Roiva Verified AI Rating
The Roiva Verified AI Rating
An opt-in rating that verifies a company's own evidence of AI in use, Responsible AI and realized value: a maturity level from 1 to 5, private unless published.
Separate from the AI in Practice Index. The rating is opt-in. A rating never changes an index score, a leader decision or a profile, and what we learn from a submission is never used in index research.
The Roiva Verified AI Rating is for companies that want their AI work assessed on more than what they've said in public. A company sends us evidence it holds: documents that show an AI use running or a Responsible AI practice operating, and figures for the value an AI use has returned, which the top level needs. A reviewer verifies it, and we place the company on a maturity level from 1 to 5.
The rating is opt-in, and private unless the company chooses to publish it. It is separate from the AI in Practice Index, which is built only from public evidence. Nothing a company sends us for the rating ever counts toward the index, in any edition.
How it differs from the index
| AI in Practice Index | Verified AI Rating | |
|---|---|---|
| Who is in it | every company in an edition's population | companies that apply |
| Evidence | public sources only | public sources, plus the company's own evidence once we've verified it |
| Consent | none needed: leaders are named from their public statements | needed for anything we publish |
| Result | AI in Practice Leaders, named alphabetically | a maturity level and, from Level 3, a tier if its score clears one and it shows at least 1 point of Governance, Responsible AI and people |
| How long it lasts | dated, never expires | valid for 12 months |
| What it costs | nothing | $2,500, $7,500 or $15,000 a rating, set only by the company's revenue |
The two badges are separate marks and never combined. Holding one says nothing about the other.
The five levels
Each level builds on the one below. A company sits at the highest level whose requirements, and all lower ones, it meets. Its scorecard says "Level N of 5" and lists what the next level needs.
| Level | Name | What it takes |
|---|---|---|
| 1 of 5 | Exploring | at least one named AI use, and an AI policy, a named person accountable for AI, or board oversight of AI |
| 2 of 5 | Live | at least one AI use running in day-to-day operations |
| 3 of 5 | Scaling | at least three distinct AI uses running, and at least one handling at least half the eligible work of its process across a whole business unit, or running in two or more units |
| 4 of 5 | Governed | an AI policy and a Responsible AI policy; board oversight of AI or a C-suite AI leader; four core Responsible AI practices in operation (an AI inventory, risk assessment before deployment, monitoring with incident handling, and oversight of third-party AI); AI literacy and training for staff; records of each of these practices operating over the 6 months before the rating; the AI inventory checked against the AI products and services the company's books show it paid for; and an actual measured result for at least two running uses |
| 5 of 5 | Value realized | at least two running uses, each with verified realized value greater than its full cost, and a combined annualized net value of at least 0.1% of the revenue of the business units they serve |
A company that meets none of Level 1's requirements isn't rated.
Levels 1 to 4 rest on documents, and Level 5 on figures. Every requirement up to Level 4 is met by documents a reviewer reads, or by public evidence. Level 4 also needs a history of its practices operating, and the AI inventory checked against the company's books (below). Level 5 is met only by realized-value figures that pass the full check below.
What we ask for
For each criterion, the rating's questionnaire asks for the information the criterion requires. For an AI policy, for example: its title and owner, the body that approved it and when, that it was in force on the rating date, and that it names AI. Any document from the company's own records that shows that information counts, if it is dated and current on the rating date. Records of a practice operating may be older, and Level 4 asks for them to go back six months. The documents we name are only examples. Our evidence guidelines list the information required for every criterion.
- Documents, for every criterion except realized value. A reviewer reads each document whole and checks that it shows the information required. Nothing is recomputed, and we don't take screen-shares or live demonstrations. A Responsible AI practice also needs records of it operating, not only a policy describing it. A measured result for Level 4 is read from its report, with its method and period.
- Two more checks at Level 4. Each practice Level 4 requires needs records of it operating across the six months before the rating date: at least one dated six or more months before it, and at least one dated within the three months before it. For risk assessment, each assessment is dated before its system went live, and at least one is dated six or more months before the rating date. And the company sends a list, from its accounts payable or general ledger, of the AI products and services it paid for in the 12 months before the rating date, signed off by its finance officer. A reviewer checks that every one appears in the AI inventory and in the company's oversight of third-party AI. A practice that misses either check still counts toward the rating score, but doesn't meet Level 4 until the company sends the records or updates its inventory and oversight. A practice shown by public evidence can meet its Level 4 requirement without the six months of records, but the AI inventory and the oversight of third-party AI still need the check against the books.
