Door count data is the starting point for serious staffing decisions in telecom retail, but the number by itself does not tell you how many people to schedule. Traffic volume is one input. You also need conversion rate, average transaction duration, and the number of visitors who require attention without completing an activation. Put those together correctly and you get a staffing model that reflects actual demand. Leave any of them out and you are guessing at a slightly higher confidence level than you were without the door counter.
This post walks through how to build a basic staff forecast from door count data for a single telecom franchise store. The math is straightforward. The harder part is making sure the data inputs are clean, because garbage door counts produce garbage schedules just as surely as no data at all.
The four inputs you need
Raw door count by half-hour interval. A door counter that gives you a single daily total is minimally useful for forecasting. You need intraday distribution, because a store with 80 visitors in a day is understaffed at 2pm if 45 of those visitors came between 1:30 and 3:00. Most door counter systems log at 15-minute or 30-minute intervals. If yours gives you hourly totals, that is workable but less precise.
Conversion rate: the share of visitors who complete a transaction. In Korean telecom retail, conversion rates vary substantially by store type and day of week. A shopping mall store in a commercial district typically runs 15 to 25 percent on weekdays: roughly 1 in 5 to 1 in 6 visitors leaves with an activation, accessory purchase, or plan change. A residential district store may run higher on weekends when customers are making planned purchases rather than browsing. If you do not have a historically calibrated conversion rate, start with 20 percent as a reasonable baseline for a generalist urban location and adjust based on what your own data shows over time.
Average transaction duration in minutes. This is the time from greeting to closing a transaction, including the POS entry and any paperwork. An activation (especially 번호이동 or 신규 with plan selection) typically takes 25 to 40 minutes. An accessory-only purchase runs 5 to 10 minutes. A plan inquiry that does not result in a sale takes 10 to 20 minutes and still consumes staff time even though it does not count as a conversion. Blended average transaction duration for a store with a typical activation mix is usually 20 to 30 minutes. Track your own by timing a sample of transactions over two weeks; self-reported estimates from staff tend to run low.
Non-transaction engagement rate. Some share of visitors require attention (asking about plan pricing, checking on a device they ordered) without completing a transaction. This time is real staff time that does not appear in your activation count. In most locations, non-transaction visits consume 10 to 20 percent of total staff engagement time. If you do not include this in your model, you will systematically underestimate required staffing on heavy-traffic days.
Building the forecast
With these four inputs, the basic calculation for required staff capacity in a given time window looks like this:
Take the expected door count for the window (say, 1:00pm to 3:00pm on a Saturday = 35 visitors based on historical data for that slot). Apply conversion rate to get expected transactions: 35 x 0.20 = 7 activations or purchases. Apply average transaction duration: 7 x 30 minutes = 210 minutes of transaction time.
Add the non-transaction engagement load: 35 x (1 - 0.20) = 28 non-purchasing visitors, at an average 12 minutes of engagement each = 336 minutes. But not all non-purchasing visitors need engagement: some browse and leave without requesting help. A realistic non-transaction engagement demand in Korean telecom retail is about 30 to 40 percent of non-purchasing visitors, so 28 x 0.35 = 10 visitors requiring engagement x 12 minutes = 120 minutes.
Total staff time required for the 2-hour window: 210 + 120 = 330 minutes. Divide by 120 minutes (the window length): 2.75 staff required to handle demand at full capacity. Round up to 3 for that window, or 2 if you accept a modest wait time for some customers.
This is a simplified version of the model. Real deployments would use smoothed historical door counts rather than a single past-Saturday data point, adjust conversion rate for whether a promotion is active that weekend, and apply different transaction durations by activation type rather than a single blended average. But the structure is the same: traffic volume x engagement rate x time per engagement = staff minutes required per window.
Where the model breaks down
There are two situations where a door-count-based staffing model produces bad outputs even with good input data.
The first is promotion launch days. When a carrier launches a high-profile 번호이동 incentive or a new flagship device goes on sale, door count can spike 50 to 100 percent above the prior-week comparable without any advance signal in the historical data. The model cannot predict this from door count history alone. You need to supplement it with carrier campaign calendar information, and schedule above the model's recommendation on any weekend where a major promotion is active.
The second is transaction duration variance on complex plans. When a carrier releases a new rate plan structure or bundles a promotional tariff with specific eligibility criteria, average transaction duration increases because staff need more time to explain the plan and process the eligibility check. A day where 40 percent of activations involve a complex new promotional plan will run significantly longer per transaction than a day with straightforward existing plan selections. The model built on historical average transaction duration will underestimate staffing need on those days.
We are not claiming these failure modes invalidate the approach. A model that accounts for 80 percent of staffing variance and fails on the remaining 20 percent is still a large improvement over no model. The point is knowing where to apply human judgment on top of the model's output, rather than scheduling from the model blindly.
Data quality prerequisites
The model is only as reliable as the door count data. Three things undermine door count quality and should be verified before building a forecast on the data.
Counter placement. A door counter placed inside the store rather than at the entrance may miss visitors who turn around immediately without entering fully. A counter too close to a window may log reflected movement. The counter should be at the entry point, triggered by actual threshold crossings, and calibrated to your specific door width.
Staff crossings. Many door counter systems count all threshold crossings including staff entries and exits. On a busy day with staff arriving, leaving, going to the back for inventory, and returning, staff crossings can add 10 to 20 percent to the raw count. Some systems allow a staff identification card to subtract from the count; without that feature, you need to manually estimate and subtract average staff crossings from the raw number before using it as a visitor count.
Gaps in the record. If the door counter loses power, the device reboots, or the data export fails for a day or more, the historical record will have gaps. Filling those gaps with interpolation rather than flagging them as missing will make your historical baseline appear smoother than it actually was, and the forecast built on it will behave unexpectedly when it encounters a real future day that matched one of those missing historical days.
Clean data with documented gaps is more useful than smoothed data that hides them.
Getting to a schedule from the forecast
A staffing forecast gives you required capacity in hours. Translating that into an actual schedule requires knowing the shift structure your staff can work and the minimum staffing floor below which you will not schedule (most operators maintain a minimum of one person in the store during all open hours regardless of traffic forecast).
For a store open 10:00am to 8:00pm, a 10-hour operating day with door count data broken into half-hour windows, the forecast output is a series of capacity requirements by window. From there, the scheduling question is which shift combinations cover the required capacity with the minimum total staff hours, given that most staff work 8-hour shifts and you cannot split shifts into 30-minute windows.
In practice this is a coverage optimization problem with a small set of shift options. For a two-to-four staff store, the number of viable schedule combinations is small enough that it can be worked through manually in 15 to 20 minutes once you have the forecast output. Automating the optimization step is useful for operators running multiple locations from the same scheduling process, where doing it manually for each location is not feasible.