The business services sector doesn't look like one economy anymore. Depending on which
sub-sector you're watching, you're either witnessing a cyclical recovery powered by
AI spending — or a structural collapse that no cycle will reverse.
We tracked 12 filing-extracted ML classifiers ("primitives") across 300+ staffing, IT
services, and consulting companies from 2005 through early 2026. The divergence between
sub-sectors has never been wider. And the direction of travel for each is now unmistakable.
How This Works
Every chart in this report is built entirely from trained machine learning classifiers —
not from earnings calls, analyst estimates, or alternative data vendors. Each
"primitive" is a binary classifier that reads a single sentence from an SEC filing and
answers one yes/no question. We run 12 of them across every filing from every business
services company in our universe, then aggregate the signal across hundreds of companies
each quarter.
SEC Filing
→
143M sentence corpus
→
12 ML classifiers
→
Per-filing score (0–1)
→
Quarterly aggregate
→
Net signal chart
The 12 Primitives
Each primitive was trained on thousands of human-labeled sentences from SEC filings using
Claude as the labeler. All 12 achieve AUC above 0.93 — meaning the classifier almost
never confuses a genuine "demand inflecting" sentence with boilerplate or unrelated text.
Six are bullish signals, six are bearish. Every chart in this report is the net spread
between the paired classifiers.
is_demand_inflecting
Order/booking trends turning positive or accelerating
"We saw bookings inflect positive in the second quarter, with IT services demand accelerating across all verticals."
AUC 0.937
is_demand_softening
Order/booking trends weakening or decelerating
"Order volumes decelerated across service lines, particularly in discretionary consulting projects."
AUC 0.968
is_pricing_power_services
Bill rates or pricing power strengthening
"Bill rates increased 8% year-over-year, reflecting strong demand for specialized AI implementation skills."
AUC 0.980
is_pricing_pressure_services
Competitive undercutting or client pushback on rates
"Pricing pressure intensified as clients pushed back on rate increases and competitive pressures mounted."
AUC 0.974
is_utilization_improving
Consultant/resource billable utilization rates rising
"Utilization rates improved to 78% from 72%, driven by strong demand for cloud migration projects."
AUC 0.933
is_utilization_declining_services
Billable utilization rates falling, benched headcount
"Utilization declined to 68% as project completions outpaced new starts in the quarter."
AUC 0.945
is_backlog_building
Contract backlog or future revenue visibility growing
"Contract backlog grew 15% year-over-year to $2.4B, providing strong revenue visibility into 2026."
AUC 0.962
is_backlog_shrinking
Contract backlog declining, forward visibility weakening
"Backlog declined to $1.8B from $2.2B last year, reflecting reduced discretionary project activity."
AUC 0.931
is_dso_improving
Days Sales Outstanding shrinking — clients paying faster
"Days sales outstanding improved to 58 days from 67 days, reflecting stronger client payment discipline."
AUC 0.959
is_dso_deteriorating
DSO expanding — B2B payment stress signal
"Days sales outstanding increased to 72 days from 61 days, reflecting slower collections from enterprise clients."
AUC 0.974
is_same_client_revenue_growing
Organic growth from existing accounts — retention signal
"Revenue from existing clients expanded 12% as we deepened wallet share with our top 20 accounts."
AUC 0.951
is_client_concentration_risk
Revenue dangerously concentrated in few accounts
"Our top three clients accounted for 47% of revenue, representing significant concentration risk."
AUC 0.931
Net signal = bullish primitive score − bearish primitive score.
Each filing gets a probability (0–1) for each primitive. We take the maximum score
across all sentences in the filing (so one strong sentence = filing fires). Then
we average across all filings in a quarter for each sub-sector. Net demand of +2.7
means is_demand_inflecting scores averaged 2.7 percentage points higher
than is_demand_softening across all IT services filings that quarter.
IT Services
Accelerating
Demand: +2.7 (best since 2015)
Backlog: 10.1 (near record)
Utilization: 2.5 (highest ever)
Staffing
Structural Decline
Demand: negative 5 straight quarters
Utilization: never recovered
Pricing: near zero (0.3)
Consulting
Entering the Drop
Demand: just turned negative
Pricing: went negative (first time)
Backlogs still holding... for now
The Full Picture: 2005–2026
Before examining the divergence, it helps to see the full historical arc. The chart below
shows quarterly net demand — the difference between "demand is expanding" and "demand is
softening" signals — across all business services companies combined.