- Figures, for realized value. Each figure takes the full check set out under realized value below. Only a company that wants Level 5, or wants realized value counted in its rating score, needs to send figures.
Realized value
Realized value is a result an AI use has already produced, in cash or in hours, measured against a baseline. For Level 5 it is counted net of the use's full cost: licenses and vendor fees, computing, staff time, a share of shared platform costs, and build, integration, implementation and change costs spread over three years (or under the company's own accounting policy).
A figure is verified when: - the company shows its method, how it attributes the result to AI, and its baseline, measured before the use went live or from a group that never used it; - it shows at least six months of actual results, ending within the 12 months before the rating; - a reviewer with accounting competence recomputes the figure from an extract of the company's own source-system data, to within 10%, and checks its costs against invoices or the general ledger; - a finance officer (the CFO, the controller or the head of internal audit) signs that the figure, its costs and the data are accurate; - for Level 5 and the tier, the company states the revenue of the business units the uses serve, from its segment reporting or management accounts, and the sign-off covers it.
A figure whose value is verified but whose costs aren't counts as gross value: it can meet Level 4, but not Level 5 or the realized-value part of the rating score.
For every verified figure in hours, the company also tells us where the released time or roles went: redeployed to named work, absorbed growth without new hires, reduced through attrition, or roles eliminated. A reviewer checks it against the company's own headcount and redeployment records, and the finance officer's sign-off covers it. With the company's consent it is shown beside the hours figure. It is shown, never scored: no answer earns more than another.
A figure measured in Roiva's own software also needs backing from outside Roiva, and the scorecard says Roiva's software was used.
A verified figure is still the company's own measurement: we check its backing, we don't audit it. We never compare figures between companies, and never present them as what another company should expect. A company that hasn't shown us a figure is shown as "not disclosed", never as zero.
What "verified" means
"Verified" has two meanings in the rating, one for figures and one for documents. Every scorecard carries both: the draft and final scorecards the company always receives, and the published scorecard if it chooses to publish one.
- For a realized-value figure: "Verified means Roiva's reviewer recomputed the figure, to within 10%, from source-system data the company provided; checked its costs against the company's invoices or ledger; and received a signed confirmation from the company's finance officer. It is the company's measurement. Roiva didn't audit or examine the company's financial statements or controls, didn't test its source data, and doesn't establish that AI caused the result."
- For every other item: "For every other item, 'verified' means a Roiva reviewer read the company's documents and found that they show the information the method requires. Roiva didn't test the company's controls or systems, or confirm the documents beyond reading them."
For documents, we also rely on the accountable executive's attestation that the evidence is accurate and current.
Responsible AI and people
The scorecard shows the company's AI governance (board oversight of AI, an AI leader, an AI policy and a Responsible AI policy) and eleven practices, each marked verified, public or not shown. "Not shown" means not shown to us, not absent. The practices are:
- an AI inventory;
- risk and impact assessment before deployment;
- testing and validation, including bias testing;
- monitoring in production;
- incident handling;
- oversight of third-party AI;
- human oversight of consequential decisions, with review or appeal;
- data governance;
- AI literacy and training;
- transparency to the people AI decisions affect;
- workforce transition and staff voice: what happened to the roles AI changed, and how staff were informed or consulted before AI went into their work.
Level 4 requires the four core practices (1, 2, 4 with 5, and 6) and AI literacy and training (9). The others count toward the rating score but don't decide a level. Each practice needs records of it operating, not only a policy describing it.
Next to each criterion the scorecard names where the criterion appears in widely used frameworks: the NIST AI Risk Management Framework, the NAIC Model Bulletin on insurers' use of AI, and the EU AI Act. We'll add ISO/IEC 42001 only once we've checked our references against the standard itself.
The rating doesn't assess conformity with any standard or compliance with any law, and isn't a review of an insurer's AIS Program under any state's bulletin. It gives no score against any framework.
