Website traffic as a proxy for private company revenue growth

Website traffic works as a proxy for the revenue growth of a private company in one case: the website is where the customer buys, so more visits can bring more orders. Everywhere else a rise in traffic can come from a campaign, a redirect, a change in the ranking or the season, and say nothing about sales. This post gives you a table by business model, a rank correlation test to run on listed companies of the same kind, four checks on any move in traffic and an illustrative worked example.

The glossary entry on traffic rank defines the measure.

When is website traffic a sound proxy for private company revenue growth?

Website traffic can be a sound proxy for a private company's revenue growth when three things hold: the website is the point of sale, the traffic is measured the same way in every period, and the move survives checks for a campaign, a redirect, a change in the source and the season. Elsewhere it measures attention.

A visit is not an order, and official statistics show how much selling happens away from a website. The US Census Bureau estimates that e-commerce accounted for 17.1% of total US retail sales in the second quarter of 2026, seasonally adjusted, in its release of 18 August 2026. Its definition counts a sale as e-commerce when the order is placed, or the price and terms are negotiated, over an online system, whether or not payment is made online.

Eurostat's e-commerce statistics cover EU enterprises with 10 or more employees or self-employed persons, in the sectors its survey covers. In 2024 they generated 19.49% of their total turnover from e-sales, which Eurostat gives as 8.39% from orders through websites or apps and 11.07% from EDI-type messages. Web sales to other enterprises and public authorities came to 4.34% of total turnover, and web sales to private consumers to 4.04%.

In both populations most sales by value are ordered away from a website. So the first question about any company is how it sells.

Sort the companies you follow into these rows before you read a traffic figure.

Website traffic and company revenue correlation: Similarweb, Semrush and academic work

We found no statistics office, regulator or central bank that has published a measured correlation between website traffic and company revenue. A coefficient computed on Similarweb or Semrush estimates, or reported in an academic study, describes its own sample, period and traffic source. Measure the relation yourself, on listed companies of the same kind as the one you study.

Similarweb's own account of its method is set out below. Read Semrush's on its own site before you use its figures, and ask any vendor the same three things: the inputs, the unit, and what happens to past figures when the model changes.

A coefficient also ages with its source. Tranco's methodology files Alexa's list under its deprecated lists and records that it was published until 1 August 2023, so a relation measured on Alexa ranks cannot be run again on new data.

A rank correlation test on listed companies of the same kind

1. Choose the peer set. Listed companies that sell the way the private company sells, to the same kind of customer. The SEC states that the SIC code in a company's EDGAR filings indicates its type of business, which gives a first cut to refine by hand.
2. Take reported revenue. The SEC's data.sec.gov APIs return the XBRL figures companies file in forms such as the 10-Q and the 10-K, with no API key, and the frames API returns one fact for each reporting entity for a calendar period.
3. Take one traffic measure from one source for the same periods.
4. Turn both into change on the same period a year earlier.
5. Rank the companies on each change and compute Spearman's rho. NIST's Dataplot reference manual defines it as the Pearson correlation computed on the ranks, with average ranks for ties: plus 1 is perfect positive correlation, minus 1 perfect negative, and 0 no relation.
6. Repeat for each quarter and keep every result. One quarter of a dozen companies is a small sample.

A rank correlation asks only whether two orderings agree. That suits traffic, where a band or a rank is an order to begin with and a visit count is an estimate. Plot the pairs as well, and read the result within its limit: NIST's handbook is plain that an association between two variables does not imply that one causes the other. Keep the filing date beside each reported figure, which keeps look-ahead bias out of a back-test.

Similarweb methodology for website traffic data: monthly visits, engagement and traffic share

Similarweb's data page lists four kinds of input: direct measurement from the analytics of websites and apps, anonymous traffic data from Similarweb products installed on devices, public data, and data from partners. Its models normalise and extrapolate them into estimates of visits, visitors and engagement, which its API returns by day, week or month.

The API documentation defines each metric. Visits and page views are estimated numbers. Unique visitors are estimated and are not deduplicated across devices. Pages per visit and average visit duration are averages, and bounce rate is the percentage of visits that view only one page. Traffic share, in Similarweb's own explanation, is the traffic to one website divided by the total traffic to a competitive set, and for a marketing channel it is the traffic the channel generates divided by the total traffic to the site.

Monthly visits give the level, and the change on a year earlier gives the series to rank. Engagement helps to separate a rise in visits that view several pages from a burst of single-page visits, and a channel's share shows whether paid traffic carried the rise.

A public ranking reports a place or a band where a vendor reports a count. One line on each, from its publisher's own page:

• Google's Chrome UX Report gives a popularity rank: a relative measure of a site's popularity within its dataset, by total navigations, published as a band.
• Cloudflare Radar gives a domain ranking generated from DNS queries to its 1.1.1.1 public resolver: an ordered top 100 and unordered buckets below it.
• Tranco averages the ranks of the provider lists it names over the past 30 days.

