CA$2.2 Billion of Canada's Federal IT Spending Is Not IT Spending

Update, 24 September 2026: the 2026–2027 plans make the case stronger

The analysis below uses the 2025–2026 Departmental Plans, the vintage in the open dataset when it was written. The 2026–2027 plans, tabled in March 2026, now sit in GC InfoBase's own programme files (Treasury Board of Canada Secretariat, 2026b). Run the same keyword match on them and it returns CA$7.34 billion across 94 programme lines in 37 organizations, and CA$2.74 billion of that — 37.3% — is not IT. It is the same kinds of line as below, plus a new one: the Natural Sciences and Engineering Research Council renamed its Research Partnerships programme Research and Technology Partnerships, so a CA$470.14 million research-grant programme now contains technolog and enters the IT table on its name alone (Treasury Board of Canada Secretariat, 2026b).

Renames push lines out as well as in. Shared Services Canada's Data Centre Information Technology Operations became Hosting Services, CA$801.95 million that no keyword matches, and its Telecommunications, Connectivity and Security lines (CA$782.26 million together) never matched either, although all four sit under the core responsibility Common Government of Canada Information Technology (IT) Operations (Treasury Board of Canada Secretariat, 2026b). So in 2026–2027 the name match over-counts and under-counts in the same run, and the core-responsibility column is what shows both. I count Hosting Services as the data-centre line's successor; the other three stay out of the headline figure, as they do in the 2025–2026 analysis below.

Checked that way, planned federal IT for 2026–2027 is CA$5.40 billion across 36 organizations. National Defence (CA$1,404.11 million) edges Shared Services Canada (CA$1,374.23 million) for first place. ISED falls from second in the raw table to nineteenth, at CA$47.50 million, and the research council from fourth to thirty-fifth, at CA$9.00 million (Treasury Board of Canada Secretariat, 2026b).

The number everyone will quote

The Government of Canada publishes planned spending and full-time-equivalent counts for every programme in every department, updated with each Departmental Plan, free and without registration (Treasury Board of Canada Secretariat, 2026). Filter that file for programmes whose names look like information technology and you get a federal IT market in one afternoon: CA$7.31 billion across 96 programme lines in 37 organizations, with 67.4% of it inside five of them and 83.7% inside ten.

Two things about that number before anyone builds a territory plan on it.

It is planned, not actual. All 96 lines carry a null in the actual_spending column and a figure in planned_spending_1, because audited actuals for fiscal 2025–2026 are not published yet. Every figure below is therefore what departments said they intended to spend, re-derived from a live fetch of the dataset on 8 September 2026.

And it is not total federal IT spending. Only 37 of the 261 organizations in the Sagentix federal organization census report a programme line the filter recognises Sagentix Phase 01 Market Intelligence, 2026. The rest either buy IT inside programmes named something else or do not report at that granularity at all. Anyone quoting CA$7.31 billion as the federal IT market is quoting the visible part of it and should say so.

The spine. The league table is produced by matching programme names against IT keywords. The same dataset carries a second field the method never reads — core_responsibility, which the Policy on Results defines as "An enduring function or role performed by a department" (Treasury Board of Canada Secretariat, 2016). It is the department's own statement of which function a programme belongs to, and the file itself carries Information Technology Services as an Internal Services line in department after department. Settle the name match against that field and CA$2.23 billion — 30.5% of the published total — is not information technology at all. It is a broadband subsidy, defence research, clean-technology programmes, a student internship stream — all swept in because their names contain digital, technolog or cyber.

The rubric that produced it

Here is the match, stated in full so the result is reproducible rather than merely asserted. A programme line counts as IT when its name matches:

information technology | information management | im/it | digital | cyber | ict |
technolog | data (and analytics|management|services) |
(it|technology|digital) infrastructure | chief information officer |
enterprise (architecture|services)

Nothing about that is unreasonable. It is what anyone would write, and I wrote it (in 2026, for a client engagement) — this is the classifier inside the federal organization census Sagentix runs on its own engagements, which is the reason I went looking. It finds the programmes you would want. It also misses some — CA$970.76 million of Shared Services Canada's own IT operations (Cloud, Telecommunications, Networks, Security) carry no IT keyword — but the bigger problem is everything else it finds (Treasury Board of Canada Secretariat, 2026).

