HomeBlogThe New Zealand Small Business Web Performance Index 2026
Original research

The New Zealand Small Business Web Performance Index 2026

We are measuring a thousand New Zealand business websites — speed, mobile usability, whether anything is being tracked at all, and whether the business shows up when someone searches for what it sells. This page publishes the method first, before we have a single result, so the method can be judged on its own terms.

Status: fieldwork open

  • Methodology published — 4 August 2026 (this page)
  • Sample frame locked — 15 August 2026
  • Measurement window — 18 August to 12 September 2026
  • Findings and open dataset — October 2026, published at this URL

Nothing on this page is a finding. The numbers below are thresholds and quotas, not results.

Why we are counting this

Every agency in the country, ours included, will tell you that a slow website costs you customers. Almost none of us can tell you how slow the average New Zealand business website actually is, or whether a plumber in Whangārei is worse off than a physiotherapist in Wellington, or how many small businesses are spending money on advertising while measuring nothing.

That gap matters more than it sounds. When a business owner asks "is my site bad?", the honest answer today is a shrug and a benchmark borrowed from a US study of ecommerce sites with a hundred times the traffic. New Zealand is a small market with its own hosting economics, its own agency landscape, its own long tail of sites built once in 2017 and never touched again. Overseas averages are the wrong yardstick and we all keep using them anyway.

So we are going to measure it. Not a poll, not a survey of what people think about their websites — a direct technical measurement of a thousand live sites, cut by industry and by region, published in full with the raw data attached.

We have a commercial interest in the answer, obviously. We sell website development and conversion work, and a study showing that most sites are slow is convenient for us. That is exactly why the method is on this page before the numbers are: so you can check whether we drew the lines to get the answer we wanted.

How the sample is built

The target is 1,000 sites, drawn as a stratified random sample across six industries and ten regions. Stratified rather than random-across-the-board, because a purely random draw of New Zealand businesses would be swamped by Auckland trades and tell us nothing about, say, Southland healthcare.

The sampling frame

The frame is built from publicly listed New Zealand businesses with a live, publicly reachable website on a .nz, .com or .co.nz domain. A business enters the frame if it has a physical service address in New Zealand, a website that returns a 200 status on both its root URL and at least one internal page, and fewer than roughly fifty staff — the small-and-medium band this study is about.

Enterprise sites, government departments, franchises operating on a shared national template, and anything that resolves to a social media page rather than a website are excluded. Franchises are excluded specifically because a single template replicated ninety times would distort a regional cut badly.

Industry and regional quotas

IndustryTarget nWhy it is in the study
Trades & home services250Largest SME web population; near-total reliance on mobile search
Professional services175High value per enquiry, historically low technical investment
Healthcare & allied health150Booking-driven, accessibility-sensitive, heavily local
Automotive150Competitive local search; strong existing benchmark from our own work
Retail & ecommerce150The only segment where speed is already a known revenue lever
Industrial & B2B125Long sales cycles, minimal tracking, largely unstudied

Regional quotas are proportional to business counts rather than equal, with a floor of thirty sites per region so that smaller centres remain reportable. The ten regions are Auckland, Wellington, Canterbury, Waikato, Bay of Plenty, Otago, Manawatū-Whanganui, Hawke's Bay, Northland and Southland. Regions falling below the floor after cleaning get reported as a combined "other regions" line rather than being quietly dropped.

Selection within each cell is by seeded random draw. The seed publishes with the dataset, which means anyone can reproduce the same draw from the same frame.

What we measure, and with what

Four things, in descending order of how much we trust them.

1. Field performance — real users, real devices

Chrome UX Report data via the PageSpeed Insights API, which reports what actual visitors experienced over the preceding 28 days. This is the closest thing to ground truth available, and it is the primary measure wherever a site has enough traffic to have a field record at all.

We record Largest Contentful Paint, Interaction to Next Paint and Cumulative Layout Shift, at the 75th percentile, for mobile and desktop separately.

2. Lab performance — the fallback

Most small New Zealand business sites will not have enough traffic for field data. For those, a Lighthouse lab audit through the same API, run on a simulated mid-tier mobile device on a throttled connection. Lab results are recorded in a separate column and never mixed with field results in the same average — a distinction most published "state of the web" reports get wrong.

3. Mobile conversion readiness

Static analysis of the homepage markup for the things that decide whether a mobile visitor can actually contact the business:

  • Whether a phone number exists as a tappable tel: link, and its pixel depth from the top of the mobile viewport
  • Whether an email address or enquiry form is reachable without leaving the homepage
  • Whether a viewport meta tag is present and correctly configured
  • Tap target sizing on primary calls to action
  • Whether the site is served over HTTPS with a valid certificate

4. Measurement and visibility

Whether the business can tell if any of this is working:

  • Presence of GA4, Google Tag Manager, a Google Ads conversion tag, or a Meta pixel
  • Whether a Google Ads tag exists without any conversion action configured on the page — spending money, counting nothing
  • Presence and completeness of LocalBusiness or Organization structured data
  • Whether a matched Google Business Profile exists, and whether its category, hours and website field are populated

Every measurement is taken from publicly accessible pages. Nothing requires a login, nothing touches a private system, and nothing is recorded that a visitor to the site could not see.

The thresholds, set in advance

These are fixed now, before any data exists, and will not move.

2.5sLCP pass threshold, 75th percentile mobile
200msINP pass threshold, 75th percentile mobile
0.1CLS pass threshold, 75th percentile mobile
3Metrics that must pass for a site to count as passing

A site "passes Core Web Vitals" only if all three metrics pass on mobile. Partial passes are reported separately rather than being rounded up, because a site that passes two of three is not a site that works.

