Age Verification Solutions: Doc vs Database vs Estimation

Age verification solutions for online gambling fall into four families: document plus selfie with liveness, database and bureau checks, facial age estimation, and reusable digital IDs. A common design combines an estimation or database check with a document fallback. Regulators require identity validated before account creation (AGCO, 19+) or before the customer is permitted to gamble (UKGC).

This vendor-neutral guide compares the four families on accuracy evidence, friction, regulator acceptance and cost drivers, then shows how to set an estimation threshold with a buffer for a 19+ market. If you are already at the buying stage, our guide to KYC providers for iGaming covers vendor categories, pricing and pilot design; this page helps you decide which method mix belongs in the RFP.

The four families of age verification solutions

Age verification is the process of confirming that a person meets a minimum age before granting access. In gambling the method has to satisfy two audiences at once: the regulator, who wants evidence, and the player, who wants to deposit quickly. The four families below reflect how the market has organised itself; Yoti's gambling page, for instance, lists document, database, facial age estimation and reusable Digital ID as its routes and describes "six proof of age methods" in total, including anonymous yes/no checks.

Document plus selfie and liveness

Document verification asks the player to photograph a government-issued ID, extracts the date of birth and checks the document's security features. A selfie is matched to the document photo, and Liveness detection confirms the selfie came from a living person rather than a printed photo or a screen. It is the most evidence-rich method and accepted under every rule cited here (AGCO 3.04, UKGC 17.1.1, FINTRAC photo-ID method), but also the highest-friction path. We cover the document flow in depth – accepted documents, capture failures and fraud checks – in our guide to KYC ID verification documents, so this page treats it as the fallback the other methods lean on.

Database and bureau checks

A Credit bureau check or other database method matches the name, date of birth and address the player typed against records held by a bureau, electoral roll, telecom or government source. No camera, no upload – the player never leaves the form. FINTRAC's Guide 11 recognises the credit file method and the dual-process method (two independent reliable sources) alongside government-issued photo ID, the affiliate or member method and reliance, which makes database checks a legitimate Canadian baseline when the sources are strong. The weakness is coverage: young adults, recent arrivals and people with thin credit files fail to match and must fall back to documents. Sanctions and PEP screening is a separate step, covered in KYC screening for iGaming.

Facial age estimation

Facial age estimation uses a model trained on faces to estimate a person's age from a selfie – no identity, no document, no database. It is the lowest-friction method for players clearly above the threshold and the one with the most public accuracy data, because NIST now evaluates it. It estimates, so it needs a buffer above the legal age and a fallback for anyone near the line.

Reusable digital IDs and wallets

A Reusable digital ID lets a person verify once with a wallet provider and then share a yes/no age attribute – "over 19" – with each site without re-uploading documents. It is the least friction of all for repeat verification, but depends on the player holding a wallet and on the regulator accepting the wallet's original verification.

How accurate is each method? What the evidence says

Vendor pages are full of accuracy claims and short on sources. Two sources are worth citing because they publish numbers with methods attached.

NIST FATE: MAE 3.1 years and demographic caveats

In May 2024 NIST published the first results of its age estimation software evaluation, reported in NIST IR 8525. Across the six algorithms evaluated, the Mean absolute error of facial age estimation had fallen from 4.3 years to 3.1 years since NIST's 2014 evaluation. NIST found no single standout algorithm, reported that error rates were almost always higher for female faces, and said it planned to publish updates every four to six weeks as more developers submitted software.

Vendor white papers: Yoti 2.1 years for 18-24

Yoti's July 2025 white paper – the vendor's own figures, not an independent test – reports a mean absolute error of 1.1 years for 13-17-year-olds and 2.1 years for 18-24-year-olds, and states that 99.3% of 13-17-year-olds were classified as under 21. That last figure is the useful one for gambling: it describes how well a buffer of a few years keeps minors out, which is exactly the design question in the threshold section below. Treat vendor numbers as a claim to test in your own pilot, not as a specification.

Why document checks fail differently

Document verification has no mean absolute error; it either extracts the right date of birth or it does not. Its failures are capture failures (glare, blur, cropped edges), unsupported document types, and fraud – forged, borrowed or synthetic documents. That is why the selfie match and liveness step exist, and why a document method with weak liveness may be less safe against minors than a well-buffered estimation check.

Friction and conversion: the trade-off nobody quantifies neutrally

Vendor pages commonly say their method raises sign-up conversion. Jumio's gaming industry page claims that its "course correction" during capture lifts conversion by 15% to 20% and that it serves nine of the ten largest operators in Europe. Neither claim is independently measured.

Friction is real, measurable, and yours to measure. The number to track is first-attempt pass rate: the share of players who complete verification on the first try without a fallback or manual review. Estimation and database methods raise it for the majority; document fallback lowers it for the minority who need it. The pilot design described on our KYC providers page – same traffic split, same period, same definition of "pass" – is the right way to compare method mixes before committing.