Business Services Net Demand Index
is_demand_inflecting minus
is_demand_softening
All companies (SIC 7361, 7363, 7371, 7373, 7374, 8742) · Quarterly 2005–2026 · net = bullish primitive score minus bearish primitive score
The sector's history is visible in the data: the 2008-2009 crash sent demand deeply
negative (-2.3 by Q1 2009), the post-GFC recovery was slow but steady through 2019,
COVID created a brief but sharp shock in 2020, and the 2021 boom was real. What happened
next is the story worth telling.
The Divergence Begins: 2022–2026
Starting in mid-2022, the sector began splitting apart. The aggregate numbers masked a
bifurcation that has since become a trifurcation. Each sub-sector is now on a completely
different trajectory.
Sub-Sector Net Demand: The Great Divergence
is_demand_inflecting minus
is_demand_softening
· by sub-sector
IT Services vs Staffing vs Consulting · Q1 2020 – Q1 2026
IT Services (green): Bottomed in mid-2024 at +0.8 and has since accelerated
to +2.7 in Q1 2026 — the strongest reading since 2015. The backlog signal hit 10.1 in
Q1 2026, near a 20-year high. The recovery is real and broadening.
Staffing (red): Net demand has been negative for 5 consecutive quarters.
Utilization has been structurally negative since at least 2019. This is not a cyclical
trough — staffing companies haven't been able to generate positive utilization trends
even during the 2021 peak. This is a structural story.
Consulting (yellow): Peaked in Q1 2023 and has been declining steadily.
Q4 2025 net demand hit -1.6. More ominously: pricing power just turned negative for the
first time in our dataset (Q1 2026: -0.2). When consulting firms can no longer raise rates,
the next step is margin compression and headcount reduction.
The Pricing Signal: Where It Gets Interesting
Demand tells you about volume. Pricing tells you about power. The two usually move together
— except right now.
Net Pricing Power by Sub-Sector
is_pricing_power_services minus
is_pricing_pressure_services
Pricing power signal minus pricing pressure signal · Q1 2020 – Q1 2026
IT Services has maintained positive pricing even through the 2023-2024 demand slump.
That's the AI premium: companies with AI-enhanced services are charging more per engagement
even when deal volumes are down. Pricing hit +3.2 in Q4 2025 and remains strong.
Staffing pricing has nearly collapsed to zero. When you're competing with AI automation
for temp placements, there's no pricing power to find. Consulting pricing just crossed
below zero for the first time — a leading indicator that margin pressure is coming next.
The Utilization Story: Who's Actually Busy
Net Utilization by Sub-Sector
is_utilization_improving minus
is_utilization_declining_services
Utilization improving minus declining signal · Q1 2020 – Q1 2026
+2.5
IT Services utilization Q1 2026 — highest reading in 20 years of data
-2.6
Staffing utilization Q1 2026 — negative in every quarter since at least 2018
+1.1
Consulting utilization — the one bright spot for consultants, but demand is the warning
Backlog: The Forward Indicator
Backlog signals — whether companies are reporting growing vs shrinking order books — are
the most forward-looking metric in our dataset, typically leading reported revenue by 1-2
quarters.
Net Backlog Signal by Sub-Sector
is_backlog_building minus
is_backlog_shrinking
Backlog growing minus shrinking signal · Q1 2020 – Q1 2026
IT Services backlog hit 10.1 in Q1 2026 — a level last seen in 2012 during the post-GFC
cloud migration wave. This strongly suggests IT services revenue will continue accelerating
through H1 2026.
Staffing backlogs have collapsed to 2-6 (vs. 12-15 in good times). Consulting backlogs
are holding at 7-8 for now — which explains why consulting hasn't fully broken yet — but
the demand and pricing signals suggest that's temporary.
Why the Divergence Is Happening
2022–2023
The demand pause: Tech company cost-cutting drove broad-based
weakness across all business services. IT project deferrals, hiring freezes,
budget cuts. All three sub-sectors weakened together.
Late 2023
Staffing goes negative first: As enterprise clients began
implementing AI tools for automation tasks, demand for temporary workers —
particularly in back-office, data entry, and administrative functions — didn't
recover with the broader economy. Staffing demand turned structurally negative.
Mid-2024
IT Services inflects: AI infrastructure buildout, cloud
migration acceleration, and digital transformation projects begin driving IT
services demand higher. Companies need people to build, deploy, and manage AI
systems. Pricing power emerges for firms with AI capabilities.
H2 2025
Consulting starts to crack: As AI tools become capable of
strategy analysis, market research, and benchmarking — tasks that generated
premium consulting fees — clients begin questioning the ROI of expensive
engagements. Demand turns negative in Q4 2025, pricing follows in Q1 2026.
