How to build a total addressable market from job postings

Say you want to find every company running Google Ads. What company filter would you use? Industry? Size? Geography? None of them tells you that. But a job posting might.

This guide shows how to build a total addressable market from what companies write in their job postings, with the same examples as the video: why company filters miss part of your market, how TAM by Jobs turns postings into companies, and how crossing several searches tells you which companies really belong.

What is a total addressable market?

A total addressable market, or TAM, is everyone who could buy what you sell. Finance teams count it in revenue. Sales and marketing teams need it as a list: the companies that could become customers, before they choose which ones to work first.

Definition Total addressable market also called TAM

Everyone who could buy what you sell. Finance teams count it in revenue; go-to-market teams turn it into a list: every company that could become a customer, before you decide which ones to work first.

PreciselyTotal addressable market (TAM) is the total demand for a product if it won every possible customer, usually expressed as annual revenue. Serviceable addressable market (SAM) is the part your offer can reach, and serviceable obtainable market (SOM) the part you can realistically win. In B2B sales, a TAM list is the account-level version: the companies that make up that market.

Most teams build that list the same way. They open a company database and filter it by industry, size and geography. The filters are quick, and they all read the same thing: what a company says it is.

What can a job posting tell you that a company filter can’t?

What is actually happening inside the company. To attract the right candidate, a company has to be specific: the tools it uses, the equipment it runs, the teams it is building, the projects it is starting. Three postings from the video show it.

Job postings are also unusually honest. No company publishes its equipment list or its software stack, but every company publishes what its next hire needs to know. The requirements are agreed between the hiring manager and HR, and a requirement that isn’t real costs the company failed interviews. Nobody asks for FANUC experience without FANUC robots on the floor.

How does TAM by Jobs turn job postings into a list of companies?

It searches what companies write in their job titles and descriptions, and returns the companies behind the matching postings: each company once, with the number of postings that matched. You describe the evidence; the search finds the companies.

Definition TAM by Jobs also called a company search that runs on job postings

A search that reads what companies write in their job postings and gives you back the companies behind them, each one once, with the number of postings that matched.

PreciselyThe BlitzAPI endpoint POST /v2/company/tam-by-jobs filters live job postings by title, description, AI keywords, field, seniority, employment type, work arrangement, job location and date posted, together with company firmographics, and returns the distinct companies behind the matching postings, each with a matched_jobs count. One record per company returned.

Each company comes back with its profile too: industry, size, headcount, headquarters and website. The list is ready to sort, without a second lookup.

Why does a label-based total addressable market miss companies?

Because a label is what a company says about itself, not what it does. Take a harder example: industrial automation. A normal TAM might start with three industries: Industrial Automation, Manufacturing, maybe Automotive.

Now you are trusting the label. The search can’t tell which of those companies actually run automated lines, and it never looks at the pharmaceutical or aerospace companies whose plants are full of them.

So ask a different question: what would a company have to reveal for you to know automation is really happening inside it?

Which hiring signals prove a company belongs in your market?

The ones that would be hard to post without the real thing behind them. For industrial automation, the playbook behind the video splits the evidence into six families: roles, equipment and vendors, technology stack, infrastructure, process technology and project intent.

Each family proves something different. A Controls Engineer tells you the function exists. FANUC, a robot maker, gives you evidence about the equipment. “Commissioning” or “greenfield” tells you something even more valuable: something is happening now.

In the actual playbook, those six families became seven independent detectors, one search each:

The point is not to write one perfect query. Each query is one piece of evidence: broad ones give you coverage, specific ones carry more weight, and project intent gives you timing.

Why are duplicates evidence?

Because each duplicate is a company seen by one more independent search. Run the detectors and the signals start to cross: the same company comes back, again and again. Normally you would deduplicate it and move on. Do the opposite: count it.

Definition Signal convergence also called convergence scoring

Counting how many independent searches found the same company. Found once, a company may be a coincidence; found by four different kinds of evidence, it almost certainly belongs to your market.

PreciselyConvergence scoring ranks each company by the independent signals that matched it. In the TAM by Jobs playbook, a company's score is the sum of its signal weights (1 for a broad role, up to 5 for a live project) multiplied by a convergence factor: ×1.00 for one signal, ×1.15 for two, ×1.30 for three and ×1.50 for four or more, plus a capped bonus for hiring volume.

