Why NAICS Codes Fail to Describe What a Plant Actually Makes

NAICS codes tell you what industry a plant belongs to. They do not tell you what that plant actually makes, what equipment sits on its floor, or what it needs to buy to keep running. Sales organizations that build territory plans and prospecting lists on NAICS are borrowing a system built for a different job entirely, and that borrowing is the reason so many industrial sales cycles start pointed at the wrong plant. A better NAICS lookup will not solve this. The path forward is dropping NAICS as a targeting tool altogether and replacing it with something that describes the plant floor instead of the tax filing.

The North American Industry Classification System comes from federal statistical agencies, built to collect and publish economic data across the U.S., Canada, and Mexico. Its organizing principle groups establishments by the similarity of the processes they use to make goods or provide services. Each establishment gets a six-digit code, and Census extends that further into ten-digit product codes for manufacturing and mining data specifically. The whole taxonomy gets reviewed on a five-year cycle, so it lags behind real industrial change by design. That lag reflects the system working as intended, for a purpose that has nothing to do with sales targeting.

NAICS sorts industries into buckets that economists and policymakers can compare over time. Its design mandate stops well short of describing what a single plant on a single street corner in Ohio actually produces on a Tuesday. Sales targeting, territory planning, account intelligence: none of that appears anywhere in that mandate. Applying census-grade statistical categories to commercial prospecting is where things break, and the break is structural, not incidental.

How a single code gets assigned — and what gets left out

Every establishment gets exactly one NAICS code, tied to its primary activity, usually whatever generates the most revenue. That single-code rule is where most of the distortion starts, and it throws away more than most buyers of this data seem to realize.

Take a plant running stamping alongside precision machining. It files under one code, and the secondary operation — potentially a large share of what the plant does — disappears from the record entirely. Census has walked through this exact kind of case: an establishment producing a mix of computers, storage devices, and semiconductors gets classified as a semiconductor manufacturer. The majority of its output vanishes, at least as far as the code is concerned.

The problem compounds at coarser levels of the hierarchy. Five-digit codes collapse more variation than six-digit codes, and a company-level database often collapses further still, attaching one NAICS code to an entire enterprise. A parent company might legitimately run a manufacturing plant, a retail storefront, and a headquarters office, each with a different primary activity and a different valid code. A rep who pulls up that company's record typically sees one code, usually whatever sits at headquarters, and every plant that doesn't match HQ's primary activity gets misrepresented before the call ever happens.

Self-reporting and the accuracy problem baked into the data

NAICS codes are self-reported, and the nature of self-reporting means codes routinely get misidentified, left blank, or keyed incorrectly. That is a documented feature of how the data gets collected, not a rare edge case. Treating it as anything less than a routine condition of the file is the mistake most buyers of this data make: if a database vendor doesn't flag self-reporting error as a known limitation, that omission is the tell.

The risk compounds once you factor in how commercial databases actually get built: multiple lookup tools, manual entry, records passed between systems that don't agree with each other. The taxonomy itself invites confusion. A bakery trying to classify itself has to choose among multiple plausible codes covering retail baked goods stores, retail bakeries, commercial bakeries, cookie and cracker manufacturing, and other snack food manufacturing. All five are plausible depending on how the operation runs, and a self-reporting owner has no strong reason to land on one over another. That kind of fork runs across hundreds of manufacturing subsectors, not just food.

The errors stack in a predictable sequence: first the structural truncation built into the one-code-per-establishment rule, then whatever self-reporting mistakes sit underneath it. A seller pulling a prospect list from a NAICS-based database inherits both layers at once, and neither shows up as a flag anywhere in the file. The list looks clean, and that appearance is exactly the danger, because a clean-looking list is the one nobody thinks to question.

Two plants, one code, completely different operations

Machine Shops is the cleanest illustration of how far this drifts. That single code can cover high-volume turning, small-batch grinding, and broaching: operations with almost nothing in common when it comes to metalworking fluid, coolant, or lubricant needs. A supplier treating "Machine Shops" as one homogeneous market is really selling into three or four distinct chemistries under a single label, and most never find that out until the pipeline stalls.

Fluid milk is stranger still. A single fluid milk code can cover conventional dairy production and milk made through fundamentally different production processes — and the input chemistry and production infrastructure behind each have almost nothing in common.

NAICS groups by how something is made, which was supposed to produce statistical comparability. In practice it produces categories broad enough to hold operations that share a label and nothing else. For an industrial seller, the code names the neighborhood; it says nothing about what equipment runs on the floor, what material gets worked, or what gets consumed in the process. A specialty chemical or metalworking fluid rep working off "Machine Shops" as a prospect pool is selling blind, in the most literal sense of the phrase.

Why process type — not product category — is the real signal for purchasing needs

Manufacturing breaks into a handful of major process families: casting, molding, forming, machining, joining, surface treatment, additive manufacturing. Each one creates its own distinct set of consumable, chemical, and tooling requirements, and that layer is exactly the one NAICS never touches. Process family, more than product category, determines what a plant needs to buy. Confusing the two is where most territory plans go stale, and it is the single most avoidable mistake in industrial targeting.