The control, in both directions

A count from a keyword match is only as good as the match, and re-running a flawed one against a fresh download reproduces the wrong number perfectly. Reproducibility is not accuracy. So the rubric itself has to be the thing under test, with two lists: strings it must match, and adversarially chosen near-misses it must reject.

Eight positives — Information Technology Services, Data Centre Information Technology Operations, Workplace Technologies, Operational IM/IT Services, Digital Government Program, Cyber Operations, Information Management Services, and Defence's ICT systems acquisition line — all match. Those eight all pass.

Nine negatives, drawn from the programme names the match actually returned, all match too. Every one of them should have been rejected:

Programme name it accepted	Should it be IT?
Bridging Digital Divides	No
Science, Technology and Innovation	No
Energy Innovation and Clean Technology (EICT)	No
Quantum and Nanotechnologies	No
Collaborative Science, Technology and Innovation Program	No
Clean Technology and Clean Growth	No
Youth Employment and Skills Strategy — Science and Technology Internship Program (Green Jobs)	No
Science and Technology	No
Biotechnology and Genomics	No

Nine of nine over-matched. Nanotechnologies contains technolog. So does Biotechnology. Over-matching is the failure that ships, because an inflated count reads as a stronger finding while nothing in the output flags it — a broken filter and an honest one produce output of exactly the same shape.

This failure has a name, and a famous precedent

Nothing here is novel, and that is the useful part. Matching a word list against text to sort it into categories is what political scientists call a dictionary method, and its failure mode has been documented for over a decade.

Justin Grimmer and Brandon Stewart set out the problem in Political Analysis: "applying dictionaries outside the domain for which they were developed can lead to serious errors" (Grimmer & Stewart, 2013). Their diagnosis of why it keeps happening is the sentence worth pinning above a desk. "But measures from dictionaries are rarely validated," they write; standard practice "is to assume the measures created from a dictionary are correct" and apply them to the problem (Grimmer & Stewart, 2013, §5.1). That is precisely what I did, and I have been doing this long enough to know better.

The canonical worked example comes from finance. Tim Loughran and Bill McDonald tested the Harvard psychosocial dictionary — a general-purpose list of "negative" words — against 50,115 10-K filings, and found 73.8% of the negative word counts it produced came from words "typically not negative in a financial context", and found that "word lists developed for other disciplines misclassify common words in financial text" (Loughran & McDonald, 2011, pp. 35–36). The scale of it is the part that lands: in a large sample of filings from 1994 to 2008, almost three-quarters of the words the Harvard list flagged as negative "are words typically not considered negative in financial contexts". Their examples are ordinary. Depreciation, liability, foreign and board carry no tone at all in an annual report, and the Harvard list counted every one of them as negative (Loughran & McDonald, 2011, p. 61).

Liability is a bad word everywhere except an accounting statement. Technolog is an IT word everywhere except a file that also contains biotechnology, nanotechnology and clean technology. It is the same error, in a different domain, fifteen years later.

What the literature prescribes is also what saved this analysis. Grimmer and Stewart tell researchers to "directly establish that word lists created in other contexts are applicable to a particular domain, or create a problem-specific dictionary" (Grimmer & Stewart, 2013, §5.1). Loughran and McDonald built one — six word lists tuned to financial text. That is real work, and most of us will not do it for a market-sizing exercise.

Which is why the next section matters more than the literature does. In this case, someone had already built the problem-specific classification, published it in the same file, and left it in a column the method never read.

The second field in the same file

The fix was not a better regex. It was a column already sitting in the same rows.

Every programme record carries core_responsibility (a required field, not an optional one), and the Policy on Results defines a Core Responsibility as "An enduring function or role performed by a department" (Treasury Board of Canada Secretariat, 2016). Departments assign it themselves; Treasury Board prescribes the vocabulary. Until 10 October 2025 the Directive on Results also listed Information Management Services and Information Technology Management Services among seven Internal Services categories that carried mandatory indicators for large departments (Treasury Board of Canada Secretariat, 2016b, App. D, s. D.2.2.3–D.2.2.4); that standard is now rescinded, and the dataset still files both as Internal Services lines.