Q1 2026
IT Services accelerates: Backlog at 10.1, demand at +2.7,
utilization at a 20-year high of 2.5. The AI spending cycle is in full swing
for the companies that build and implement technology.
The Recession That Already Happened
For context: what does a real recession look like in this data? The 2008-2009 financial
crisis provides the clearest reference point.
2008–2009 vs Today: A Comparison
is_demand_inflecting minus
is_demand_softening
· all business services · aligned to cycle start
Both cycles indexed to Q1 of the respective slowdown · net demand quarterly signal
During 2008-2009, net demand for the sector as a whole dropped to -2.3, pricing turned
deeply negative (-2.1), and backlogs fell to 5-6. Recovery took nearly 5 years to return
to 2007 levels.
Staffing's current readings are comparable to 2009 recession lows. For staffing
companies, the recession isn't coming — it's been underway since 2023, and unlike 2009,
there's no obvious catalyst for recovery.
What to Watch Next
Three signals will determine whether the current trajectory continues or reverses:
1. Consulting backlog durability: Backlogs are holding at 7-8 despite
demand and pricing weakness. If backlogs break below 5 in the next 1-2 quarters, that
would signal that the existing project pipeline is being worked down without replacement —
the same pattern that preceded staffing's structural decline.
2. IT Services pricing sustainability: Pricing has held at +3.2 even as
demand fluctuated. If pricing can be maintained while demand grows, IT services margins
could significantly expand. Watch for any sign of competitive pricing pressure as more
firms try to capture the AI services market.
3. Staffing pricing collapse: Currently near zero (0.3). If staffing
pricing goes negative, it would signal rate undercutting — firms competing on price to
fill positions, which typically precedes M&A-driven consolidation in the sector.
The Statistical View: Z-Score Analysis
The charts above show raw NET signals. But when did the divergence become statistically significant?
We normalize each sub-sector against its own 8-quarter rolling history. When z-score exceeds ±2σ,
something genuinely unusual is happening.
Sub-Sector Z-Scores: When Did Each Break?
Z-score = (current quarter − rolling 8Q average) ÷ std dev. Gray band = normal range (±2σ).
Outside the band = statistically rare event (<5% of quarters).
-4.0σ
Consulting Q2 2020
COVID shock (historic)
+2.6σ
IT Services Q1 2026
4th consecutive breakout
-2.4σ
Consulting Q4 2025
Breakdown while IT breaks out
IT Services has been in sustained breakout territory since Q4 2023.
Four of the last five quarters exceeded +2σ. This isn't noise — it's a statistically
significant multi-quarter trend driven by AI infrastructure spending.
Consulting's Q4 2025 breakdown is the warning sign.
It hit +2.0σ in Q3 2025 (briefly positive), then immediately crashed to -2.4σ in Q4 2025.
This kind of whipsaw — breakout followed by breakdown — often signals a turning point.
Staffing's brief breakout in Q3 2021 didn't last.
The +2.8σ reading was the post-COVID hiring boom. It never returned to breakout territory.
The structural decline was already underway — the boom just masked it temporarily.
Methodology
All data is extracted from SEC filings (10-K, 10-Q, 8-K, and other periodic reports)
using machine learning classifiers trained on sentences from the ARGOS corpus — a database
of 143+ million sentences extracted from U.S. public company filings since 1993.
The 12 binary classifiers shown above were trained using the following pipeline:
(1) a PostgreSQL full-text search query targets sentences likely to contain the signal;
(2) 2,500–5,000 sentences are sampled (targeted + random mix) and labeled by
Claude Haiku; (3) sentence embeddings (all-MiniLM-L6-v2, 384 dimensions) are computed;
(4) a logistic regression classifier is trained on the embeddings. All 12 classifiers
achieve AUC 0.93–0.98 on held-out test sets.
Each filing is scored by running all sentences through the classifiers and taking the
maximum probability across sentences (so one strong sentence is sufficient to fire the
primitive). Quarterly aggregates are the mean filing-level score across all filings in
the quarter for each sub-sector. Net signal = bullish mean − bearish mean.
Results are stored in the business_services_filing_primitives table
(3.68M rows, ~300 companies, 1970–2026). Companies covered: SIC 7361, 7363 (staffing),
7371, 7373, 7374 (IT services), 8742 (management consulting).
Data note: Q1 2026 figures are based on filings received through
February 21, 2026. Annual report season is ongoing; Q4 earnings filings will continue
through mid-March. Current quarter readings may shift modestly as additional filings
are processed. Historical data (pre-2024) is complete and stable.