Measuring Financial Inclusion: From Theory to Evidence

From Theory to Measurement

In a companion study, Who Really Uses Bitcoin ATMs?, we examined the published research on who relies on cash-to-Bitcoin kiosks and why. The Federal Reserve Bank of Kansas City and the FDIC both point to the same conclusion: the typical user is not a speculator but a member of a community that the conventional banking system has historically reached last — lower-income households, immigrants, and minority neighborhoods. That is the academic literature. A fair question follows: does a real, operating Bitcoin ATM network actually serve those communities, or is the inclusion story merely aspirational?

This article answers that question with measurement rather than narrative. Using a standard, peer-reviewed statistical method and the aggregate records of a national Bitcoin ATM network, we estimate the demographic composition of the activated customer base and compare it to the United States population. The result is a quantitative test of the financial-inclusion hypothesis — and it is, we believe, the first time an operator has published such an estimate openly.

How We Measured It — and How We Did Not

Estimating the racial and ethnic composition of a customer base is a well-developed problem in quantitative social science, precisely because direct self-reported data is usually unavailable. The established solution is Bayesian Improved Surname Geocoding (BISG), a probabilistic method that combines two public data sources: the U.S. Census Bureau's distribution of self-reported race by surname, and the Census distribution of race by residential geography. Bayes' theorem combines them into a probability that a given individual belongs to each racial or ethnic group. The method is not fringe: the Consumer Financial Protection Bureau adopted BISG as its standard proxy for fair-lending analysis in 2014, and it is used throughout banking supervision, public-health research, and political science for exactly this purpose.

Two points about method are essential, because they define what this estimate is and is not.

First, BISG produces probabilities, not labels. No individual customer is assigned a race. Instead, each record contributes a fractional probability to each group, and those fractions are summed across the entire base to yield an expected composition. A single person might count as, say, 0.7 toward one group and 0.3 toward another. Aggregated across tens of thousands of records, the law of large numbers makes the population-level estimate far more reliable than any individual inference would be.

Second, no facial recognition, photograph, or biometric data was used. The analysis draws only on surname and ZIP code — the same inputs a bank regulator uses — and operates exclusively on aggregate statistics. No customer photograph was processed, no identity record was exported, and no individual-level result was retained. This is a deliberate methodological and ethical choice: the privacy-respecting public-data method is also the more accurate and defensible one.

The estimate is based on a sample of 45,973 activated customers — those who completed verification and transacted, and for whom a valid surname and U.S. ZIP code were available. As with any model, the figures below carry error and should be read as directional estimates rather than exact counts.

What the Data Shows

The headline result is unambiguous. The Bitcoin ATM customer base substantially over-represents Black Americans relative to their share of the national population, and concentrates regionally among Hispanic communities in the Southwest. Set against the 2020 U.S. Census, the estimated composition is as follows:

Group	Estimated customer share	U.S. population (2020)	Representation index
Black	22.6%	12.1%	1.87×
White	64.4%	57.8%	1.11×
Hispanic	7.8%	18.7%	0.42×
Asian / Pacific Islander	2.9%	5.9%	0.49×
Native American	0.6%	0.7%	0.86×

The representation index — customer share divided by population share — is the figure that matters. A value of 1.0 would mean the customer base mirrors the country. The estimate for Black Americans is 1.87, meaning they are represented at nearly twice their national share. This is not a marginal skew; it is the defining feature of the customer base.

Why This Is the Signature of Financial Inclusion

A demographic profile is only meaningful in context, and the relevant context is financial exclusion. The FDIC's 2023 National Survey of Unbanked and Underbanked Households — the authoritative U.S. benchmark — found that 10.6% of Black households are entirely unbanked, compared with 1.9% of white households. Black Americans are, in absolute terms, the largest unbanked population in the country. Hispanic households follow at 9.5%, and Native American households are highest of all at 12.2%.

Place the two datasets side by side and the alignment is striking: the group the Bitcoin ATM network serves most disproportionately is precisely the group the conventional banking system reaches least completely. The 1.87× over-representation of Black customers is not a coincidence of geography — it is the demographic shadow of the 24.6 million American households that, as documented in the companion study, operate outside the mainstream banking system. The network is, in measurable fact, populated by the communities the financial-inclusion literature predicts.

The regional picture reinforces this. The estimate varies sharply by state, tracking local community composition rather than a single national pattern:

State	Black	Hispanic	White
Georgia	51%	3%	43%
Alabama	30%	2%	65%
California	8%	30%	47%
Texas	18%	20%	56%
Florida	22%	11%	64%

A network of physical machines, placed in the corner stores and gas stations where people already are, naturally comes to reflect the neighborhoods it sits in. In the Southeast that means a predominantly Black customer base; in the Southwest, a substantial Hispanic one. The machine does not select its community — the community selects the machine, by walking up to it. As we describe in the customer-profile study, the overwhelming majority of users arrive on foot, locally, without any digital advertising touchpoint.