One request to TAM by Jobs never returns the same company twice. The duplicates appear when you put several searches side by side, and they are the raw material of the score.

Does signal convergence make a TAM list more precise?

Yes, and the playbook measured it. Its reference run looked for industrial automation with seven searches over 180 days, on August 5, 2026. The seven searches returned 38,868 rows. Merged, they made 23,843 unique companies: 15,025 duplicates, each one a company seen by one more signal.

To check the list, the playbook counted the manufacturers in it: companies whose LinkedIn industry is one of 96 manufacturing industries. Among the companies found by a single signal, 57% were manufacturers. Among the 176 found by all seven, 87% were. That share is what the video calls precision, and it is a conservative one: a manufacturer with no industry on file counts as a miss.

So you don’t need every discovery query to be perfect. You can discover broadly, and recover precision through convergence.

Try it: how many signals should a company match?

Pick a threshold. The simulator reads the measured table of the reference run: how many companies you would work first, and how many of them are manufacturers.

The companies under your threshold don’t leave the TAM. They stay in it, lower in the list, and they move up the day a new signal finds them.

What is the hidden market?

The companies that are in your market without carrying its label. Among the companies matching all seven signals in the reference run were AbbVie, Airbus, Alstom, Amcor and American Honda: a pharma company, an aerospace company, a train manufacturer, a packaging group, a carmaker. Same operational reality, different labels.

Of the top eleven companies matching all seven signals, only two appeared in an industry search on automation: ABB and Agile Robots.

We ran a version of the robotics detector on October 24, 2026, on job postings located in Poland: postings that name FANUC, KUKA or Yaskawa robots, at least two per company, confirmed staffing agencies excluded. Fifty-eight companies came back. Six carry an automation label. Most of the others are manufacturers that run these robots on their own lines: car parts, tires, machinery, medical devices, furniture, food. ZF, Valeo, Michelin, B. Braun, Stryker and Nestlé are among them.

Why does job location matter more than headquarters?

Because headquarters tells you where the company is registered, and job location tells you where the work happens. A German group hiring Controls Engineers in Poland? That’s a Polish plant: potentially a different territory, a different owner, a different conversation.

The October run shows how common that is. Of the 58 companies hiring around those robots in Poland, 30 have their headquarters in another country: the United States, Germany, France, Switzerland and five others. Route them by headquarters, and the account lands with a team that has never seen the plant.

Job location also fills gaps. In the reference run, 4,753 companies had no headquarters country on file, while the places where they were hiring were known.

How do you build a total addressable market from evidence, step by step?

Stop asking which companies fit your filters. Start asking what evidence would prove a company belongs in your market, then let the evidence find the companies. The playbook does it in three stages, and keeps them apart:

1. Discovery: which companies could plausibly be in your market? Aim for coverage. Run one search per signal and remove only what can’t be a buyer: staffing agencies, and the industries you have measured as noise.
2. Scoring: how strong is the evidence? Simple arithmetic: the weight of each signal a company matched, multiplied by its convergence.
3. Prioritization: which accounts do you work first? This is where your ideal customer profile applies: size, geography, industry, territory. A weak score moves a company down the list; it doesn’t remove it.

A company search by industry still has a place: as one more signal, the one with the lowest weight. It adds the companies that are in your market but not hiring right now, and it never outranks what the postings prove.

Then the market becomes contacts. Once you know which companies to work, Waterfall ICP finds the right decision maker in each one, as the decision makers guide explains. And the signals that put a company in your TAM give you your first line: “I saw you’re commissioning a new line” opens a different conversation from “I see you work in manufacturing.” For the full workflow and the APIs around it, see how to build a TAM with BlitzAPI.

For builders

Everything above runs on one endpoint. This part is for the people who will wire it up: the call, a real run, how filters match, the score, and what to check before you ship.

How do you run TAM by Jobs with the API?

POST/v2/company/tam-by-jobsAPI reference

Job posting filters in, the distinct companies behind the matching postings out, each with its matched_jobs count.

Send

job.title, job.description, job.ai_keywords

What the postings must say: the role in the title; tools, vendors and projects in the description or the AI keywords.

job.location, job.date_posted

Where the job is, not the headquarters, and how many days back to look.

job.min_per_company

Keep only the companies with at least this many matching postings, up to 25.

company

Firmographic filters (industry, size, headcount, headquarters), and is_agency: false to drop confirmed staffing agencies.

max_results, cursor

Up to 50 companies per page. Send the cursor back until it comes back null.

