How we rank CDMOs
Last updated August 4, 2026 · Evidence Score formula v1.0
Our ranked lists exist because the alternatives rank by revenue or publish no method at all. Revenue tells you who is biggest, not who is right for your programme. This page is the whole method: every input, every gate, every tie-break, and everything we refuse to measure. If you disagree with a rank, you should be able to work out exactly why it came out that way.
The Evidence Score
Three weighted dimensions, every input drawn from a public FDA dataset.
| Dimension | Weight | What it measures |
|---|---|---|
| Regulatory evidence | 45% | What the FDA found when it inspected, how much inspection history exists, and how current it is. |
| Approved-product record | 30% | Products on the FDA record that this manufacturer is associated with. |
| Modality-match depth | 25% | FDA-proven operations, dosage forms, and routes relevant to the modality the list covers. |
| Total | 100% |
Missing evidence is withheld, never scored as zero
A dimension with no evidence is withheld and the remaining weights renormalize — we never score an absence as a zero. Scoring a zero would publish a damning finding about a named manufacturer that our data does not support. Awarding full credit for an unchecked record would be the same error pointing the other way.
When we hold no substantive evidence about a company at all, it has no Evidence Score. It is not yet assessed, which is a different statement from scoring badly, and the page says so.
On any ranked list, a dimension we cannot evidence for a company shows the words “No data yet” in that company’s row. There are no blank cells and no dashes, because a blank reads as a zero.
Every input and where it comes from
Regulatory evidence (45%)
Verified inspection records and their NAI / VAI / OAI classifications, the most recent inspection date, warning letters and enforcement actions (only when enforcement was actually checked).
Source: FDA Inspections Classification Database and FDA Enforcement Reports, matched to each company by FEI facility number.
Approved-product record (30%)
Count of approved products associated with the manufacturer, across NDA/ANDA, listed NDC products, and licensed biologics.
Source: FDA Orange Book, the FDA National Drug Code Directory, and the FDA Purple Book.
Modality-match depth (25%)
FDA-registered establishment operations, plus dosage forms and routes of administration proven on the FDA record, measured against the modality this list covers.
Source: FDA Drug Establishment Registration and FDA-listed product records.
How inspection outcomes are credited
The FDA classifies an inspection when it closes. We credit those classifications as follows, and we report the split rather than a verdict:
- NAI — No Action Indicated (100/100). The inspection closed without objectionable conditions being cited.
- VAI — Voluntary Action Indicated (55/100). Objectionable conditions were found but did not warrant regulatory action.
- OAI — Official Action Indicated (0/100). Regulatory action was warranted.
We never describe a manufacturer as having a “clean record”. We state the split — for example “10 FDA inspections: 0 NAI, 10 VAI, 0 OAI” — and let you read it. NAI is an FDA classification, not a bill of health.
Who is eligible to be ranked
A company appears on a ranked list only when all of the following hold.
Classification
The company is classified as a CDMO, CMO, or API manufacturer in our data. We read the classification field — we never infer what a company does from its name.
A verifiable FDA record, not just a number
An inspection count on its own is not evidence. We rank only on FDA inspection records we can trace back to a source, which means all three of:
- The profile carries real per-inspection history — individual FDA inspection records, not just a count.
- The profile carries at least one FEI number, the FDA facility identifier that anchors the record to a source.
- The profile carries an ingestion timestamp, proving the record came through the verified FDA ingestion path.
We also require the record’s own count to agree with the number of inspection records it contains. A company whose FDA record we cannot verify this way is excluded rather than estimated — even when that makes a list shorter, and even when it removes a company a reader would expect to see.
Enough companies to be a list
We publish a single-axis ranked list only when at least 10 eligible companies clear the bar, and a narrower cross only at 5. Below that, the page tells you we do not have enough verified evidence, points you to the relevant browse page, and is excluded from search indexing. We publish no partial or padded list.
Matching and scoring are separate — deliberately
Deciding whether a company belongs on a list and deciding where it ranks use different data, and it matters which is which.
- Matching may use self-reported text. To decide whether a company works in a given modality we read its stated capabilities, expertise, and therapeutic areas. That text is self-reported or scraped, so we treat it as a filter only.
- Scoring never uses self-reported text. Once a company is on a list, its position is computed exclusively from FDA-sourced fields. Nothing a company writes about itself can move it up.
We match on strong, unambiguous modality terms and require whole-word matches, so a company is not pulled onto a cell-and-gene list because an unrelated phrase happens to contain a fragment of one.
What we do NOT measure
None of the following is read by the ranking query or by the scorer. They are not weighted at zero — they are not present:
- Featured or promoted placement
- Featured rank
- Partner tier
- Verification status
- Reviews and star ratings
- Profile completeness
- Subscription tier
- Profile views
No seller can pay for or self-report their way into a rank. There is no promoted slot, no sponsored position, and no commercial relationship with us that changes where a company appears on a ranked list.
This is enforced in code, not by policy alone: our test suite flips every one of the fields above and asserts that both the scores and the ordering come back byte-identical.
Tie-breaks
The ranking is deterministic: the same data always produces the same order. There is no randomness and no hidden boost. When two companies tie, we break it in this published order:
- Higher Evidence Score first; an unscored company ranks last.
- More verified FDA inspections on record.
- More recent inspection; a company with no dated inspection ranks after one that has a date.
- Alphabetically by company name.
Editorial independence, and the limits of this score
Ranked lists are produced by applying the formula on this page to public FDA data. They are not reviewed, negotiated, or adjusted with the companies that appear on them. A company being a customer of CDMO Hub has no bearing on whether it appears or where.
Being ranked is also not an endorsement. The FDA record is one input into a sourcing decision and a backward-looking one. It does not capture technical fit for your molecule, capacity availability, tech-transfer capability, project management, responsiveness, or commercial terms — and it does not replace a site audit or your own technical and quality due diligence. The selection literature is consistent that those factors outweigh the paper record. Use these lists to build a shortlist, then go and look.
If you believe a figure attributed to your company is wrong, tell us and we will re-check it against the source. We correct errors; we do not adjust ranks on request.
Formula version
Ranked lists on this site are computed under Evidence Score formula v1.0. The version is stamped on every score and shown on every ranked page. When the formula changes materially, the version changes with it, and scores computed under a version we no longer stand behind stop being served rather than quietly ageing.