Tiers
A company at Level 3 or above can also get a tier: Platinum, Gold, Silver or Bronze. A tier needs all three of: - a rating score that clears that tier; - a minimum level: Level 5 for Platinum, Level 4 for Gold, Level 3 for Silver and Bronze; - at least 1 point of Governance, Responsible AI and people, which is more than an AI policy on its own.
Those points are counted from what the company shows, verified or public:
| Shown | Points |
|---|---|
| board oversight of AI | 1 |
| a C-suite AI leader, or another named AI lead | 1, or 0.5 |
| an AI policy | 0.5 |
| a Responsible AI policy | 1, and it also earns the AI policy's 0.5 |
| one or more of the eleven practices above, or a public source showing Responsible AI practices that we haven't tied to one of them | 1 in all |
An AI policy alone, or a named AI lead below the C-suite alone, earns 0.5, so a company with nothing more gets no tier. Gold and Platinum already need far more through Level 4's requirements, so in practice this decides Silver and Bronze. A company below 1 point keeps its level, and at Level 3 or above may use the badge with its level alone (below).
The rating score combines breadth (distinct AI uses), depth (running uses), measurement, realized value (verified net value as a share of the revenue of the business units served, in bands), Governance, Responsible AI and people, and recency, from public and verified evidence. Its weights, the tier bands and the tier rules are fixed in version 2 of the rating method before the first rating is issued. The rating score itself is never published.
- At first, a tier is a fixed score band in the rating method, set before the first rating and the same for every company. "Gold" means a rating score at or above the Gold band, with the level and the governance points the tier needs. It isn't a percentile, or a comparison with other companies.
- Once an industry has 100 rated companies, a new version of the rating method, announced in advance, moves its new ratings to percentiles: roughly the top 1%, 5%, 15% and 35% of rated companies in that industry, from a snapshot taken each quarter. Only then may a tier be described as a percentile, and always with the pool's size and quarter, for example "top 5% of 112 insurers rated by Roiva, Q3 2027".
- A tier lasts as long as the rating, 12 months, even if its industry moves to percentiles.
Levels 1 and 2 get no tier and no badge.
How a rating is decided
- The company applies, signs the rating agreement and pays the fee. It names an accountable executive, who confirms the evidence is accurate, current and the company's to share, and, if it claims Level 4 or above, a finance officer for the sign-offs. The review starts once the fee is paid.
- The company answers the questionnaire, criterion by criterion, with the documents that show each one, and figures for realized value if it wants Level 5 or wants realized value counted in its rating score. See what to send and how we handle it.
- AI agents may do a first pass (Claude, made by Anthropic), such as matching each item to what it shows and checking dates, under the direction of Roiva staff. They never decide whether an item counts. A company can ask for a review by people only, at any time before its review starts, at no extra cost and with no effect on the result.
- A reviewer decides every item and the rating: Roiva staff or a contractor, with no commercial interest in the company. If the company is a Roiva customer, prospect, partner, supplier or investor, that reviewer holds no stake in Roiva and earns no commission. The rating fee alone isn't such an interest: it is set only by the company's revenue band, is the same for every result, and is paid before the review starts. Realized value, and so Level 5, is decided by a reviewer with accounting competence. Each reviewer declares any conflict before reviewing, and a reviewer with one doesn't review that company. Where more than one reviewer records an item and they disagree, the item doesn't count and the lower level stands.
- The company sees a draft scorecard first, and has 30 days to correct facts or add evidence. If it disagrees with the decision, it can ask us to look again, and we give our reasons in writing.
- A rating is valid for 12 months. Renewing it is a new review under the method current at the time, at the published fee for the company's revenue band then.
What's published, and what isn't
Nothing about a company's rating is published without its written consent: not that it applied, was rated or withdrew, and not its level, tier or scorecard. A lower level at renewal is never published either. The one exception: once a rating a company published or announced is no longer valid, because it expired, was withdrawn or its scorecard was unpublished, the page its badge links to shows only that, with the date. If the company keeps showing the badge or announcing the rating, we may say the same to anyone who asks, or publicly. We never publish our reasons for withdrawing a rating.
