PMI explained

How to read the Report on Business.

What the ISM publishes

ISM's panel members — purchasing and supply executives at hundreds of stratified-by-industry firms — answer one question per business measure each month: was activity better, same, or worse than last month? The diffusion-index calculation is 1 × (% Better) + ½ × (% Same) + 0 × (% Worse), producing a number between 0 and 100. Fifty means as many respondents reported improvement as decline. The math intentionally throws away magnitude — a small uptick at every firm scores the same as a large uptick at half the firms — which trades information density for noise resistance.

Manufacturing PMI

Running monthly since 1948, the longest unbroken monthly diffusion-index survey of US business activity. Headline PMI is an equally weighted average of five subindices: New Orders, Production, Employment, Supplier Deliveries, Inventories. The full ROB tracks five more — Prices, Backlog of Orders, New Export Orders, Imports, and Customers' Inventories — that don't feed the headline but matter for breadth. Released on the first business day of each month at 10:00 AM ET, covering the prior month. The equal- weight scheme dates only to January 2008; pre-2008 weights were 30/25/20/15/10. Anyone backtesting on the spliced 1948-present series is mixing two methodologies unless they re-derive the headline from the underlying subindices.

Services PMI

Monthly since June 1997. The headline was originally the Business Activity Index; ISM introduced the composite NMI in January 2008 and renamed it Services PMI in August 2020, alongside the launch of the Hospital PMI. The composite is an equally weighted average of four subindices: Business Activity, New Orders, Employment, Supplier Deliveries. Released on the third business day of each month at 10:00 AM ET. The live Services callout above pulls from data/nmi-wayback.json; three of the four interactive modes (Timeline, Decompose, Heatmap) already render Services alongside Manufacturing. Fed Chair remains Manufacturing-only, since its scenarios all pre-date the Services PMI's 1997 launch.

The annual SA refresh

Each January, ISM re-estimates seasonal-adjustment factors for the entire history. A number that looked like 49.5 in November can be revised to 50.1 the following February. The autumn-2008 cluster was the textbook example: the 2012 SA announcement noted those prints "may not have been adequately handled with default settings", and ISM extended its revision window from the customary four years to seven to incorporate the corrections. Anyone modeling against pre-refresh vintages of GFC-era months will see a different recession bottom than someone using the current vintage.

How to read a release

Three numbers, in order

1. The headline level — above or below 50, and by how much.
2. The change vs. last month — the rate-of-change of a rate-of-change index, so this is the acceleration. A move from 51 to 53 is a different signal than 49 to 51.
3. The breadth — how many of the ten subindices are above 50. A headline above 50 with breadth below 5/10 is a fragile expansion; the reverse is a resilient one.

The 50-line is not the only line

The 50-line tells you whether the surveyed sector expanded. The lesser-known thresholds — derived by ISM via regression against BEA GDP — tell you whether the overall economy is expanding: Manufacturing PMI above ~42.5 → overall economy expanding; Services PMI above ~48.6 → overall economy expanding. ISM re-fits the regression periodically and the Manufacturing anchor has been published as 42.3, 42.5, and 42.7 across the past decade (42.5 in the current ROB) — always check the live release for the active value rather than memorizing.

Surprise vs. consensus

Markets don't trade levels, they trade surprises. A Manufacturing PMI of 51.5 is bullish if consensus was 50.0 and bearish if consensus was 53.0. Every sell-side desk and every Bloomberg/Citi surprise index frames the print this way. If you only check the level, you will systematically read every release backwards relative to how the bond market reads it.

The respondent comments

Every report ends with quoted respondent comments. These are the only qualitative data in the release and are routinely under-weighted by quants. Comments lead the quantitative subindices when something structural is shifting — supply-chain dislocations in spring 2020, semiconductor shortages in 2021, tariff repricing in 2025 all showed up in comments before they showed up in the diffusion math.

The most common misreads

• Crossing vs. magnitude. A print of 50.2 vs. 49.8 is statistically indistinguishable. The press will report a regime flip; the data doesn't support one.
• Levels vs. changes. "PMI is still above 50" says nothing about whether the trend is up or down. Pair the level with the delta.
• Single-subindex stories. A Prices spike with no New Orders confirmation is a one-month commodity shock, not an inflation regime shift. The signal is in the cluster.
• Conflating ISM and S&P Global. They are independent surveys of different panels using different question wording and frequently disagree, especially at turning points.

Reading guides by audience

The framework above is universal. What you do with it depends on whose decisions you're informing. Pick the lens that matches your work.

Worked example — the latest prints

The framework is most useful applied to both reports together. A divergence between Manufacturing and Services is itself information — Mfg has historically led at cyclical turning points, while Services dominates the GDP-tracking composite. Working through both anchors today:

Manufacturing

The most recent observation is September 2026 at PMI 54.5, down 0.1 points from August 2026's 54.6. Apply the framework: the headline is above the 50-line (manufacturing-sector expansion), and above the 42.5 anchor (consistent with overall-economy expansion). The rate-of-change reading — the acceleration — is deteriorating. For a full read you'd add the breadth count from the five subindices on /decompose, the per-industry growth/contraction split on /heatmap, and the policy regime context on the timeline.

Services

The most recent observation is September 2026 at NMI 54.9, down 0.5 points from August 2026's 55.4. Apply the framework: the headline is above the 50-line (services-sector expansion), and above the 48.6 anchor (consistent with overall-economy expansion). The rate-of-change reading is deteriorating.

Read together

Services is roughly 70-75% of US GDP, so a value-added-weighted composite would be roughly 54.8 — Services-weighted ~75% / Manufacturing-weighted ~25%. That blended read stays above the 50-line. The two reports are within 3 points of each other this month — broadly aligned, no sector-divergence story to tell.

ISM does not publish an official Manufacturing + Services composite; the 75/25 blend above is a back-of-envelope GDP-weighting, not an ISM number. The widely cited US Composite PMI is a separate product from S&P Global Market Intelligence, built from S&P's own PMI panels rather than ISM's.

Note: the Wayback series have structural gaps — Manufacturing 2015-04 → 2020-05, Services 2015-04 → 2020-06 (the two series resume in different months because ISM's ROB URL structure stabilised at slightly different points). Recent headline prints come instead from ISM's monthly press releases on PRNewswire and are usually current within days of release; subindex breakdowns on /decompose and /heatmap still rely on Wayback and can run a few months behind. For the live ROB go to ismworld.org directly.