How to track enterprise technology spending with alternative datasets

A spending figure for enterprise technology arrives late and in aggregate: a vendor reports revenue with its periodic results, and a statistics office publishes investment after the quarter closes, then revises it. To track enterprise technology spending trends between those dates, research teams license alternative datasets, each observing a different step of a purchase. This guide sets out what each kind measures, builds three measures from adoption and hiring data with their queries (net adoptions on a fixed cohort, roles that name a tool and weekly series by industry), and names the official statistics that carry the amounts.

Which alternative datasets track enterprise technology spending trends?

Seven kinds of dataset track enterprise technology spending trends, and each measures something different. Technographic data counts companies adopting and removing a tool. Job postings count roles that name it. Payment panels, invoice records and public procurement records carry amounts for the buyers they cover. Official statistics and company filings carry the totals, after the period closes.

The first two rows are counts of companies and roles, each dated to the day it was observed. The other five are amounts or survey totals, each for the buyers its source covers. Read the counts for direction and timing and the amounts for level, and test one against the other.

How job postings and technographics data track cloud, software and IT spending trends

Job postings and technographics data track cloud, software and IT spending trends by direction and breadth. An adoption dates the week a company starts running a tool, a removal dates the week it stops, and a posting that names a platform dates an employer's decision to staff it. Each is a count, read over a fixed cohort of companies.

Fokals delivers both as company-level datasets joined on one company ID. The marketing stack dataset holds Technology Stack, with the first-seen and last-seen dates of every technology a company runs, and Technology Changes, with a dated event for every adoption and removal. The hiring dataset holds Job Postings: every role a company publishes, labelled by job function and seniority, with the tools it names.

The catalogue recognises 6,298 technologies in 69 categories. Those that bear on enterprise technology include workplace software, email providers, content delivery and security, hosting, CRM, analytics, and domain verifications with software and AI vendors.

Net adoptions on a fixed cohort

Net adoptions are the companies that added a technology in a window, less the companies that removed it. Take both over a fixed cohort, a same-store cohort: companies on record before the window opened and still observed when it closed. A count over a growing set partly measures the growth of the set. A first observation sets a baseline and is never counted as a change, so every row of Technology Changes is a change seen between two dated observations.

-- the fixed cohort: enterprises on record before :from and still observed at :to
create table cohort as
select company_id
from company_technologies
where company_id in (select company_id from company_headcounts where employees >= 5000)
group by company_id
having min(first_seen_at) < :from
   and max(last_seen_at) >= :to;

-- companies adding and removing one technology in the window, on that cohort
select
  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 in ('technology', 'dns')
  and key = :technology
  and observed_at >= :from
  and observed_at <  :to
  and company_id in (select company_id from cohort);

The first statement builds the cohort from Technology Stack (company_technologies) and Employee Headcount (company_headcounts): companies with a stated headcount of 5,000 or more whose stack was first seen before the window and last seen at or after its end. The second reads Technology Changes (company_tech_events), where key holds the technology ID of a technology event. Events in the category dns carry a technology name and category as technology events do, and the query reads both. Additions less removals are the net adoptions, and dividing by the cohort gives a rate.

Roles that name a tool

A posting that names a platform is an employer stating in public that it will pay people to run it. Job Postings carries the software named in each role, and tool families as true-or-false labels, among them Snowflake, Databricks, AWS, Azure, Google Cloud and SAP. Two more labels mark roles that select tools and roles that own a budget. Measure the share of new roles that name the tool, so that a hiring freeze or wave across the cohort does not pass for a change in demand for one platform.

with roles as (
  select
    company_id,
    date_trunc('month', first_seen_at) as month,
    coalesce(labels->>'tool_aws', '')   = 'true'
      or coalesce(labels->>'tool_azure', '') = 'true'
      or coalesce(labels->>'tool_gcp', '')   = 'true' as names_cloud
  from job_postings
  where found_on_first_read = false
    and company_id in (select company_id from cohort)
)
select
  month,
  count(*)                                                  as new_roles,
  sum(case when names_cloud then 1 else 0 end)              as roles_naming_cloud,
  count(distinct case when names_cloud then company_id end) as companies_naming_cloud
from roles
group by month
order by month;

In Job Postings (job_postings), found_on_first_read marks a posting already published when a company's roles were first observed, so leaving those rows out keeps first_seen_at an opening date. The query reads an empty tool family as not named. For one product in place of a family, test the technologies list of the posting for its catalogue ID.

Weekly series by industry

Market Series carries adoption and hiring measures already aggregated: 28 families of weekly series across 21 industries, countries, company size bands and markets, over 7, 30 and 180 days, on same-store cohorts. Each row holds a count, the cohort, a rate, the previous window, growth and an index. The families include technology added, technology removed and technology prevalence for each technology with a series, the share of new postings in each job function, and the share that are AI roles.

-- :technology is a technology id from the catalogue
select
  a.dimension                                         as industry,
  a.unit,
  a.cohort,
  a.count                                             as added,
  coalesce(r.count, 0)                                as removed,
  a.count - coalesce(r.count, 0)                      as net_adoptions,
  100.0 * (a.count - coalesce(r.count, 0)) / a.cohort as net_per_100
from market_series a
left join market_series r
  on  r.metric         = 'tech_removed:' || :technology
  and r.dimension_kind = a.dimension_kind
  and r.dimension      = a.dimension
  and r.window_days    = a.window_days
  and r.as_of          = a.as_of
where a.metric         = 'tech_added:' || :technology
  and a.dimension_kind = 'industry'
  and a.window_days    = 30
  and a.as_of          = (select max(as_of) from market_series)
order by net_per_100 desc;

The query reads market_series for the latest as-of date, a Sunday, and the 30-day window. metric joins the family to the technology ID, cohort is the same-store base of the row, and unit names the unit of the rate, per 100 companies or per 100 websites. Set dimension_kind to size for the bands of stated employees, where 1001-5000 and 5000+ are the enterprise rows, or to country. Each row is written once for its as-of date and never revised.

