Digital bottlenecks

Shopping malls. Some love ‘em, some hate ‘em. We tend to frequent Meadowhall outside Sheffield – or Meadowhell as it’s affectionately known around Christmas time. Its food court offers the usual mix of fast food and bars and, with a small person in tow, we often frequent McDonalds out of sheer laziness/ease on the eardrums.

Recently, it’s become bedlam.

Since the introduction of the self-service kiosks, the overall experience has deteriorated. To be fair to McDonalds, they’re not alone in this shift for the worse due to the march of technology that supposedly make our lives more convenient; more streamlined. So what’s wrong?

It’s all about throughput

With my manufacturing background, I have some perspective on why processes do or don’t work, and it comes down to a very simple concept: flow.

Before the introduction of the self-service checkouts, the store was arranged as follows:

• A full-width counter at the far end, with a handful of POS terminals for taking orders.
• Collection areas interspersed between the checkouts, with space to wait and exit once your order was complete.
• A central pillar with the condiments/napkin trolley adjacent to it. That’s a problem they can do nothing about due to limitations of the physical store space they have.
• Two optional “express” lanes down either edge of the store, with:
  • A single line in which to queue.
  • Their own dedicated operator who took orders.
  • Their own, separate dedicated payment kiosk to pay for the orders a few paces further on.
  • An area to wait for your order a few paces closer to the counter, and an exit path along the other side of the express lane barrier.

That was brilliant. It was fast, super efficient, and the dude who entered your order with his stylus was trained so well he could do it almost in real-time as you spoke. He verbally gave you an order number, you moved on to the payment kiosk, waited less than a minute to pay for your order that was already loaded onto the system, then waited less then a minute for your food. Out. Done. Fast. It was fascinating to see how slick the operation was.

One queue to rule them all

The reason it worked so well was pipelining. Anyone who’s been caught in a traffic jam knows that the biggest source of problem – besides dickheads constantly changing lanes – is when two lanes go into one. Merging is hard work and ripples back. It’s the problem that plagues road and roundabout design, and traffic light placement and sequencing. Get it wrong and nobody goes anywhere.

In McDonalds, funneling people into a single lane, having a superfast order point, a separate superfast checkout point, and dedicated staff serving those customers meant the entire process was seamless. No bottlenecks.

Even though the express queue often extended out of the store, the fact you were almost constantly moving forward every fifteen seconds or so gave the perception of speed. It delivered a good customer experience and was an optimised process, each station taking roughly the same amount of time to handle each part of the transaction chain.

Multiple queues to bind them

So what is there now? Why is replacing the two express lanes with twelve self-service kiosks so drastically wrong?

There are just two collection areas towards the centre of the counter, and a couple of part-manned kiosks either side if you wish to fight your way past the customers idly loitering to use the ill-placed kiosks and through the crowd of people milling to await their order number shouted from the counter.

The dude who was unbelievably talented at operating the POS now just calls out to herd people away from the very middle of the store so that you don’t spill anything on your tray as you fight your way back out.

You can be standing there waiting, waiting, waiting, five, six, seven, eight minutes for an order that once took three.

The reason there is now so much confusion and so much waiting is simple: the pipeline is broken. The contributing factors are:

• The kiosk software is not well organised, resulting in people unfamiliar with it having to wade through the illogical layout and ill-optimised categorisation (v2 introduced in the last few months is an improvement but still unfit for general purpose usage).
• The kiosks act as both order points and payment points, increasing the time spent by each customer at the screen, resulting in more waiting by others to find a free kiosk.
• A further bottleneck has been created, with twelve “order lanes” going into two “supply lanes”, resulting in far more waiting.

The input-output disconnect

This is not a new phenomenon. Back in my Marconi days, there was a technology called PDH that was developed – Plesiochronous Digital Hierarchy. Ignoring the stupid name, it was born of the idea that if you interleaved a whole bunch of stuff into a pipe, the parallelism would mean switching would go faster and you could route more traffic.

It didn’t. It was unscalable. The bandwidth was capped because nothing was synchronised. Data was coming into the pipe, having to wait for a slot, multiplexed, and then sent on its way, adding delays to delays.

Some bright spark analysed this and invented SDH, where the ‘S’ stands for Synchronous. It used consistent frame sizes, effectively creating a bunch of single lanes end-to-end that were synchronised throughout the network, making multiplexing faster and simpler.

It was more costly to implement but, most importantly, it had about eighty times the bandwidth capacity of PDH. All through, pretty much, just synchronising the pipeline from the input at one end to the output at the other.

That’s the only way to make an efficient system: control the flow from input to output. It’s counter-intuitive but by slowing down the entry and exit points, making the multiplexing more efficient and consistent, achieves not only general improvements at low-load times, but the ability to handle peak demand far more effectively.

So, McDonalds and other kiosk-providers, I offer you this nugget of wisdom: before you implement a system using the latest whizzy tech, ensure said tech is not only fit for purpose but analyse the impact it will have on your pipeline. And if the answer is that it’ll negatively affect the customer experience, run like hell and stick with the expertise of the dude with his stylus. The world needs more people like him and fewer automated kiosks.