So the government has already answered the question the keyword match was guessing at. A programme filed under Internal Services, Common Government of Canada Information Technology (IT) Operations, Administrative Leadership, or Information Delivery and Services for Other Departments is the organization's IT function, on its own classification. A programme filed under Science and Innovation, People, Skills and Communities, Fisheries or Polar Science and Knowledge is not, whatever words are in its title.

Take the largest single line the match swept in. Bridging Digital Divides sits at Innovation, Science and Economic Development Canada, carries CA$949.39 million of planned spending against 105 planned FTEs, and is filed under the core responsibility People, Skills and Communities. The department's own planning note in the same record — the planning_explanation field on programme BNN03 — attributes the change to the winding down of the Universal Broadband Fund, which it states ends in 2026-27 (Treasury Board of Canada Secretariat, 2026). It is a broadband subsidy programme. No reasonable person would put it in a federal IT market, and the department never did.

What the second field removes

Eleven programme lines, across six organizations, totalling CA$2.23 billion and 3,011 planned FTEs (3,010.75 unrounded; the rounded rows in the table sum to 3,010):

Spend and staff	Programme (organization)	Filed under
CA$949.39M 105 FTEs	Bridging Digital Divides ISED	People, Skills and Communities
CA$729.65M 1,481 FTEs	Science, Technology and Innovation National Defence	Future Force Design
CA$330.14M 834 FTEs	Energy Innovation and Clean Technology Natural Resources Canada	Innovative and Sustainable Natural Resources Development
CA$56.96M 239 FTEs	Quantum and Nanotechnologies National Research Council	Science and Innovation
CA$54.60M 42 FTEs	Collaborative Science, Technology and Innovation Program National Research Council	Science and Innovation
CA$38.88M 30 FTEs	Clean Technology and Clean Growth ISED	Companies, Investment and Growth
CA$24.12M 170 FTEs	Digital Technologies National Research Council	Science and Innovation
CA$15.14M 6 FTEs	Youth Employment — Science and Technology Internship (Green Jobs) Natural Resources Canada	Globally Competitive Natural Resource Sectors
CA$13.58M 25 FTEs	Science and Technology Polar Knowledge Canada	Polar Science and Knowledge
CA$11.72M 51 FTEs	Digital Service ISED	Companies, Investment and Growth
CA$2.76M 27 FTEs	Biotechnology and Genomics Fisheries and Oceans Canada	Fisheries

(Treasury Board of Canada Secretariat, 2026)

The arithmetic closes: CA$4.12 billion of back-office IT, plus CA$955.30 million of mission ICT, plus CA$2.23 billion that is not IT, equals the CA$7.31 billion the name match reported (Treasury Board of Canada Secretariat, 2026).

One dataset, three instruments, three answers

• Programme-name keyword match (as published)CA$7.31B
• Name match, checked against core responsibilityCA$5.08B
• Internal Services and IT Operations lines onlyCA$4.12B

The ratio that started this

I did not go looking for a classification error. I went looking for a screening signal: planned spend per IT full-time-equivalent, which nobody publishes and which anyone can compute from two columns of the same file.

Start with what that quotient actually is, because the obvious reading of it is wrong. A programme, in this file, is "a group of related resource inputs and activities ... treated as a budgetary unit", and a full-time equivalent is "a measure of the extent to which an employee represents a full person-year charge against a Departmental budget" (Transport Canada, 2016). The people are charged against the same budget the spending column reports. So the quotient is not what an organization spends on each IT employee — not tools, licences and contracts per head. It is the programme's entire cost divided by its own staff, and for most IT programmes the staff are the largest single thing inside it.

Read properly, then, a figure near CA$0.13M a head is mostly salary, and a figure well above it is mostly something else: capital, contracts, or money going out the door as transfers. That is a labour-intensity measure. It was never a build-versus-buy measure, whatever I called it at the time.

Under the raw table it looked compelling anyway. Shared Services Canada came in at CA$0.267M per IT FTE against 4,791 staff. ISED came in at CA$1.804M against 575 — and against the Department of Justice's CA$0.061M, a 29.5-fold spread across the 31 organizations reporting at least 50 IT FTEs, and the most quotable number in the whole analysis (Treasury Board of Canada Secretariat, 2026).