A count supports a percentage change. A band supports a crossing from one band to another, and nothing inside it.

Limitations of website traffic data for private companies, from public ranks to Similarweb estimates

Website traffic data has five limitations for private companies. It counts visits, not orders. Every source observes its own population or models an estimate, and Similarweb's documentation labels its visit figures as estimated. A rank is relative to other sites. A band hides movement inside it. And a company can sell through more than one site.

• Visits, not orders. The share of visits that become orders and the value of an order can both change while visits hold.
• A population or a model. A ranking drawn from one browser or one DNS resolver describes that population, and an estimate reflects its vendor's inputs and model.
• Relative position. A site can keep its visitors and still lose places when others gain.
• Coarse bands. A site can double its visits and stay in its band, or cross a boundary on a small gain.
• Several doors. Country sites, a shop on a separate domain and an app all carry demand that a single-site series misses. Reduce each company to all its websites first, as in domain matching.

A ranking also ends: Cloudflare Radar's largest bucket holds one million domains. Treat a company with no entry as unranked, never as zero.

Using website traffic data to compare the growth of private companies

To compare the growth of private companies with website traffic data, take one source for every company, sum each company's websites, compare each with itself a year earlier, and rank the companies by that change within a group that sells the same way. Then test every large move for the four causes below before reading it as growth.

A source changes: the release notes of the Chrome UX Report record that the October 2022 release split its rank bands into half steps. A site reported in the top 10,000 before that release could be reported in the top 5,000 after it with the same visitors. And NIST's handbook defines seasonality as periodic fluctuations, calls it common in economic time series and gives retail sales, which peak for the Christmas season, as its example.

Keep the group fixed as well: compare only the companies present in both periods, the rule behind a same-store cohort.

Worked example: Acme Robotics against two groups of listed peers

This example is illustrative. Every company, website and figure in it is invented.

Assumptions:

• Acme Robotics is a private company that sells robot kits to consumers through its own online shop at acme-robotics.example.
• Group 1 is six listed companies that also sell through their own online shops, shop-a.example to shop-f.example. Group 2 is six listed companies that sell industrial robots through a sales team, maker-a.example to maker-f.example.
• For each listed company you hold two changes for one quarter against the same quarter a year earlier: estimated visits, from one source, and reported revenue.
• No two companies tie, so Spearman's rho reduces to 1 minus 6 times the sum of the squared rank differences, divided by n times (n squared minus 1). With n = 6 the divisor is 6 times 35 = 210.

Step 1: the online shops. The rank within the group is in brackets.

The squared differences sum to 2. Rho is 1 minus 12 divided by 210, which is 0.94.

Step 2: the makers sold by a sales team.

The squared differences sum to 32. Rho is 1 minus 192 divided by 210, which is 0.09. In this group the order by traffic says almost nothing about the order by revenue.

Step 3: the four checks on Acme Robotics. Its main site had 400,000 estimated visits in the quarter a year earlier and 640,000 in this one, a rise of 60%.

• Redirect: during the year Acme Robotics redirected a second site, acmebots.example, to the main one. That site had 100,000 visits in the earlier quarter. Like for like, 500,000 visits became 640,000, a rise of 28%.
• Campaign: the three months stand at 200,000, 220,000 and 220,000, which sum to 640,000. No single month carries the rise.
• Source: the six shops moved in both directions, from up 42% to down 15%, so the figures show no shift common to every site.
• Season: both quarters are the same quarter of the year.

Step 4: the reading. At 28%, Acme Robotics stands second of seven by change in visits, between shop-a.example and shop-b.example. In a group where the two orders agreed at 0.94, that supports a place in the upper part of the group's order by revenue growth. It supports an order and no revenue figure, and six pairs in one quarter are a small sample: repeat the test over several quarters.

The monthly traffic tier in Fokals data

Web Traffic, a dataset of the announcements and scale product, delivers the monthly traffic tier of each website: the smallest tier the website falls in that month, where a smaller tier means more traffic. Each row names the month, the website's domain, its company and company ID, and carries an attribution. The data dictionary defines each field, and the sourcing statement is public.

A tier is a band on an ordinal scale, which is what the method above runs on: rank the companies you follow by tier, compare each website with itself across months, and count the crossings from one band to another. Each row carries the company ID, so a crossing reads beside other signs of growth for the same company: stated headcount over time in Employee Headcount, daily open, new and closed postings in Hiring Activity, and private capital raises in Company Funding. The guide to sourcing private companies for venture capital puts such signals to work together.

The weekly Market Series carry traffic up and traffic down families over 30-day and 180-day windows. The first compares the latest month with the one before, the second with the month six earlier, each on a same-store cohort.

Frequently asked questions

What this page says about the products it names was checked against their public documentation on 7 October 2026. Product and company names are trademarks of their owners. Fokals is not affiliated with them or endorsed by them.