A worked example: one data platform over one quarter

Acme Robotics is an invented listed vendor of a data platform, which its customers verify their domains with and name in their job postings. The example and every figure in it are illustrative.

Assumptions:

• The cohort is 2,000 enterprises, fixed for the quarter as in the first query.
• At the start of the quarter 240 of them run the Acme platform. Summed over every platform in its category, installs are 800.
• During the quarter 46 companies add the Acme platform and 14 remove it. Summed over the category, net adoptions are 80.
• The cohort opens 18,000 new roles in the quarter, and 396 name the platform. In the quarter before, it opened 17,500, and 315 named it.

The base grew by 32 companies, or 13.3%, on a cohort that did not change, so the growth is adoption and not wider coverage. The platform took 40% of the category's net adoptions against 30% of its installs, so it gained share within the category. The share of new roles naming it rose from 1.8% to 2.2%, so employers in the cohort are staffing it. Every result is a count or a ratio of counts. What customers pay is read, in total, from the vendor's reported revenue.

Job postings or credit card data for enterprise IT spending?

Job postings and credit card data observe different parts of enterprise IT spending. A card panel records the amounts of purchases paid by card. A posting records that an employer is staffing a tool, with a date and no amount. For contracts settled by invoice and bank transfer, postings and adoption data date the decision, and filings give the amount.

The Federal Reserve Payments Study's findings for 2024, which cover consumers and businesses together, put cards at over three quarters of US noncash payments by number and ACH transfers at almost three quarters by value. A card panel samples the most frequent kind of payment, while most of the value moves by ACH transfer. Before you rely on card data for a technology vendor, put three questions to the provider:

• Which payment methods does the panel record, and are its accounts consumer, small-business or corporate?
• What share of the vendor's reported revenue did the panel capture in past quarters, and how steady was it?
• Is the set of accounts held fixed between periods, or does it grow?

A posting records an intention to hire, published by the employer on its own careers page, and it is observed however the employer pays its suppliers. The guide to hiring data in equity research shows how to read postings between reporting dates.

Procurement, invoice and technographics datasets on enterprise software spending

Procurement and invoice datasets record what a buyer agreed to pay for software: a purchase order, an invoice or a contract award, in currency, for the buyers whose records are in the dataset. Technographics datasets record which companies run the software and when that changed. One gives the amount per buyer, the other the number of buyers.

Public procurement is the open part. The US Treasury describes USAspending.gov as the official source for spending data for the United States government, and its API gives access to spending on awards, including who received federal contracts and grants. For a vendor that sells to government, those awards are dated amounts, buyer by buyer.

For private buyers, ask a provider of invoice or purchase-order data where the records come from, how many buyers contribute, and whether that set is held fixed between periods. A panel that adds buyers shows rising totals, as a growing set of websites shows rising installs.

The two kinds combine as price and quantity: an invoice panel shows what one buyer pays, Technology Stack gives the installed base of companies running a product, and Technology Changes gives its net movement week by week. The guide to reading technology adoption as a signal of vendor revenue sets installs against the customer figures a vendor reports.

Official statistics for the amounts

Statistics offices publish the amounts. These are the series to hold beside an adoption measure.

They are revised. The Bureau of Economic Analysis normally publishes an advance, a second and a third estimate for each quarter, each built on more comprehensive source data, with the third near the end of the third month after the quarter closes. A dated adoption series is written once and never revised, as why we write each table once explains.

Employment is already an input. The Bureau measures investment in own-account software, the software an establishment develops in-house for its own use, as the sum of production costs, based mainly on Bureau of Labor Statistics data on employment and wages for occupations such as computer programmers and software developers. Its 2023 update drew on research that used online job postings to estimate how much of each occupation's time goes to that development. Hiring in software and data functions is therefore a natural series to read beside the official amount.

Cloud and AI are recorded as adoption. The last two rows are shares of firms, the same kind of measure as prevalence in a cohort. A Census Bureau report of 26 May 2026 put overall AI use among US businesses between 17% and 20% from December 2025 to May 2026, and at 37% for firms with at least 250 employees. Use such a share as an outside check on a prevalence you compute, cut by the same size class.

A test to run on sample data

Run these six steps on sample data for the companies you track.

1. Define the category. List the technology IDs and tool families that make it up, and name each one's vendor in a table of your own. The guide to mapping alternative data to a security master covers joining a company to its securities.
2. Fix the cohort. Build it once for each window, as in the first query, and record its size beside every rate.
3. Check the baseline. Confirm that the earliest observation of each company in the sample carries no adoption events.
4. Compute the three measures. Net adoptions, the share of new roles that name the tool, and the weekly series by industry and by size band.
5. Check against an official share. Compare the prevalence in your cohort with the official adoption share for the same size class, and explain the gap.
6. Set the series beside reported amounts. For each closed period in the sample, compare each measure with the vendor's reported revenue and with official industry revenue, using the date on which each figure was published. Decide beforehand what relationship you would accept as evidence, and keep a measure only where it holds.

The data dictionary defines every dataset and field used above, and the guide to point-in-time datasets for back-tests covers storing dated observations so that each test reads only what was known on its date.

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.

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.