How do I build a bottom-up market-size estimate?

Build a bottom-up market-size estimate by defining the unit a customer would buy, identifying distinct eligible buyers and applying a supported price and purchase frequency. Check the source's coverage, date and counting unit before multiplying. Keep the resulting potential separate from expected sales, accepted orders and delivery capacity.

A large number of businesses can be a useful starting point. It becomes misleading when every location is treated as a separate buyer, every buyer is assumed suitable and every suitable buyer is assumed ready to purchase.

Define the offer and who makes one purchase

Write the actual product or service, scope, period and pricing unit. Is the proposed fee per person, business, location, transaction or year? A count is useful only if it matches what generates one purchase.

Identify the buying decision. Several stores may buy centrally, while another organization may allow each location to purchase separately. A legal entity, an operating site and a procurement decision are not automatically the same unit.

The SBA's market-research guidance separates questions about demand, market size, reach, competition and pricing. Answer them explicitly rather than letting one industry total stand in for all of them.

Also define what this estimate is for. Comparing two possible customer groups calls for a different level of evidence from committing to staff and equipment. State which decision the estimate can inform and which later questions remain open.

Read the data definition before the number

Record the source, reference period, geography, industry classification, included population and counted unit. The year you download a table is not necessarily the year its businesses were measured.

The Census Bureau's Statistics of U.S. Businesses overview describes employer-establishment data and fields for firms and establishments. Its methodology explains that an establishment is a physical business location and an enterprise can have several. The program also has exclusions, including businesses without employees.

Those definitions matter if your intended customers include solo operators or companies with many sites. An employer-only table does not automatically count every possible business customer, and a site count does not automatically count distinct purchasing organizations.

Keep a copy or precise reference to the actual table and its notes. Do not replace a suppressed or missing value with zero, or quietly combine incompatible years and categories merely to complete a spreadsheet.

Public aggregate data can establish a useful universe without giving you a contact list or revealing which organizations have the need your offer addresses. The next classification work still belongs to your estimate.

One original roster, not a Census statistic

Consider a wholly fictional business considering an annual catalog-checking service at a complete $1,200 per buying organization. The defined package has one annual purchase per organization; adding locations does not increase the fee in this supplied offer.

An invented research roster contains 1,200 physical locations. Its documented ownership and purchasing relationships connect those sites to 800 distinct buying organizations. Some have multiple sites but would make one purchase for this particular package.

These counts are original teaching inputs. They are not downloaded Census figures, actual research findings, contacted prospects or an estimate for a named city or industry.

Multiplying all 1,200 locations by $1,200 would suggest $1.44 million of annual-equivalent contract value. Correcting the counting unit to 800 buying organizations reduces that preliminary figure to $960,000.

The $480,000 difference comes from counting the same buying relationship more than once. It does not mean those additional sites lack business activity; they simply are not additional purchases under the defined offer.

Apply eligibility filters in a reproducible order

Of the 800 fictional buying organizations, 600 have the operating profile the service is designed for. The other 200 do not. Within those 600, only 400 have the defined catalog workload that makes the specified package relevant.

Among those 400, current permitted sales channels can reach 200. Another 150 are outside those channels, and access to fifty remains unresolved. The current supported reachable base is therefore 200 organizations.

These classifications are supplied facts for the invented roster, not assumptions that a public industry code reveals need or that finding an email address establishes permission to contact someone.

Scroll across to follow the original sequential count. Each row starts from the surviving group above it.

Retain the reason and evidence for each organization's classification. Applying filters sequentially avoids subtracting the same unsuitable organization several times because it fails more than one criterion.

Reconcile exclusions and the unresolved group

The original 800 organizations reconcile as 200 outside the operating profile, 200 without the required workload, 150 outside current channels, fifty with unresolved access and 200 in the supported reachable base.

The fifty unresolved organizations are not proven unreachable, but they are not confirmed reachable either. Keep them in a separate investigation group with the specific missing access information.

If every one later becomes confirmed reachable and the other supplied criteria remain satisfied, the base would become 250. At $1,200 each, its annual-equivalent potential would be $300,000, or $60,000 above the main case.

That is a conditional classification change, not a prediction that fifty customers will buy. If no access is confirmed, the supported base remains 200. An unresolved field should not automatically select the larger estimate.

Preserve the original roster and classification version. When a record changes, identify whether new evidence changed the buyer, need or channel status. Do not rewrite an earlier estimate so it looks as though the new information was known all along.

Multiply by a price that matches the offer

The main reachable estimate is $240,000: 200 organizations multiplied by the assumed $1,200 annual package price. It describes the amount if every organization in that defined group bought one package at that price.

It does not establish willingness to pay, timing, renewals, accounting revenue or cash collections. A full annual contract beginning late in a calendar year is also different from a full year's earned revenue during that calendar year.

If the actual offer is priced per location, per user or by workload, rebuild the model with that price structure. Do not keep a per-organization buyer count beside an unrelated per-user price because the multiplication produces a more attractive total.

Check whether an annual figure includes setup, required usage, discounts or other necessary charges. If you use a price assumption rather than an established offer, label it. A competitor's headline price is evidence about that offer, not proof your own customers will pay the same amount.

In a separate price branch, reducing the assumed package price to $900 makes the same 200-organization potential $180,000. The $60,000 reduction changes a scenario input; it does not demonstrate that the lower price generates enough extra demand to compensate.

Do not call an adoption target an observed rate

Suppose the founder sets a first target of winning 10 percent of the 200 reachable organizations. That means twenty packages and $24,000 of annual-equivalent contract value if they buy on the supplied terms.

The target does not become a forecast merely because 10 percent is easy to enter. This fictional file contains no accepted orders. Contracted value from these leads is currently zero.

The new-business sales-forecast guide handles the next task: specific prospects, decision stages, dated delivery and collections. Market sizing supplies a defined pool to investigate; that forecasting process establishes how actual sales could develop.

Likewise, twenty sales would require the actual work, staffing and funding to deliver. This estimate supplies no delivery-capacity finding and no whole-business profit calculation. A reachable customer pool is not an invisible production team.

Fund the research without pretending it sells the service

For this bounded research exercise, a paid analyst has thirty scheduled hours at a complete $40 per hour. Ten hours are already reserved for other accepted work, leaving twenty for the defined roster and classification review.

Those twenty hours cover identifying sources, reconciling the supplied buying relationships, applying the stated filters and documenting the estimate. Public source access costs zero in this example, and there are no other research payments or unpaid founder tasks.

The research costs $800, funded from a separate $1,000 allocation after protecting other obligations. It leaves $200. All thirty analyst hours are funded across their respective assignments; the market review does not take the ten hours belonging to existing work.

No customer interviews, outreach campaign, legal review, purchased contact database or launch activity is included. If evidence requires one of those, establish its actual scope and resources separately. An orderly estimate is not proof those later tasks have happened.

Leave with an estimate someone can challenge

Keep the offer, buyer unit, source reference and year, coverage limitations, distinct buyer count, sequential filters, unresolved records and price assumptions together. Label the final amount with its scope and period.

Ask what would materially change it. A centrally purchased service, a different workload threshold or an unavailable channel can alter the answer much more than another decimal place in a market-share percentage.

Use the business-test guide for a separate evidence decision about paying demand and the operation you would actually run. A market estimate can identify where to investigate without establishing that an expansion or launch is ready.