ISED's figure was the broadband fund. CA$949.39 million of ISED's CA$1,037.53 million — 91.5% — is that one transfer programme, carried by 105 people, because handing out money needs fewer staff than running a data centre (Treasury Board of Canada Secretariat, 2026). Following that number is what sent me to the core-responsibility column, and it is the reason the CA$2.23 billion above is on the record.

It is also where the ratio stops being a measurement.

The second instrument, and the same twenty-minute test

Put the ratio through what the keyword match had just failed, and it does not survive either — for a plainer reason. The two columns are not commensurable. planned_ftes_1 is gross headcount, while the pattern in the data is consistent with planned_spending_1 being reported net of recoveries (author's inference; the published definitions do not state the treatment). Divide the first by the second and you do not get a cost per employee.

The file announces this in its own programme names:

Programme	Planned spending and staff	Implied cost per person
Real Property Services Public Services and Procurement Canada	CA$0.70M 3,689 FTEs	CA$190
Intellectual Property ISED	CA$11.63M 1,295 FTEs	CA$8,981
Cost-Recovered Statistical Services Statistics Canada	CA$13.77M 1,169 FTEs	CA$11,779
Information Technology Services Indigenous Services Canada	CA$5.29M 379 FTEs	CA$13,961

(Treasury Board of Canada Secretariat, 2026)

Nobody employs 3,689 people for CA$700,000. Those organizations recover their costs from the departments they serve or charge them to a separate account, so the spending column reports what is left after the recovery while the headcount column reports everyone. This is not a handful of exceptions: 199 of the 761 programme lines carrying 50 or more planned FTEs — 26% — report less than CA$130,000 per person, which is roughly one fully-loaded federal position on my working assumption (Treasury Board of Canada Secretariat, 2026).

Seventeen of those lines sit inside the corrected IT table, Indigenous Services' own IT line among them at CA$13,961 a head. So the residual 5.7-fold spread that survives the transfer-payment correction is largely telling you which organizations recover their costs. It is not telling you which ones build and which ones buy.

The spend figures themselves are unaffected: they are dollars against programmes, and the ranking below is built from them. What does not survive is the per-employee denominator.

Where the league table moves

Ranked by keyword match

Ranked after the second field is read

• #1 National Defence — CA$1,812.44M→ #2, CA$1,082.79M (defence R&D removed)
• #2 Shared Services Canada — CA$1,281.46M→ #1, CA$1,281.46M (unchanged)
• #3 ISED — CA$1,037.53M→ #21, CA$37.54M (broadband fund removed)
• #5 Natural Resources Canada — CA$383.50M→ #20, CA$38.23M (clean-tech programmes removed)
• #9 National Research Council — CA$193.02M→ #15, CA$57.33M (research centres removed)

What is still a judgment call

The honest answer to "how much does Canada plan to spend on federal IT" is a band, not a number, and the reason is worth stating rather than hiding behind a single figure.

Some lines are unambiguous. Information Technology Services under Internal Services is IT. Biotechnology and Genomics under Fisheries is not. Between them sits a set of programmes that deliver information and communications technology as part of a mission rather than as back office: Defence's ICT systems acquisition line, its Ready Cyber and Joint Communication Information Systems forces, the RCMP's Operational IM/IT Services and its National Cybercrime Coordination Unit, the Canada Border Services Agency's Field Technology Support, Elections Canada's Electoral Data Services, and four smaller cyber and research-IT lines. Together they carry CA$955.30 million.

Count them and the total is CA$5.08 billion across 36 organizations. Exclude them and it is CA$4.12 billion across 35. Keep the raw keyword match and it is CA$7.31 billion across 37 (Treasury Board of Canada Secretariat, 2026). Three defensible implementations, three answers, and the honest sentence is that one rather than any single figure. If you want a single number for a board slide, use the middle one and say which programmes it includes.

Note what does not vary. Whichever reading you take, ISED and Natural Resources Canada are not top-five federal IT buyers, and the CA$2.23 billion the keyword match added is not in dispute — the department's own classification puts every one of those eleven lines somewhere else.