Machining is subtractive and CNC-driven, and it demands tight dimensional tolerances, which means precision coolants and cutting fluids matched to the specific material and operation. Grinding, turning, and gun-drilling each carry their own thermal and lubrication profile; a fluid that works for one can be wrong for another. Stamping and forming live in a different world, high-volume sheet metal work where the lubricant's job is to build a controlled film between the workpiece and the tooling, a different product family entirely from anything used in machining. Forging is different again: high-strength components like gears, crankshafts, and aerospace parts, made under extreme pressure, with fluid requirements that barely overlap either of the above.

Tailor a product to the specific process, the equipment, the operators, the facility, and a supplier can actually help a plant run better. Sourcing an account off a NAICS list that never revealed the process to begin with makes that kind of tailoring impossible. Before the first call, a seller needs to know what process family the plant runs, what materials it works, and what equipment sits on its floor. NAICS reliably tells you none of it, and pretending otherwise is where sales cycles start bleeding time.

The commercial stakes in markets where process fit is everything

The metalworking fluids market shows what is actually at stake. Global Market Insights valued it at $13.6 billion in 2025, projecting growth to $25.7 billion by 2035, a 6.7% compound annual growth rate. Straits Research reads it more conservatively: $11.45 billion in 2024, growing to $16.3 billion by 2033 at a 4.1% CAGR. The two forecasts disagree on pace but agree on the range: current global value sits somewhere between $11 billion and $14 billion, and aerospace is also the fastest-growing segment of it.

In a market that size, with process specificity baked into every purchasing decision, a misclassified prospect is not a rounding error. It is a costly misdirection of sales resources inside a cycle that already runs long, often months, sometimes well over a year, and touches procurement, operations, engineering, and safety along the way. Every stage that follows a bad target compounds the cost of that first mistake. Metalworking fluids is one vertical among several; the same process-fit logic governs specialty chemicals, coatings, water treatment, packaging, and plastics. Sellers who get the targeting right in any of these markets are working with an advantage that competitors sourcing off NAICS simply don't have, and it shows up first in win rate, not in list size.

What plant-level production data captures that NAICS cannot

Plant-level data starts from the facility, not the filing. It profiles what the plant makes, how it makes it, what equipment it runs, what materials it processes, how much it produces, and what its environmental footprint looks like. That starting point sits far apart from a six-digit code assigned once and revisited every five years.

It also carries a time dimension NAICS simply lacks. Facility expansions, new production line announcements, EPA and building permits, leadership changes: a new VP of Operations often signals a strategic review underway, and with it, an opening for a new supplier to get in front of the decision before it closes. Equipment profiles come closest to a real purchasing signal. A plant running a fleet of CNC machining centers has a predictable coolant draw; a plant that just added a grinding line has added a fluid requirement that shows up in the equipment record and stays invisible in the classification code.

Most mature manufacturing facilities run more than one process, and they only get described accurately when each process gets captured on its own, not when the whole facility gets flattened into its single dominant revenue activity. That is also where plant-level data breaks from ordinary firmographic databases. Headcount and revenue bands describe how big a company is; they say nothing about what a given facility does on its floor, and therefore nothing about what it needs to buy. For an industrial sales team, that is the line between a list and an actual territory: knowing an industry exists somewhere in a region versus knowing exactly which plants in that region run the process a given product serves.

What changes when territory planning and account intelligence are built on process data instead of codes

Territory planning built around manufacturing density and process type beats territory planning built around geography or NAICS-code counts, and it isn't close. A rep assigned to "machine shops in the Midwest" does not need a count of how many establishments share a six-digit code in that region; she needs to know which of those shops run the specific processes her product actually serves. Anything short of that is guesswork dressed up as a territory map.

The same shift changes account growth work inside existing customers. A customer filed under one NAICS category may run several different process families across its various plants, and each one is a cross-sell or upsell opportunity that stays invisible for as long as the account gets viewed through a single industry label.

CRM enrichment changes too. A record with a NAICS code and a headcount number attached is a starting point and nothing more. Add process type, equipment profile, and facility-level activity signals, and that same record turns into something a rep can act on. The first conversation itself changes shape as well: walking into a call already knowing what a plant makes and how it makes it turns a cold, diagnostic conversation into something closer to a consultation. Platforms that index facilities at the plant level, tracking process type, equipment, and production signals across hundreds of thousands of manufacturing sites, are what make that shift practical for a sales team covering real territory rather than a handful of accounts.

NAICS still has its place, and it will keep doing exactly the job it was built to do: sorting industries into comparable statistical buckets for economists and policymakers. Asking it to also answer what a particular plant, on a particular street, actually makes is asking it to fail. A cleaner spreadsheet will not fix that; the right response is to stop using the code for that purpose entirely.