What regulators accept and when verification must happen

Method choice is bounded by two rules per market: the minimum age and the moment by which verification must be complete. Where we could not verify a rule, the table says so.

Ontario: AGCO 19+ and validation before account creation

The Registrar's Standards for Internet Gaming, published by the Alcohol and Gaming Commission of Ontario and updated May 14, 2026, set the minimum at 19: Standard 3.01 makes individuals under 19 not eligible to play, with an exception for those 18 and over to purchase lottery tickets only. Standard 3.04 requires name, date of birth, address, identification method, contact information and the information required under the Proceeds of Crime (Money Laundering) and Terrorist Financing Act to be complete, accurate and validated before a player account is created. Standard 3.12 requires players to be authenticated before accessing their accounts, and Standard 1.09 requires compliance information, including logs, to be retained for a minimum of three years. For method selection the key phrase is "identification method": whatever you use must be recorded, and validated before the account exists.

UK: verify before permitted to gamble

Licence condition 17.1.1 of the UK Gambling Commission's Licence Conditions and Codes of Practice requires remote licensees to obtain and verify the customer's name, address and date of birth before that customer is permitted to gamble, and says customers should be told what documents will be needed before deposits are made. According to Experian's May 2024 guide, this dates from the 2019 changes that removed the earlier 72-hour window; the same guide, citing UK Gambling Commission survey data, reports that 26% of 11-17-year-olds surveyed had gambled their own money in the previous twelve months and that operator fines surpassed GBP 200 million in 2023.

Canada AML: FINTRAC identity methods

Ontario's Standard 3.04 points at the federal statute, and FINTRAC is the federal body that says what "verifying identity" means. Guide 11 lists the five identity methods named above and requires photo ID to be authentic, valid and current. Facial age estimation is not one of them; it can gate access, but cannot on its own satisfy AML identity verification, which is why estimation-first designs still need a document or database step before the account is fully opened.

Quebec and other provinces

Quebec's minimum gambling age and Loto-Québec's verification rules: DATA UNAVAILABLE at the time of writing. Check the provincial regulator directly rather than assuming Ontario's 19 applies nationally.

Designing thresholds and buffers for facial age estimation

Because estimation has an error distribution, the pass threshold is never set at the legal age. You set it at the legal age plus a buffer wide enough that a minor is very unlikely to be estimated above it, and you send everyone between the legal age and the threshold to a document or database fallback.

Worked example for a 19+ market

The table is illustrative – it applies published error figures to Ontario's age of 19 to show the logic, not to prescribe a number. Your vendor's error at the ages that matter, measured in your pilot, sets the real value.

Yoti's own figure that 99.3% of 13-17-year-olds were classified as under 21 is the kind of number you want from any vendor: not "how accurate are you on average" but "how often does a minor land above my threshold". Ask for it split by demographic.

When to fall back to documents

Route to document verification when the estimated age is below the threshold, when liveness fails, when the player's typed date of birth conflicts with the estimate by more than the buffer, and – for AML purposes – before the account is fully opened, because estimation does not verify identity. Design the fallback screen to explain why, so the near-threshold adult does not abandon.

Combining methods: patterns that pass audits

Estimation-first with document fallback

Facial age estimation at registration; clearly-adult players proceed, near-threshold players go to document plus selfie. The identity step (document or database) still runs before the account is opened, to satisfy AGCO 3.04 and FINTRAC.

Database-first for returning players

Where bureau coverage is strong, a database match on typed details verifies identity and age with no camera; thin-file players fall back to documents. This suits markets where the regulator recognises the credit file or dual-process method, as FINTRAC does.

Reusable ID for repeat verification

For multi-brand operators, a reusable digital ID lets a player verified on one brand share an "over 19" attribute with another; keep the wallet's evidence reference in your records so you can show the original verification met the regulator's standard.

Questions to ask age verification vendors

• Which of the four families do you offer natively, and which through partners?
• For facial age estimation: what is your mean absolute error at the ages just below and above our legal age, and how does it differ by sex and other demographics? Is your algorithm in the NIST evaluation?
• Which FINTRAC identity methods does your database check satisfy, and what are your sources in Canada?
• What evidence is returned for each method, and can we retain it for at least three years to meet AGCO Standard 1.09?
• What first-attempt pass rate did you achieve in a comparable operator's pilot, and will you run one with us on our definition of "pass"?

Feed the answers into the RFP checklist and pilot design on our KYC providers for iGaming guide.

Age verification exists to protect minors and self-excluded players. Nothing on this page should be read as guidance on loosening or evading it. Players must be 19 or older in Ontario and 18 or older in most other regulated markets; if gambling is a problem for you or someone you know, contact your local support service.

18+ only. Gambling can be addictive — please play responsibly and only bet what you can afford to lose. If gambling is affecting you or someone you know, contact a local support service. This content is informational and never a guarantee of winnings.

Written and reviewed by the iGaming Expert Hub editorial team. Facts checked against primary sources; see the reference above.

← All articles