Summary. Training video that streams over a store network competes with payment traffic and loses, and a video that stalls does not get watched. Research on 23 million video views found abandonment begins at about two seconds of startup delay. This paper explains why the store network is the wrong place to solve that, what resident video changes for IT and for the sales floor, and how to price the difference before you commit to a platform.

Two seconds is the entire budget

The tolerance for a video that does not start is far shorter than most training teams assume. Krishnan and Sitaraman, working with data from Akamai and the University of Massachusetts Amherst, analyzed 23 million video views by 6.7 million unique viewers across 216 million minutes of playback. Viewers begin abandoning a video if it has not started within roughly two seconds, and every additional second of startup delay raises abandonment by about 5.8 percent.¹

Interruption after playback starts is worse than it sounds. A viewer whose video froze for just 1 percent of its duration watched about 5 percent fewer minutes.¹ A small amount of buffering costs a disproportionate amount of attention.

That research was conducted on consumers watching video they chose. Your associate is watching something they were assigned, standing up, mid-shift, with a customer about to walk in. There is no reason to expect more patience, and every reason to expect less.

Put the two findings together and the operating requirement becomes clear. It is not enough for the video to arrive eventually. It has to start effectively instantly and never pause, because the penalty for either failure is not a delay, it is an abandonment.

The store network was never yours to spend

The instinct is to treat this as a bandwidth problem and buy more bandwidth. That misreads what a store network is for.

A modern store network carries point-of-sale terminals, self-checkout, payment processing, inventory scanners, cameras, guest Wi-Fi and digital signage, and it does so under a compliance regime that has tightened rather than relaxed. The new requirements of PCI DSS 4.0 took full effect on March 31, 2025, and the standard goes deeper than its predecessor on network segmentation, pushing retailers to isolate and regularly test sections of their network and to control privileged access at each boundary.²

Segmentation is not a nuisance to be worked around. It is what keeps the cardholder data environment small, which is what keeps audit scope, control count and compliance cost down. A training platform that asks to push large media files across that environment, or that needs its own exception, is asking the store technology team to make their most scrutinized system more complicated in exchange for a video about a new handbag.

Even where there is bandwidth to spare, it is usually reserved deliberately. Payment authorization and customer-facing systems come first, and they should.

“Just add bandwidth, it is cheap now”

Connectivity has genuinely become cheaper, and this objection is made in good faith by people who have seen the quotes. It fails for four reasons that have nothing to do with price per megabit.

The constraint is usually not the pipe. It is contention on the in-store wireless, the age of the access points, the physical layout of a stockroom with metal shelving, and the number of devices already associated. A faster line into the building does not fix any of those.

The weakest stores set the schedule. A fleet-wide launch has to work everywhere on the same day. Upgrading the average store does nothing for the rural location, the mall unit on a shared circuit, or the store whose upgrade is scheduled for next year.

Simultaneity is not solved by capacity. Every associate opening the same large file within the same hour is a spike, and spikes are sized by the peak rather than by the average that the bandwidth quote assumed.

The compliance cost is not a bandwidth cost. Whatever the line costs, putting a new class of traffic near the cardholder data environment buys you scope, testing and review, and those are paid in specialist time rather than in monthly rental.

The fair concession: on a modern, well-segmented network with a dedicated non-payment segment and current access points, streaming can work acceptably. If that describes every one of your stores, this is a smaller problem for you than for most. The question worth asking is whether it describes your worst fifty.

Why “it works on my phone” is not a fair comparison

The most common complaint a retail IT leader fields is that a colleague can watch video instantly on a phone, so training video should behave the same way. The comparison omits everything that matters. A phone on a carrier network is one device on a connection nobody else is depending on. A store is dozens of devices sharing a link that is also settling card transactions, and the training video arrives not as one person’s casual request but as the same large file requested by every associate in every location at once, because that is what a launch is.

Simultaneity is the part that breaks. Content that is fine for one associate is a spike when a fleet opens it the same morning.

What resident video changes

The alternative is to stop streaming. High-definition video is downloaded quietly in the background, stored in the device’s own memory, and played from there. The associate taps and it plays, because at the moment of playback nothing is being fetched.

• Playback stops depending on the network. The two-second budget becomes irrelevant, because there is no startup fetch to lose it on. This holds in a store with poor connectivity as well as a flagship.
• Mission-critical traffic keeps priority. Background transfer works with whatever capacity is genuinely spare, rather than competing with payment authorization at the worst possible moment.
• Launches stop being events for IT. A simultaneous fleet-wide release becomes a series of background transfers rather than a coordinated spike.
• The helpdesk queue changes shape. “The video keeps buffering” is a ticket generated by architecture, not by users. Removing the cause removes the ticket.
• Content can be pre-positioned. Material for Monday’s launch can be on every device by Friday, which means the launch date becomes a publishing decision rather than a network event.

This is the approach behind our patented delivery model, and it is the reason clients describe the experience as coming on “like a light switch, regardless of store bandwidth.”

The trade-off is honest and worth stating: resident video consumes device storage, and storage is finite, particularly on older or personal handsets. That makes retirement policy a real design question rather than housekeeping, and it is the one part of this model that needs active management. On personal devices it interacts with the whole device decision.

What it costs when video does not work

The cost is paid twice. Once in the minutes lost waiting for content that stalls, and again in the training that quietly stops being attempted, because an associate who has been burned by a buffering module does not volunteer for the next one. The second cost is the larger of the two and it never appears in a helpdesk ticket.

You can put a figure on it with the abandonment research above and your own player data.

1. Measure real startup time in your worst quartile of stores, not in the office. Ten stores, five attempts each, timed by hand if necessary.
2. Apply the abandonment rate. Each second beyond the first two costs roughly six percent of viewers.¹
3. Convert to associates not reached. That percentage, applied to the headcount in those stores, is the population your last launch did not brief.
4. Weigh it against the performance gap between your best and worst stores on the metric the launch was meant to move. Some part of that gap is simply people who never saw the content.

The second-order effect is harder to quantify and more expensive: every failed attempt teaches the floor that this channel does not work, and that reputation is carried into the next launch regardless of what you fix in between.

What to ask any vendor

1. At playback, is anything fetched from the network? If yes, the two-second budget still applies.
2. What happens when 400 stores open the same module within an hour of each other?
3. Does the content transfer yield to payment and customer-facing traffic, or compete with it?
4. What does the platform require inside the cardholder data environment? The correct answer is nothing.
5. Can a store with weak connectivity run the same content as a flagship, on the same day?
6. How is content retired from the device, and who decides? Storage is the constraint this model creates.

Ask the demonstration to be run on your own network, in your worst-connected store, on a Saturday. Every platform performs well in a conference room.

Put your own numbers through the ROI calculator to see what returning that time to the floor is worth.

References

1. S. Shunmuga Krishnan and Ramesh K. Sitaraman, Akamai Technologies and University of Massachusetts Amherst. Video Stream Quality Impacts Viewer Behavior, Proceedings of the ACM Internet Measurement Conference (IMC), 2012. 23 million views, 6.7 million unique viewers, 216 million minutes of playback. people.cs.umass.edu
2. PCI DSS 4.0 future-dated requirements took effect March 31, 2025, following the March 31, 2024 documentation and self-assessment milestone; the standard expands guidance on network segmentation and privileged access. Summarized for retail IT by BizTech Magazine. biztechmagazine.com