From Access to Wealth-Building

Access is the beginning of the story, not the end. The deeper significance of these numbers lies in what these communities are gaining access to. A check-cashing window converts cash into cash, minus a fee. A money order moves value sideways and then stops. A Bitcoin ATM, by contrast, converts cash into a bearer asset — a digital instrument the holder controls outright, with no custodian, no minimum balance, and no counterparty — that has, over its history, appreciated.

This is a categorical difference. For the first time, a household that the banking system offered only fee-extracting, value-flat products can access, in cash, at 11pm, a savings instrument with the potential to grow. The behavior we observe in the data is consistent with this: many users do not trade, they accumulate, returning to convert modest amounts of cash on a recurring basis — the pattern we examine in The Gas-Station Savings Plan. It is dollar-cost averaging, conducted in twenty-dollar bills, by people for whom a brokerage account was never on the table.

Intellectual honesty requires a caveat, and it is an important one. Our data also shows that the highest-balance segment of the customer base skews closer to the national wealth distribution — that is, the largest accumulators are disproportionately drawn from groups that already hold more wealth. This mirrors the well-documented racial wealth gap measured by the Federal Reserve's Survey of Consumer Finances, in which median white household wealth remains several times that of Black and Hispanic households. Bitcoin ATMs do not erase that gap overnight, and we make no such claim. What the data does show is that they widen the door: they place an appreciating, self-custodied savings instrument within physical reach of communities that were previously offered only the depreciating or extractive end of the financial product spectrum. Wealth-building begins with access, and access is what these numbers measure.

The Broader Significance

Financial inclusion has, for decades, been framed as a problem of extending conventional banking to the unbanked — opening more branches, lowering minimum balances, simplifying account terms. Those efforts matter. But the data here points to something the inclusion debate has largely overlooked: a parallel, cash-native, physically-distributed financial infrastructure has quietly grown up in exactly the neighborhoods the branch network has been leaving, and it connects those neighborhoods not merely to payments but to ownership of an appreciating asset. Whether or not one is bullish on Bitcoin as an investment, the inclusion fact stands on its own: a regulated, KYC-compliant, cash-accessible on-ramp to digital savings is now disproportionately used by the communities most excluded from traditional finance.

This also reframes the policy stakes, and it does so with arithmetic rather than rhetoric. When a jurisdiction bans Bitcoin ATMs outright — as Indiana did in 2026, shutting roughly 900 machines overnight — the access that is removed does not fall evenly across the population. It is withdrawn first and most completely from the communities this data identifies: the unbanked, the cash-dependent, and the disproportionately Black and Hispanic neighborhoods that the conventional branch network had already been leaving. The fraud that typically motivates such bans, as we document in the companion study, simply migrates to less-regulated channels — wire transfers, gift cards, cash by mail. The inclusion does not migrate. It ends. A measure framed as protecting vulnerable people can, when measured honestly, remove a regulated and supervised financial lifeline from precisely the communities that have the fewest alternatives to fall back on.

None of this argues against addressing fraud, which is real and serious. It argues for addressing it with the instruments that preserve access — KYC, transaction monitoring, live human intervention, and the five-layer safeguards a regulated operator already runs — rather than with prohibition, whose costs land hardest on the people least able to absorb them. That is a story worth measuring carefully and telling accurately — which is why we have published the method alongside the result.

Methodological Notes and Limitations

In the interest of academic transparency, the principal limitations of this estimate are stated plainly:

BISG is an estimate, not ground truth. It infers probable race and ethnicity from surname and geography; it does not observe self-reported identity. Validation studies (including the CFPB's own) find BISG most accurate for the White, Black, and Hispanic categories and less precise for multiracial and certain Asian subpopulations. Individual-level inference is unreliable by design; only the aggregate is meaningful.

Coverage is incomplete. Records whose surname or ZIP code is absent from the Census reference tables cannot be resolved and are excluded, which can introduce mild bias if those exclusions are non-random.

The comparison baseline is the general population. We benchmark against the 2020 U.S. Census. A more refined analysis would benchmark against the population within each machine's local catchment area; the state-level figures above are a step in that direction.

No personal data was exposed. The analysis was conducted on aggregate statistics derived from internal records; no individual customer's name, location, photograph, or inferred demographic was printed, stored, or transmitted as part of this work.

With those caveats, the central finding is robust to reasonable variation in method: a real, national Bitcoin ATM network is used at nearly twice the expected rate by the single largest unbanked population in the United States. The communities the headlines miss are not a rhetorical device. They are, by measurement, the customer base.