32 Million Cannot Be Scored: What an Expanded-Data Model Reads

There are two ways an application comes back with nothing useful in it. One is a number you do not like. The other is no number at all, which is harder, because there is nothing to argue with and nothing obvious to fix.

Experian sells a product called Lift Premium that it says expands the scoreable universe to 96% of U.S. adults, against 81% using conventional scores. One thing first: it already exists and has for years. It did not launch this month, whatever the coverage suggests. What is happening in 2026 is adoption, with issuers and lenders reaching for AI credit and alternative data in volume.

Fifteen percentage points is the whole claim. This is what they are made of, what can honestly be said about the data behind them, and the two things that matter when you hold the empty file: how an approval turns into a conventional credit file, and what you are owed when the answer is no.

The whole subject rests on a distinction between two groups that get discussed as one.

Credit invisible means there is no file. The nationwide bureaus hold nothing under your name: no tradelines, no collections, nothing to compute against. There is no input to score.

Unscoreable means a file exists and is not enough. It might hold one account opened four months ago, where the model wants six. It might hold accounts nobody has updated in years, so every field is stale. It might hold a single collection and no credit account. There is something there; the model reads it and declines to return a number.

One caution on the 32 million: whether a tally folds the invisible in with the unscoreable is a definitional choice, not a fact. When a headline hands you a number here, ask which group it counts.

Do the subtraction.

If conventional scores reach 81% of U.S. adults, 19% sit outside them. Experian's figure for Lift Premium is 96%, which leaves 4% outside after the expansion. The difference, 96 minus 81, is 15 percentage points.

Fifteen of the nineteen points that were outside is about 79%, or four in five. So roughly four of every five adults a conventional score cannot reach, an expanded-data model can. The other one in five stays unreachable even with the wider inputs, with no route in on offer.

Experian also reports coverage of an estimated 65% of the credit-invisible population, lower than the four-in-five overall, and it should be. A file with something in it is easier to extend than one with nothing, so the group with no bureau record is where an expanded model does its worst work.

Apply the four-in-five ratio to the Federal Reserve's 32 million and you get roughly 25 million adults moving inside the scoreable universe and roughly 7 million staying outside. That arithmetic is mine, not Experian's: the company does not break its coverage down that way, and the two populations are not measured on the same basis. Treat it as an order of magnitude. The honest version of the claim is the ratio, and most of the gap, not all of it, and least of all where it is widest.

So what is a model of this kind reading, when the conventional file is nearly blank. Start with the limit on what can honestly be said.

Experian does not publish a ranked input list for Lift Premium. What its public description identifies is expanded FCRA-regulated data, including public records and Clarity alternative-finance data. That is the substantiated part, and narrower than most coverage implies.

Around it sits the wider category: recurring obligations the core bureau file never collected: rental payments, utility and telecom accounts, bank and cash-flow data, alternative lending records. Which of those any model weighs is not public. Read it as a category, not a ranking: the order is what nobody outside the vendor knows.

An expanded-data score is not a FICO score. Three things follow, and the third is the one worth acting on.

First, the expanded score itself is not written into your conventional score. It is a separate product computed for a lender that chose to buy it; if a mortgage underwriter later pulls a classic FICO, nothing the expanded model concluded shows up there.

Second, no lender is required to use one. Adoption is a commercial decision, made lender by lender and often product by product, so 96% describes what could be scored if a lender bought the capability, not what will be scored when you apply.

Third, and this is the part worth the whole article: the account is not the score. If an expanded-data model gets you approved and that account reports to the nationwide bureaus, you have a tradeline you did not have before. Once it meets FICO's eligibility requirements, your file can support a conventional FICO score that did not previously exist. The expanded model did not give you a FICO score, because there was none to give. It got you the account that produces one.

That is the practical route out of an empty file, and why an approval from an alternative underwriter is worth taking seriously even if the product is not one you would have chosen. Ask before accepting: does this account report to all three nationwide bureaus. One that reports nowhere builds nothing.

A decision made with an alternative model is still a credit decision, and that is where your leverage is. The two statutes are not coextensive and they do different jobs.

The Equal Credit Opportunity Act governs adverse action on a covered credit application, whether or not a credit report was involved. Section 1691(d)(3) says a statement of reasons qualifies only if it contains the specific reasons. The familiar wording comes from Regulation B, 12 CFR 1002.9(b)(2): the statement must be specific and indicate the principal reasons.

The Fair Credit Reporting Act carries separate notice duties that turn on whether a consumer report was used, and they are not a second helping of the same thing. Section 1681m(a)(3)(B) requires the FCRA notice to state that the bureau did not make the decision and cannot give you the specific reasons, so it expressly disclaims the job ECOA assigns to the creditor.

One provision matters more here than the rest. Under section 1681m(b)(1), where credit is denied or the charge for it increased because of information from a source other than a consumer reporting agency, the user must disclose the nature of that information on written request made within 60 days, and must tell you at the time of the adverse action that the right exists. That is exactly the alternative-data case: information that never came through a bureau can fall outside section 1681m(a) entirely and still be reachable here.

