 65.9 FTE to cover the curve with real shifts.
This is one typical day. Pebble WFM does every day from your forecast, then builds and publishes the roster. Free month, no card needed.
Common mistakes
- Forecasting contacts directly instead of driver × contact rate, which makes it impossible to say what was wrong.
- Using the mature handle time for the launch month.
- Staffing the middle of the range and calling the shortfall a surprise.
- Adjusting the launch forecast for months instead of replacing it with the actuals.
Where next
- How to calculate FTE from contact volume: the chain each of the three runs goes through.
- How to staff for a spike or campaign: the same method for a known uplift on an existing line.
Frequently asked questions
- How wrong will a first forecast be?
- Plan on plus or minus 30 per cent on volume in the first month and 15 per cent on handle time. That is not a failure; it is the width of the range you should be staffing. The first two weeks of actuals will narrow it faster than any amount of pre-launch analysis.
- Should I use the vendor's benchmark for contacts per customer?
- As one input, with the range it came from. Benchmarks are averages across products and customer bases that are not yours. Where you can, find the closest comparable in your own history: a similar product launch, a similar customer segment, and scale from that.
Stop doing this one interval at a time
Pebble WFM forecasts your demand, computes the staffing requirement for every interval, builds the roster and publishes it, with self-service for agents and a copilot that can do what a planner can. Explore a sample organisation on day one.