Phygital Network Simulator

How the model works

It is a discrete-event simulation: each customer is an event with their own arrival time, transaction type, duration, and abandonment threshold. The engine steps from event to event (arrival, end of service, abandonment, break) throughout opening hours, and then serves anyone still inside.

All three scenarios run on exactly the same customers. Same arrivals, durations, and thresholds; the only things that change are the available capacity and the routing rule. Any difference between scenarios is therefore due to the service model, not to chance.

Demand: the proposed curves

Each branch receives arrivals following a non-homogeneous Poisson process. Daily volume is visits per position × positions × relative demand × day factor, with random variation by branch (±8% by default). That volume is spread across 30-minute slots according to one of four profiles:

The proposal is to combine profiles in the same network. Their peaks don’t coincide, and that offset is what lets a shared video contact center (VCCaaS) absorb overflow that a single branch can’t handle. The day factor (off-peak ×0.6, average ×1.0, peak ×1.35) reproduces the ratio between the lowest and highest day of the week observed in a healthcare client’s live video-call channel (January–July 2026).

The queue-ticket system

1. If a fixed position is free, the customer goes to the fixed position.
2. Scenarios 2 and 3. If no position is free, the customer is eligible for video, a kiosk is free, and the estimated wait at the VCCaaS is shorter than at the branch, the customer goes to the kiosk. The kiosk connects to the first available VCCaaS agent in the network.
3. Scenario 3. If the kiosks are busy and the wait exceeds the QR threshold (or the customer’s own tolerance), the QR is offered. If the customer accepts, they make the video call from their phone, with a maximum of 5 simultaneous calls per kiosk. Kiosk customers have priority in the VCCaaS queue: a free agent serves customers at a kiosk first, and only then those who came in via QR. The QR thus works as an overflow channel and doesn’t take agents away from the kiosks; you can switch back to first-come, first-served in Assumptions.
4. In any other case, the customer checks the estimated wait. If it exceeds their threshold, they leave without taking a ticket (abandonment at the door). If not, they wait; and if the actual wait exceeds their threshold × the ticket tolerance, they abandon the queue.

In addition, when a kiosk becomes free, the first eligible customer in the queue is called. When a VCCaaS agent has no one left in queue, the VCCaaS calls an eligible customer from the branch with the longest queue that has a free kiosk. And when a fixed position becomes free with no queue, it takes back that branch’s customer who was still waiting for the VCCaaS.

Estimated wait = (people ahead + 1) × average duration ÷ active positions. This is the figure the customer would see on the ticket display. At the VCCaaS, a kiosk customer counts only other kiosk customers as being ahead; a QR customer counts the whole queue.

Staffing and hours

Fixed positions serve from opening to closing, with a 60-minute lunch break staggered between 12:00 and 16:00 (about 25% of positions are unavailable during that window). VCCaaS agents work the full day with no break; this is a simplification. After closing, no one else enters and anyone already inside is served.

Costs

Total cost = in-person interactions × in-person rate + video interactions (kiosk or QR) × assisted remote contact rate + daily fixed kiosk cost (0 by default). A customer who abandons generates no interaction cost, but its margin is lost.

Two cost models

• Per-interaction cost (default). Applies the report’s rates to each interaction. Capacity has no cost of its own: a position or an agent only costs money when serving. Useful for comparing service level and unit cost, but not for sizing.
• Capacity cost. Each fixed position, VCCaaS agent, and kiosk has a daily cost, whether it is used or not. This is the model for deciding how many positions to deploy. By default it is calibrated with the report’s own rates: rate × interactions per position per day in Scenario 1 (about 30), i.e., 410 USD per fixed position and 220 USD per agent.

Levers and sizing matrix

The Levers tab adjusts cost, value per interaction, and capacity (fixed positions and kiosks per branch, VCCaaS agents) all at once, and recalculates all three scenarios on the same day. Adjusting positions changes only capacity, never demand: branches receive the same customers. The matrix tests 42 combinations of fixed positions and kiosks and marks as “best” the one with the best result among those that don’t exceed the maximum acceptable share of lost customers.

Customer value and net result

The report doesn’t publish what a served customer is worth, so the model handles it in two layers.

• Break-even value per recovered customer. (Scenario cost − Scenario 1 cost) ÷ (lost in Scenario 1 − lost in the scenario). It is the minimum margin each recovered customer must bring in to cover the additional service cost. It doesn’t depend on any value data.
• Margin generated. Simple transactions × direct margin + advisory transactions × conversion × margin per sale. On video, conversion is multiplied by the video conversion factor.
• Margin lost to abandonment. Lost customers × their expected value. It is an upper bound: some of those customers may come back another day.
• Net result. Margin generated − total service cost.

The +33% conversion cited in the report compares video-advised sales with voice or unassisted channels, not with in-person service. That’s why the model assumes the same conversion on video as in person.

Metrics

• Average wait. Counts only customers who were served. Those who left on seeing the queue record no wait, which is why a low average wait can coexist with many abandonments.
• Fixed-position utilization. Time serving ÷ time available (opening hours minus lunch). Idle hours are position time with no customer.
• Simultaneous Service Ratio. Customers being served ÷ people present in the branch, at each branch’s peak hour, averaged across the network. With fixed positions only, it can’t exceed 1.
• Branches per agent. Average number of different branches each VCCaaS agent served during the day.
• Cost avoided vs. in person. Video interactions × (in-person rate − remote rate).

Model limitations

• Rates are market medians (mostly from English-speaking markets) compiled in the report, not measured costs from a specific network.
• The report doesn’t publish a cost per video call. The cost of an inbound contact center call is used as a reference; replace it with the actual cost of the service.
• The abandonment threshold applies equally to all branch profiles. Retries, appointments, and extended kiosk hours (Scenario 4) are not modeled.
• VCCaaS agents are dedicated to this network only; they don’t share a queue with other campaigns.

Sources for assumptions