Reading technology adoption as a signal of vendor revenue

If you cover a listed software vendor whose product leaves a mark on its customers' websites, the number of sites running it, and the net change in that number week by week, is a dated proxy for customer growth that you can read between reporting dates. This guide shows how to count installs and net adds from technographic data, how to set them against the customer figures a vendor reports, and how to read a web-visible measure so that it stays a sound signal.

Which vendors leave a footprint

Fokals detects technologies from page content, scripts, network requests, response headers and DNS records, and recognises 6,283 technologies in 68 categories. A vendor is a candidate when its product shows in one of those views: a script or pixel for analytics, advertising, consent management, chat or reviews, a commerce or content platform, a payments or buy-now-pay-later tool, or a mail or verification record in DNS. The marketing stack dataset holds Technology Stack, which lists each technology a site runs with the dates it was first and last seen, and Technology Changes, which holds a dated event for every adoption and removal.

The strongest candidates are products whose footprint sits on the customer's website: the visible layer of a commerce, chat, analytics, consent, payments or content product. The datasets identify a technology by its ID, name and category, not by its vendor's listing, so keep a small mapping table of your own from technology ID to the ISIN of the listed vendor, and review it whenever the catalogue changes.

The measures

Three properties of the events make them cleaner than the difference between two install counts. A website's first observation sets a baseline and writes no events, so an add is a change observed between two readings and never a site discovered with the tool already on it. A removal is confirmed before it is written, which keeps out tags that come and go with consent banners and tests. A new account ID under a technology the site already runs is a separate technology_id event, so you can count it as expansion within a customer and keep it out of the adds.

If the vendor sells a commerce or content platform, read the platform category as well. A replacement of one platform by another is recorded as a platform change, with the platforms before and after in before and after, so a migration to or from your vendor is dated and attributed to both sides.

select
  date_trunc('quarter', observed_at)                                as quarter,
  count(distinct case when change = 'added'   then company_id end) as companies_added,
  count(distinct case when change = 'removed' then company_id end) as companies_removed
from company_tech_events
where category = 'technology'
  and key = 'acme_chat'   -- the technology id from your mapping table
group by 1
order by 1;

Take the base from the same sites. The index grows every day, so a raw count partly measures coverage, and a rate is comparable over time only when it is taken over websites that were already observed before the window began. The weekly Market Series do this for you: technology added, technology removed and technology prevalence by technology, on a same-store cohort and over 7, 30 and 180 days, and the series metrics endpoint of the REST API lists the technologies that have a series. The guide to measuring an installed base reads them. The API's technology list reports, for each technology, the companies running it and its added, removed and net counts over 30 days on websites already watched.

An illustrative example

Acme Robotics is an invented listed vendor whose chat widget loads on its customers' websites. Every figure below is illustrative.

Suppose the vendor reports customer growth of 7% and 8% for the same quarters. The net add rate runs a little above it, and that gap is the first thing to explain: free plans, customers with several sites, or customers the vendor counts differently. What makes a signal is that the ratio between the two stays steady, so a break in the ratio is the finding. The vendor's own revenue and customer counts come from its filings and results, and the installs series is what you read between them.

Reading the series

Read the mix as well as the total. Market Series can be cut by industry, country and size band, which shows whether a technology's growth is coming from the vendor's core market or from somewhere new. Pair the install series with the vendor's own sales hiring in Sales Team Metrics (company_sales_weekly): installs that flatten while the vendor accelerates its sales hiring ask a different question from installs that flatten while hiring stops. The guide to tracking sales team build-out shows how to read that dataset.

Choose the denominator with care. A net add rate over every site suits a vendor whose product fits every kind of site, and the install base of a niche tool is small enough for a handful of sites to move the rate. Measure it against the matching industry cut of Market Series, and report the count beside the rate.

Match the vendor's definition of a customer. If it reports only customers above a certain size, or only paying customers, restrict the installs you count to the size bands that fit, since Market Series carry a size dimension built from each company's stated employees. A free tier shows up in installs and not in a paying-customer figure, so compare the free base and the paying base as two series.

Subtract the category as well. If every consent-management vendor adds sites in the same quarter because of a regulatory deadline, the signal is the vendor's share of those adds, not its add count. Set the vendor's adds against those of the other technologies in its technology category and watch the share. If the vendor sells more than one product, give each a technology ID in your mapping table and keep them apart where their pricing differs.

Testing it against reported figures

1. Align the periods. Use the vendor's fiscal quarters and the dates on which it reports, not calendar quarters.
2. Compare rates. Set the net add rate against reported customer growth for one vendor over time, and across vendors in one category by rank correlation.
3. Test the lead. Ask whether the net add rate in the first half of a quarter predicts the quarter's reported figure better than last quarter's figure does.
4. Respect the dates. An event is known from its observation time. The marketing stack is refreshed daily to weekly depending on the company, and a removal is confirmed before it is written, so build the lag into the test. Every observation is dated and written once, never revised, so the same test run today and next year returns the same answer. The guide to point-in-time datasets for back-tests covers the dating.

Decide in advance what counts as a pass: a rank correlation clearly above zero across the vendors you cover, and a lead that survives when you lag the events by the time a removal takes to confirm. Keep the vendors where it passes as your working list.

How to read an install count

• An install is a website. A website is a unit of count, not a contract. A company with five sites counts five times unless you count by company ID, and a free plan counts as much as the largest contract, so read price, plan mix and usage from the vendor's results.
• A detection shows presence. A detection records that the technology runs on the website. A vendor that changes how its tag loads can move its own series, so check the dates of any step change against the vendor's release notes.
• Companies are resolved. Each website belongs to one company ID, so a count by company is a count of distinct customers in the index, and a brand carries the identifiers of its listed parent.
• A new technology starts from a baseline. When a technology joins the catalogue it is observed first as a baseline on every website, so its arrival is never counted as adoption. Additions are listed in the changelog of the methodology.

Frequently asked questions

The queries and code on this page are examples to adapt. Test them in your own environment before you rely on them.