A company that chooses to publish gets a scorecard page on roiva.ai showing: - its level, tier, rating date and "valid until" date; - its AI uses, with their status and whether each is backed by public or verified evidence; - its governance and Responsible AI practices, with the framework references above, the version of each framework, and the statement that the rating isn't a conformity or compliance assessment; - how many uses have verified realized value. Figures appear only where the company agrees to each one; - its commercial relationship with Roiva, if any (customer, partner, supplier or investor), and whether Roiva's software produced any verified figure; - that the company paid Roiva's published fee; - the line "Includes company-provided evidence verified by Roiva", and both meanings of "verified" above.
A published scorecard never shows the rating score or names other companies. The only comparison it shows is a percentile tier, once its industry has moved to percentiles, with the pool's size and quarter.
The badge
A company at Level 3, 4 or 5 with a valid rating and a published scorecard may use the badge, which links to that scorecard. Its wording is fixed, for example:
Roiva Verified AI Rating: Gold, 2026 Level 4 of 5 · valid to November 2027 Includes company-provided evidence verified by Roiva
A company at Level 3 or above with no tier, because its score clears no tier it has the level for or it has under 1 point of Governance, Responsible AI and people, uses the same badge with its level alone, for example:
Roiva Verified AI Rating: Level 3 of 5, Scaling, 2026 Valid to November 2027 Includes company-provided evidence verified by Roiva
The badge may not be shortened or changed, combined with the AI in Practice Leader badge, or used to compare the company with named companies. It doesn't mean Roiva certifies, audits or endorses the company, its AI or its products. Nor does it show the company complies with any standard or law, and a company may not cite the rating that way, including in a regulatory filing or exam. When the rating expires or the scorecard is unpublished, the company updates or removes its references to the rating within 30 days; after a withdrawal, at once. It doesn't cite the rating in any new filing, regulatory exam or announcement.
Independence
- Nothing bought counts. A rating never depends on whether a company buys, uses or discusses any Roiva product. Roiva customers are held to the same standard and process as everyone else, and pay the same fee as any company in their revenue band, with no discount, priority or separate route.
- The fee, and nothing more to buy. Nobody can pay for a level, a tier, a faster review or for a figure to count. No one's pay, sales target or commission depends on any rating.
- No advisory work. We don't sell advisory or implementation services to a company from the day it accepts the rating agreement until its rating ends, and we don't rate a company we sold them to in the 12 months before it applied.
- The index stays public-only. A rating never changes an index score, a leader decision or a profile, and what we learn from a submission is never used in index research.
- No trading. Roiva, our staff, our reviewers and their households don't buy, sell or short a company's securities, or trade options on them, from receiving its evidence until 30 days after the decision, and after that for as long as we know we hold material non-public information about it. That holds whether or not the company ends up rated.
Our independence policy covers the rest.
What it costs
The fee depends only on the company's annual revenue:
| Annual revenue | Fee per rating |
|---|---|
| under $1 billion | $2,500 |
| $1 billion to $10 billion | $7,500 |
| over $10 billion | $15,000 |
Revenue is the total revenue of the company and its subsidiaries for its most recent completed fiscal year, as reported in its financial statements with an independent auditor's report, or, without them, as its accountable executive attests. For a bank, it is net interest income plus noninterest income. The band is set when the company signs the rating agreement.
Every company in a band pays the same fee, whatever level it claims and whatever the result. The fee never depends on the level, the result or any relationship with Roiva. A renewal is a new rating, at the published fee for the company's band when the renewal is signed, with the band set again from its revenue then.
- It is due when the company signs the rating agreement. The review starts once it is paid, so no result is known when a company pays.
- It isn't refunded for a lower level, no tier or a "not rated" result, nor if the company withdraws its submission, its submission lapses, or it ends the rating agreement.
- It is refunded in full if we stop offering the rating while the company's review is in progress.
- It buys an assessment under the method, never a level, a tier, a faster review or a figure counting.
- Every scorecard says that the company paid Roiva's published fee.
- We announce any change to the fee at least six months ahead. A change never applies to a rating already applied for.
The AI in Practice Index stays free: any company in an edition's study can ask for its own score and the sources behind it, and claim, correct or remove its profile, at no charge.
How to apply
Write to data@roiva.ai. We'll send the rating agreement for electronic signature, the questionnaire and a secure upload link. The fee is due on signing. Please don't send documents by email.