Measuring Speed From Where Your Customers Are

The defaults flatter

Speed testing tools have to choose a device, a connection, and a location, and the defaults are usually generous.

A desktop machine, an unthrottled connection, and a server in whichever region the tool prefers.

That produces a number that is true and is not about your customers.

The gap between that number and what somebody in your area on a phone experiences is frequently a factor of three or four.

The three settings that matter

• Device: mobile, not desktop.
• Connection: throttled, not full speed.
• Location: near your customers.
• Cache: empty, as a first visit.
• Repeat: three runs, not one.

The fourth is the one people forget. A test run twice in a row uses stored files on the second run and produces a much better figure, which is the experience of a returning visitor rather than of somebody arriving from a search.

Location matters more than expected

Since distance is time.

A request travelling to a server on another continent and back adds a measurable delay before anything begins, and it is paid on every file the page loads.

A site hosted far from its customers can be well built and still slow for them.

Test from the nearest location the tool offers to where your customers actually are, and compare that against the default.

Where the difference is large, the hosting location or a content delivery network is the thing to look at rather than the page itself.

For a local business serving one region, that is a straightforward decision with a clear answer.

Throttle it honestly

Since connection quality varies more than device speed does.

Most tools offer a simulated mobile connection, and it is worth using rather than treating as pessimistic.

Your customers include people in vehicles, in basements, in rural areas, and in buildings with poor reception.

A page that is acceptable on a throttled connection is acceptable for everybody, and one tested only at full speed has been tested for the easiest case.

Run both and note the difference, since the gap tells you how much of your page is weight rather than structure.

A worked example

A business tested their site and got a load time under two seconds, which they were pleased with.

Retested as a mobile device on a throttled connection from a location near their customers, the same page took just over nine seconds to become usable.

Most of the difference was images sized for a desktop being downloaded in full to a phone, and a font service adding a delay before any text appeared.

Neither showed on the original test because the connection was fast enough to hide both.

After fixing the images and the font loading, the realistic figure came down to about four seconds.

The flattering figure had barely changed, which is why it had never prompted anything.

Field data is the real answer

Where you have enough traffic for it to exist.

Some tools report measurements collected from actual visitors rather than from a test, which is the only figure describing your real audience.

That covers their devices, their connections, and their locations, without you configuring anything.

It lags by several weeks and needs a reasonable amount of traffic, so a small site may have none.

Where it exists, prefer it to any test result, and use tests for diagnosing rather than for judging.

Where it does not, a carefully configured test is the best available substitute.

Test on an actual phone as well

Since a simulation is not the same as the thing.

Load your site on the oldest phone anybody in the business has, on mobile data rather than the office network, standing outside.

That takes two minutes and produces an impression no number conveys: whether it feels slow, whether anything jumps, whether the text appears before the images.

Do it away from the premises, since the connection there is not representative and may be unusually good or bad.

The number tells you what to fix and the experience tells you whether it matters.

Compare against your competitors

A useful calibration, since speed is judged relatively.

Run the same configured test on two or three competitors' sites.

Being slower than everybody in your trade is a real disadvantage, and being similar to them means the effort is better spent elsewhere.

It also gives a realistic target, since a figure achieved by a similar business on similar hosting is achievable and an abstract ideal may not be.

Note the figures with the date, so the comparison can be repeated.

Note the conditions with the number

Since a figure without its settings is not comparable to anything.

Write down the device, the connection setting, the location, and the tool alongside the result and the date.

Otherwise next quarter's test is run with different settings by a different person and the comparison is meaningless, which is how a site appears to have improved or declined for no reason.

Keep it to one line per test in the same place you keep the quarterly record.

The consistency of the configuration matters more than which configuration you chose, provided it is a realistic one.

The counter-case

Pessimistic testing can mislead too.

Simulated throttling is an approximation and may be harsher than real conditions in your area.

Chasing a good score on the worst-case configuration can also mean optimising for a small share of visitors while the majority already have a fine experience.

And a modest difference matters less than most performance writing suggests, particularly for a business whose customers arrive knowing the name.

Configure the test to a phone, a throttled connection, and a nearby location, and check the real experience on an actual device.

What to do

1. Set the device to mobile.
2. Throttle the connection.
3. Choose a location near your customers.
4. Clear the cache between runs.
5. Run it three times and take the middle.
6. Check field data if you have any.
7. Load it on an old phone outside.

Step three is the one nobody changes, and for a business hosted far from its customers it is frequently the largest single factor in the figure.

Which numbers to watch is covered in measuring speed the way Google does.

Frequently asked questions

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.