The composite Index score runs 0–100 and weights the four measurement areas at 40 per cent field or lab performance, 25 per cent mobile conversion readiness, 20 per cent measurement, and 15 per cent local visibility. That weighting is a judgement call and we will say so in the findings; the component scores publish separately so anyone who disagrees with the weighting can reweight it themselves from the raw file.

What we expect to find

Stating predictions before the data lands is uncomfortable, which is the point. If we are wrong about these, that will be in the report too.

  1. Most sites will fail mobile Core Web Vitals. Our working expectation is a failure rate above half, driven by unoptimised hero imagery and page-builder overhead rather than hosting.
  2. The gap between industries will be larger than the gap between regions. A trades site in Invercargill and a trades site in Auckland are usually built by the same class of tool; a law firm site and an ecommerce site are not.
  3. Tracking will be the worst-performing category, not speed. We expect a substantial share of sites running paid advertising with no conversion action configured — the failure we find in nearly every audit we run.
  4. Structured data adoption will be under a quarter across the whole sample, and lowest in trades despite trades having the most to gain from it.
  5. Google Business Profile completeness will outperform website quality in every region, because profiles are easy and websites are not.

What we are deliberately not measuring

Design quality, because it is not measurable and any attempt to score it would be us dressing up taste as data. Content quality, for the same reason. Traffic or revenue, because we have no legitimate access to either and estimating them from third-party tools would import someone else's error into our numbers. Rankings, because ranking is personalised and location-dependent to the point where a single measured position means very little.

We are also not naming any business. Every result reports in aggregate. There is a version of this study that publishes a leaderboard of the worst-performing sites in the country and gets ten times the attention; we are not doing that, because the businesses in this sample did not volunteer and most of them are one bad month away from not being able to afford a fix.

Where this method is weak

Four places, named now rather than buried in a footnote later.

The frame is not a census. There is no complete public register of New Zealand SMEs with websites attached, so the frame is assembled from multiple public sources and will inherit their gaps. Businesses with no web presence at all are invisible to this study by definition, and that is a real limitation on any claim about "New Zealand small business" as a whole.

Lab data is not field data. A simulated mid-tier device on a throttled connection is a modelling choice, and it systematically differs from what real visitors on real New Zealand connections experience. We keep the two apart and report the split, but the sites without field data are the smaller sites, which means our field-data findings skew toward better-trafficked businesses.

Homepage-only analysis misses a lot. A fast homepage over a slow booking flow reads as a pass here. Extending to a second measured page is on the list for the 2027 edition.

Point-in-time measurement is fragile. A site mid-redesign during our measurement window gets recorded as whatever it was that week. At a thousand sites this washes out in aggregate, but it is noise in any individual row of the dataset.

The open dataset

The complete row-level results publish alongside the findings as a CSV under a Creative Commons Attribution licence. One row per site, with the business name replaced by an anonymised identifier and the domain removed, retaining industry, region, every raw metric, and the derived scores.

Also publishing: the collection script, the seeded sample draw, and the exact API parameters used. If our numbers are wrong, we would rather someone found out from the data than from a press release.

Use it for whatever you like — a competing analysis, a university assignment, an article arguing we got it backwards. Attribution to this page is the only condition.

Running this on your own site

You do not need us or this study to find out where your own site sits. The same field data we are collecting is free and public:

  1. Open the PageSpeed Insights tool and enter your homepage URL.
  2. Read the field data section at the top, not the lab score. The number in the coloured circle is a lab simulation; the four field metrics above it are what your actual visitors experienced.
  3. Check the mobile tab, not desktop. For most New Zealand service businesses the majority of enquiries start on a phone.
  4. Open your own site on your own phone, on mobile data rather than wi-fi, and try to contact yourself. Time it. That test finds more problems than any tool.

Our hour-long self-audit walkthrough covers the rest of it, and the speed and conversions piece explains which of these numbers actually move revenue and which are vanity.

Questions we have already been asked

Why publish the methodology before the results?

Because a method you only see after the numbers is impossible to judge. Publishing the sampling frame, thresholds and hypotheses in advance means we cannot quietly redraw the lines to make a more dramatic headline, and it gives anyone who wants to challenge the findings something concrete to challenge.

Will the raw dataset be released?

Yes. The full row-level results publish as a CSV under a Creative Commons Attribution licence at the same time as the findings, with individual business names removed and replaced by an anonymised identifier.

Are individual businesses named in the report?

No. Every measurement is taken from publicly accessible pages, but results are reported only in aggregate by industry and region. No business is identified, ranked or singled out.

Can my business be included in the sample?

The sample is drawn randomly within quotas, so it is not possible to opt in without breaking the sampling model. Anyone can run the same measurements on their own site using the PageSpeed Insights API, and the methodology on this page describes exactly how.

How often will the Index be repeated?

Annually, using the same sampling frame and thresholds so that year-on-year movement is meaningful rather than an artefact of changed method.

Working with the data before it publishes

If you are a journalist, researcher or industry body who wants early access to the findings under embargo, or a regional business association who wants the cut for your area, email info@webpress.co.nz with what you need and we will sort it out.

Related reading

Where to go next

Want your own site measured properly?

A paid audit puts ten working days into your accounts and gives you a prioritised plan you own outright — no retainer required. Or book a free 30-minute session if you just want a second opinion.

← All articles

Ready to make your marketing pay for itself?

Book a free discovery session. We'll audit what you're running, tell you what's leaking budget, and show you what we'd do differently — no obligation.