Read

results[].company

The company profile: name, LinkedIn page, industry, size, headquarters, website.

results[].matched_jobs

How many of its postings matched: the strength of this signal for this company.

cursor

The next page, or null at the end.

Each company returned is one record; plans and rate limits are on the BlitzAPI pricing page. To read the postings behind a company, run the same job filters on Jobs Search.

One detector, one request: a real run

On October 24, 2026, we ran a version of the robotics detector on job postings located in Poland, over the last 180 days, with at least two matching postings per company and confirmed staffing agencies excluded:

{
  "job": {
    "description": { "include": ["FANUC", "KUKA", "Yaskawa"] },
    "location": { "country_code": { "include": ["PL"] } },
    "date_posted": { "last_days": 180 },
    "min_per_company": 2
  },
  "company": { "is_agency": false },
  "max_results": 50
}

Two pages later, the cursor came back null: 58 companies. Here is one of them, trimmed to the fields that matter:

{
  "company": {
    "name": "ZF Group",
    "linkedin_url": "https://www.linkedin.com/company/zf-group",
    "industry": "Motor Vehicle Parts Manufacturing",
    "size": "10001+",
    "hq": { "city": "Friedrichshafen", "country_code": "DE" },
    "domain": "zf.com"
  },
  "matched_jobs": 13
}

A German group with thirteen postings in Poland that name industrial robots: the Polish plant from the video, in the data. Adding company.industry with the two automation labels (Industrial Automation, Automation Machinery Manufacturing) to the same request returned 6 companies, all already among the 58. ZF was not one of them.

The run also shows where the noise is. Eleven of the 58 companies sit in industries that mostly host intermediaries: IT services and consulting, internet, advertising services, outsourcing. is_agency: false drops confirmed staffing agencies, not every company that republishes jobs. That is a company-level problem, and the rules below handle it.

How do filters match?

• Different filters combine with AND; values in one list combine with OR. Two titles in job.title.include return the companies hiring for either role.
• A phrase needs all its words, in any order. “TIA Portal” matches a posting that contains “TIA” and “Portal” anywhere. A phrase is only as precise as its most common word: in the reference run, “Step 7” matched 697,492 postings, because “step” and “7” are everywhere. Prefer distinctive names: SIMATIC, PROFINET, TwinCAT, Keyence.
• Titles in the title field; tools, vendors and projects in the description. In the reference run, “Controls Engineer” searched as a broad theme matched 326,130 postings. As a job title it matched 24,705, and the share on target was four times higher.
• Brackets mean an exact title. “[Controls Engineer]” matches that title only.
• job.location is where the job is; company.hq is where the company is registered. Filter on the first to work a territory, and keep the second as a column.
• Noise lives at the company level, not in the posting text. A staffing firm placing a controls engineer writes the same posting as the manufacturer hiring one. Drop intermediaries with is_agency: false, then exclude an industry only after measuring what it costs: in the reference run, Facilities Services held as many industrial maintenance providers as cleaning companies, so it stayed in.

How do you score convergence?

Keep one row per signal and company, with its matched_jobs, then group the rows by company.linkedin_url. The playbook’s score:

score = (sum of the matched signal weights) × convergence factor
        + a capped bonus for hiring volume

convergence factor: 1 signal ×1.00 · 2 signals ×1.15 · 3 signals ×1.30 · 4 or more ×1.50

Globex, found by the PLC stack (3), robotics (3), infrastructure (2) and project intent (5) searches, scores (3 + 3 + 2 + 5) × 1.50 = 19.5 before the volume bonus; a company found by the broad roles search alone, 1.

Keep this evidence score and your commercial score in separate columns. Re-scoring is a spreadsheet operation; collecting the market again is not.

Before you ship

• ✓Qualify each keyword alone before it joins a signal: one request tells you its volume and how much of it is on target.
• ✓Page every signal until the cursor comes back null, and check that none stops at a cap you set: a truncated signal is a sample, not the market.
• ✓Keep the raw rows, one per signal and company. The duplicates are the score.
• ✕Filtering on size, geography or target industry at discovery: those filters belong at prioritization.
• ✓Run the same signals every day with date_posted.last_days set to 1: new companies, new signals and hiring spikes become alerts.
• ✓Store the job locations with each company. They map the plants, and they place the companies that have no headquarters on file.