Insights
What counts as a distribution partnership, and what does not
Most partnership counts do not say what they counted. This is what we count, why the boundary sits where it does, and the two places we have already got it wrong.
The word does almost no work
Press releases call almost everything a partnership. A software licence is a partnership. A sponsorship of a stadium is a partnership. A company buying fraud scoring for its own underwriting is a partnership. Read 1420 announcements and the word turns out to describe a purchase about as often as it describes a route to a customer.
That is why a count of partnerships is nearly meaningless unless the counter says what they counted. Ours counts 1091 of 1420 arrangements, and the 329 it does not count are recorded and shown anyway, under their own heading.
One test, applied to everything
Does the capability leave the partner and reach an end customer? If it stays inside the company that acquired it, the arrangement is a purchase. If it leaves, it is distribution, and which kind depends on who the parties are and what moves between them.
What does not decide it: the money direction, because payment runs from buyer to seller in a purchase and in infrastructure distribution alike. Not a revenue share, which looks like partnership economics and changes nothing. Not the size of the contract, since several of the largest arrangements here by stated value are purchases. Not the word partnership, which appears in the headline of most of the purchases we record.
The ten types, and the question each turns on
| Type | The question |
|---|---|
| Embedded finance placement | Is the host a non financial company whose product flow the financial product sits inside? |
| Bank sponsorship | Is a charter the object of the agreement? |
| Infrastructure distribution | Does one financial company's capability reach customers through another financial company? |
| Non financial provider distribution | Same movement, but is the provider outside financial services? |
| Lender distribution agreement | Does loan volume move between the parties? |
| Carrier distribution agreement | Does an appointment exist, and can the counterparty bind? |
| Marketplace placement | Is the product listed alongside competitors on a surface the partner controls? |
| White label | Is the originating company undisclosed to the end customer? |
| Affinity and group distribution | Is the population closed and defined? |
| Referral and lead flow | Is the handoff the whole arrangement? |
And the tenth, which is counted by nobody and is the most consequential of the lot: the technology purchase.
Where we got it wrong
Two types on that list were added after the taxonomy failed on real arrangements, and both failures ran the same direction.
A payroll platform running on a banking core, and a payments company whose card programme runs on an issuing platform. Neither party is a bank, so bank sponsorship did not apply. The partner is a financial company, so embedded placement did not apply. The only type left was the technology purchase, which is the one category never counted. Filing a route to customers as a purchase understates distribution in exactly the population where it is most common, so infrastructure distribution was added.
Then the same shape appeared again with the provider outside the industry: a financial crimes agent built with an AI company and made available to banks, a card network's protocol combined with a security company's bot detection and offered to merchants. Seven arrangements in twelve months of one publication, all declined for want of a type, all filed nowhere. That type now exists too.
A missing type is never a neutral omission. It sends real distribution into the one bucket the dataset does not count, and the resulting number is quietly too low with nothing on the page to say so.
What is still missing
One shape has no type today: a financial company supplying something to a non financial company that distributes it onward to its own customers. A card network's identity data inside a security vendor's platform is the example we hit. What moves is not a financial product, so embedded placement does not fit, and the direction is wrong for the other two. It is declined rather than forced into a type that does not hold it.
That is written down here rather than left as an unexplained absence, because an arrangement we cannot classify is a fact about the taxonomy, not about the market.
Counts on this page come from the dataset on this site and were generated on 2026-08-28. Every row is listed with its source.