The company size filter you trust to protect lead quality is built on a number somebody typed into a company page years ago and never updated.
Firmographic targeting works as a floor, not as a qualifier. Employee count, industry, seniority and revenue filters are useful for keeping your ads away from obviously wrong companies. They are not accurate enough to decide which of two plausible buyers is real. In B2B paid media, the more of these filters you stack, the more error you compound, and the less you can tell what caused a bad month.
This is written for demand generation and growth leads at mid-market and enterprise B2B companies who have already tightened targeting twice and are still getting leads sales will not call.
Where firmographic data actually comes from
Firmographics describe the company, not the person. Employee count, industry, revenue band, location, funding stage, tech stack.
On LinkedIn, company attributes like company size and industry come from the organization's own Page rather than from any verified registry. LinkedIn lists them as company attribute targeting in its targeting documentation. Nobody audits them. A company that listed itself at 11 to 50 employees when it built the page stays at 11 to 50 until a marketing hire notices and edits it.
Enrichment tools fill the rest, and they bring a different failure. A lot of them count the profiles associated with a company rather than the people the company employs. Those two numbers separate fast. Contractors, alumni who never updated their profile, founders of a side project, and a few duplicate accounts all sit inside the profile count. We have opened target lists where a business shows up at three or four times its real headcount for exactly that reason, and plenty where a 900-person company reads as a startup because its page was never touched.
Neither error is random, which is the part that hurts. Both skew toward companies that manage their online presence badly, and companies that manage their presence badly are not spread evenly across your market.
Ready for paid ads that pay off?
Book your free auditThe audience size estimate is a range, not a number
Platform audience estimates inherit every problem above and then round.
Treat the number Campaign Manager or Google shows you as an order of magnitude. If it says 14,000, you know you are in five figures. You do not know whether 3,000 of those are real buyers or 11,000 are.
The damage shows up when that estimate is what you make a decision on. Sizing a budget off a precise-looking audience count, or writing off a segment as too small to run, is a decision built on a number nobody at the platform is standing behind. We wrote separately about what to do when a LinkedIn audience really is tiny, and the first move there is still to distrust the count.
Why stacking targeting layers makes lead quality worse
The sequence looks the same almost every time. Leads come in unqualified, so the team adds a job function filter. Quality does not move, so they add seniority. Then industry. Then a company list. Four layers in, volume has collapsed, cost per lead has doubled, and quality is where it started.
Every layer multiplies the error rate instead of averaging it. If each attribute is right 80% of the time, two stacked layers land around 64% and four land around 41%. Those figures are illustrative rather than measured, but the direction is the whole point. Intersecting four unreliable fields does not hand you a cleaner audience. It hands you a smaller one that holds the same share of wrong companies, plus a pile of correctly targeted buyers excluded by a stale field.
Those excluded buyers are invisible, which is why this failure survives for years. No report shows you the 200 perfect-fit accounts that dropped out because a page says the wrong industry.
Automated expansion runs the other direction and deserves a look at the same time. LinkedIn's audience expansion setting and Google's optimized targeting both serve beyond the audience you defined. Google states plainly that optimized targeting looks past your manually selected segments, and that it may reduce or stop serving on your own signals when it finds traffic it likes better. You can be four filters deep on paper and still buying clicks outside all four.
Raise one threshold instead of adding a fifth filter
The move that improves quality is unglamorous. Take the single attribute that most separates your good customers from your bad ones, usually company size, and raise it hard. Then watch who replies.
If your real customers start at 200 employees, do not set the floor at 51 and try to patch the gap with three more layers. Set it at 500 and read the reply quality. Bad data cuts both ways, so an aggressive threshold on one field survives it better than a precise threshold on four. I would rather run one filter I know is wrong a fifth of the time than four filters I have no way to audit.
→ One blunt filter carries one error rate. Four precise filters compound four.
→ A high threshold is recoverable. Companies misreported as smaller than they are still get caught by the tier below, and you can lower the floor in a single step.
→ You keep a variable you can read. When results change you know which lever moved.
Judge the threshold on who books meetings and who shows up to them, not on the audience count and not on cost per lead. Volume at a higher floor usually drops. That is the trade you are making on purpose.
Do the real qualifying after the click
Targeting decides who can see the ad. Everything that decides who converts happens later, and it is more trustworthy because it reads behavior instead of self-reported profile fields.
→ Ad copy. Naming the buyer, the price band, or the company size in the ad repels the wrong reader more accurately than a filter does, because the reader knows their own headcount. We covered how to write that in B2B Google Ads copy that repels the wrong buyer.
→ The offer itself sorts by seriousness when it demands something. An implementation call sorts. A free download sorts by nothing.
→ Business email gating on the form. Consumer-domain signups engage measurably less in sales-led B2B, so requiring a work address removes a chunk of them without a single targeting change. Asking for company size on the form is self-reported too, but that answer comes from the person rather than a page, and you can hold it against the CRM record later.
→ The landing page, which out-qualifies a fifth audience layer any day it names the exact situation you solve. That is the argument in B2B PPC landing pages.
The bias in these filters points the right way. A targeting filter excludes a buyer silently. A demanding offer excludes a buyer who told you they were not interested. Only one of those gives you information.
A build order for a B2B paid audience
1.) Define the account tier you actually sell to, in your CRM, not in the ad platform. Pull closed-won deals by employee count and find where wins really start.
2.) Set one primary firmographic filter at or above that line, deliberately too high.
3.) Turn off audience expansion and optimized targeting in the campaigns where you are testing the audience. A targeting test the platform is allowed to override is not a test.
4.) Move the qualification into the copy, the offer and the page.
5.) Add a second targeting layer only once you have data showing the first one is not the constraint.
6.) Recheck the firmographic assumption against replies and booked meetings every month. These fields decay quietly.
When firmographic targeting is worth paying for anyway
There is a real case for it, and it is narrower than how it gets sold.
Firmographic targeting earns its premium when the wrong audience is not merely useless but expensive. Enterprise-only products with long cycles, regulated buyers, and any account-based program working from a named list are worth the CPM. Company list targeting is also a different animal from company size targeting, because you supplied the names and you know what is on the list.
Where it does not earn its keep is as insurance against a weak offer. No combination of filters makes an undifferentiated ad produce sales-accepted leads, and an hour spent tuning the audience is an hour not spent on the message. That is the direction we push clients in on LinkedIn Ads and in Google Ads work, and the answer is the same in both places. It is also why our B2B SaaS paid programs start with the offer and the page rather than the audience builder.
Check this week
Pull the last 90 days of closed-won deals and their real employee counts out of your CRM. Compare that spread against the company size filter running in your ad accounts right now. If the filter's floor sits below where your wins begin, raise it to the win line and delete one other targeting layer in the same session.
Then check whether audience expansion or optimized targeting is switched on in those campaigns. If either one is, you have not been testing your targeting at all.