Reasons are not always delivered unprompted: section 1691(d)(2) lets a creditor either give them as a matter of course or notify you that you may request them, with 60 days for you to ask and 30 for it to answer. Entitled to is not the same as handed to you.

If you are the empty file right now, the expansion of somebody else's scoreable universe is not a plan. This is:

• Pull your reports from all three nationwide bureaus and find out which situation you are in. Nothing at all is credit invisible; something insufficient is unscoreable.
• Read what is on the file before assuming it is thin. One holding only a collection is unscoreable for a reason you can act on.
• Before accepting any approval, confirm the account reports to all three nationwide bureaus. That is what turns an approval into a credit file.
• Then let it age. Conventional models want an account roughly six months old before they return a number.
• Ask any lender you are considering whether it uses alternative or cash-flow data. The ones that do tend to say so.
• Keep the rent and utility record clean regardless. It is the kind of obligation an expanded model is built to read.
• If you are denied, ask in writing for the specific reasons, and if the decision drew on information from outside a credit bureau, ask what that information was. Both requests run on a 60-day clock from the adverse action.

Those last requests are worth making even when a denial feels closed. The reasons are the only direct readout of what a model saw in your file.

A word on reading the next announcement, because there will be one. The pattern in 2026 has been coverage that treats a coverage statistic as a consumer benefit. A vendor saying it can score 96% of adults is describing the reach of a model it sells to lenders, not saying 96% of adults will be approved for anything.

So: is this a launch, or an existing product written about again. Is the number about who can be scored, or who was approved. Does the piece distinguish credit invisible from unscoreable. And if it ranks what a commercial model reads, where did that ranking come from, because the vendors mostly do not publish one.

Fifteen percentage points. Four in five of the people a conventional score cannot reach, and one in five left where they were. Roughly 25 million of the Federal Reserve's 32 million moved inside on that ratio, with about 7 million outside. That is my arithmetic rather than the vendor's, and an order of magnitude rather than a count.

The part to carry out of this: an expanded-data score is not your FICO score, but an account it gets you approved for can become the tradeline that gives you one, provided it reports. That is the actual route off an empty file. If the answer is no instead, ECOA entitles you to the specific reasons, often only once you ask, and where the decision drew on information from outside a credit bureau, section 1681m(b)(1) lets you ask what that information was.

So: confirm which situation you are in, get an account that reports, let it age, keep the rent and utility record clean, and ask every lender what data it underwrites on. None of that depends on which vendor wins the alternative-data argument.

1. What does it mean to be unscoreable?

It means a credit file exists but does not contain enough usable information for a scoring model to return a number. A common cause is that the only account on the file is younger than the roughly six months conventional models require, or that nothing on the file has been updated recently enough to be current.

2. Is credit invisible the same as unscoreable?

No. Credit invisible means the nationwide bureaus hold no file at all under your name. Unscoreable means a file exists and is too thin or too stale to score. They fail at the same counter for opposite reasons, and the route out differs: one needs a file to exist, the other needs the existing file to mature or gain a second entry.

3. What is Experian Lift Premium?

It is an existing Experian scoring product that uses expanded FCRA-regulated data beyond the conventional credit file, including public records and Clarity alternative-finance data. Experian says it expands the scoreable universe to 96% of U.S. adults against 81% using conventional scores, and reports on the 300-850 range. It is not a 2026 launch.

4. Does 81% to 96% mean 32 million people become scoreable?

Not exactly. The 15-point expansion is about 79% of the 19 points that sat outside conventional scoring, so roughly four in five. Applying that ratio to the Federal Reserve estimate of 32 million unscoreable adults gives around 25 million inside and around 7 million still outside, but that is an inference from the percentages rather than a figure Experian publishes.

5. Does an expanded-data score change my FICO score?

The score itself does not. It is a separate product and nothing it concludes is written into a conventional score. But the approval can matter a great deal: if the account you are approved for reports to the nationwide bureaus, it becomes a tradeline, and once that tradeline meets FICO eligibility requirements your file can support a conventional FICO score it could not support before. Confirm the account reports before accepting it.

6. Do lenders have to use an expanded-data score?

No. Adoption is a commercial decision made lender by lender and often product by product. A coverage figure describes what could be scored if a lender bought the capability, not what will be scored when you apply.

7. What data does Lift Premium actually use?

Experian does not publish a ranked input list. Its public description identifies expanded FCRA-regulated data including public records and Clarity alternative-finance data. Expanded-data underwriting as a category also draws on rental payments, utility and telecom accounts, bank and cash-flow data, and alternative lending records, but how any one model weighs them is not public. Be sceptical of articles that rank a specific product inputs.

8. If an alternative model denies me, what am I entitled to?

9. What if the denial used data that did not come from a credit bureau?

10. Is a lender allowed to say I just failed to score high enough?

No. Regulation B, 12 CFR 1002.9(b)(2), states that reasons resting on the creditor internal standards or policies, or stating that the applicant failed to achieve a qualifying score on the creditor credit scoring system, are insufficient. That rule long predates the current argument about AI. CFPB Circular 2022-03 applies the same standard to complex algorithms: reasons must be specific and accurate and tied to the factors actually scored, and being unable to explain the model is not a defence.