Where this fits in how I work

I found this by auditing my own instrument rather than a competitor's, which is the only reason the corrected number exists. The keyword match is the one running inside the Sagentix federal organization census; six of the 261 organizations were carrying an inflated IT figure because of it, and since the 24 September 2026 census refresh every organization carries the adjudicated one. An instrument nobody re-tests is an instrument nobody can trust, including its author.

Every Sagentix Phase 01 market-intelligence engagement now checks a derived classification against a second field in the source (the column the first pass ignored) before any count from it reaches a client deliverable, because a count is only as good as the instrument behind it and the instrument is the part nobody checks Sagentix Phase 01 Market Intelligence, 2026. The full delivery system runs 6–8 weeks, draws on 1,425 curated artifacts, and is priced from CA$4,500 for Phase 1 to CA$45,000 for a full go-to-market build, depending on scope — with a Phase 1 money-back guarantee (subject to terms).

The discipline itself is smaller than the engagement and you do not need me for it. Two free files, one afternoon, and a second column you were already downloading.

Three ways to act on this

Re-run your own federal numbers against the core-responsibility column. Download the Departmental Plans expenditure file, keep your existing filter, then group the results by core_responsibility and read the groups. Anything under Science and Innovation, People, Skills and Communities or a natural-resources heading is a mission programme wearing a technology word. This takes an afternoon, costs nothing, and you keep the corrected list. If someone in-house will actually do it, this is the right option.

Test both ends of every derived number. For anything you obtained by matching, write down eight strings your filter must match and, more importantly, nine near-misses it must reject — the same words in a different role, the other language's false friend, the compound noun that contains your keyword. For anything you obtained by dividing, check that the numerator and the denominator cover the same population: a spending column net of recoveries over a gross headcount is not a cost per person, however sensible the ratio looks. Twenty minutes, and it generalises far beyond federal data.

Bring in a structured market-intelligence pass when a number is load-bearing — when a board has been shown a federal TAM, when a raise depends on it, or when the territory plan for two quarters is about to be written from a ranking. That is a Phase 01 engagement, and I would pick it last of the three if the first two are open to you, because they are cheaper and the capability stays with you.

The federal IT market is not hidden. It is mislabelled, in a file that also carries the label that fixes it.

Which of your market figures came out of a keyword match or a division — and when did you last check what that filter refuses, or what those two columns each actually count?

References

• Grimmer, J., & Stewart, B. M. (2013). Text as data: The promise and pitfalls of automatic content analysis methods for political texts. Political Analysis, 21(3), 267–297. https://doi.org/10.1093/pan/mps028 (quotations verified against the author-hosted full text; section numbers cited rather than pages, because that printing paginates 1–31).
• Loughran, T., & McDonald, B. (2011). When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. The Journal of Finance, 66(1), 35–65. https://doi.org/10.1111/j.1540-6261.2010.01625.x (the publisher's copy refuses automated requests; quotations verified against the linked full text of the published version).
• Transport Canada. (2016). Appendix: Definitions. Government of Canada. (Page last modified 21 January 2016. Reproduces the pre-2016 Treasury Board definitions of programme, full-time equivalent and planned spending used in Reports on Plans and Priorities; the Policy on Results carries its own, different definition of Program.)
• Treasury Board of Canada Secretariat. (2016). Policy on Results. Government of Canada.
• Treasury Board of Canada Secretariat. (2016b). Directive on Results. Government of Canada.
• Treasury Board of Canada Secretariat. (2026). Departmental Plans and Departmental Results Reports — Expenditures and full-time equivalents by program and organization [Data set]. GC InfoBase, Government of Canada. Open Government Licence – Canada. Re-derived 8 September 2026 from the CKAN datastore distribution, resource 64774bc1-c90a-4ae2-a3ac-d9b50673a895.
• Treasury Board of Canada Secretariat. (2026b). GC InfoBase [Program spending and full-time equivalents data files, 2026–2027 Departmental Plans, updated 23 September 2026]. Government of Canada.

Contains information licensed under the Open Government Licence – Canada. That licence covers the GC InfoBase expenditure dataset used throughout. The Treasury Board Policy on Results and Directive on Results, and the Departmental Plan definitions reproduced by Transport Canada, are Canada.ca content, which is not open-licensed: that material is quoted for commentary and is not covered by the Open Government Licence.