The Partnership That Worked May Tell You Nothing About the Next One

The first partnership can fool you

Marqeta built its business on Block. By the end of 2022 Block's Square products accounted for 74 percent of Marqeta's revenue and roughly half its gross profit. The relationship covered the Cash App card, the Square seller card, Square Canada, and the card services behind Afterpay. When founder Jason Gardner handed the CEO role to Simon Khalaf in January 2023, the first task named publicly was renegotiating that contract. It renewed through 2028.

By any reasonable measure this was an excellent partnership. It was also unrepeatable. There was no second company with four consumer-facing product lines of that scale needing issuing infrastructure at that moment. Marqeta spent the following years attempting to reduce its dependence on Block and did not find a replacement of comparable size. Reading that as a failure of business development misses what actually happened. The counterparty did not exist.

23andMe has the same shape with a sharper edge. The GSK collaboration signed in July 2018 with a $300 million equity investment and a five-year exclusive target discovery period. GSK extended it in January 2022 with another $50 million. The partnership was productive by its own terms, generating roughly 50 programs and advancing an immuno-oncology antibody into Phase I.

Then it ended, and what replaced it was smaller. In October 2023 the parties signed a one-year non-exclusive data license for $20 million upfront, under which any program GSK initiated it would own outright, with 23andMe eligible only for downstream royalties. Seventy-one therapeutics staff were laid off that August. The company's own 10-Q stated that substantially all research services revenue came from the GSK agreement and that the ability to enter new collaborations would determine future research services revenue.

No second pharmaceutical partner of comparable scale appeared. Two structural reasons are visible in the agreement itself. Exclusivity contractually barred a second partner for five years. And once a buyer has run exclusive discovery across a genotype database for that long, the marginal value of the same database to the next buyer is lower than it was to the first.

Both cases point at the same thing. Economic importance and repeatability are separate properties, and they can move in opposite directions. A relationship can be enormously valuable precisely because it is unusually specific to one counterparty, and that specificity is what closes the door behind it.

What repeats is not the relationship

The useful unit is not the partnership. It is the commercial object the partnership sold.

Some deals produce an object the next counterparty can buy substantially unchanged. Others produce something so tailored that the next counterparty would need different economics, a different integration, and a different strategic rationale, which is another way of saying a new deal.

GoodRx is the cleanest case of the first kind. The company's founders built its pharmacy and PBM relationships from 2011 and both stepped back from the co-CEO structure in 2023. Two CEO generations later, in August 2025, GoodRx announced a collaboration with Novo Nordisk making Ozempic and Wegovy available to self-paying patients at $499 a month across more than 70,000 retail pharmacies. It was the first time Ozempic had been offered at that self-pay price.

What matters is what came next. By the fourth quarter of 2025 the rebranded Pharma Direct business held roughly 200 manufacturer partnerships and more than 100 brand self-pay programs, with named counterparties including Eli Lilly, covering Foundayo from $149 a month and the Zepbound KwikPen from $299, and Viatris, covering 17 established brands. Pharma Direct revenue reached $151.4 million in 2025, up about 41 percent, with management guiding to at least 30 percent growth in 2026.

Novo was not exceptional. It was representative. Every one of those deals is the same construction: a transparent self-pay cash price for a branded drug, available across the same pharmacy network. The second manufacturer invents almost nothing.

Peloton reached the same place from a different direction. The company sold exclusively direct to consumers until August 2022, when it began selling on Amazon. Dick's Sporting Goods followed in September 2022, the first brick-and-mortar retailer to carry Peloton equipment outside Peloton's own stores. Hyatt and Lululemon came in 2023, and Costco in more than 300 stores from November 2024.

Five partnerships in roughly two years, and the commercial form barely changed across them. Place an existing product into someone else's footprint. It is worth noting who announced them. The Dick's release carried a quote from Jen Parker, senior vice president of global direct sales. The Costco announcement came from Dion Camp Sanders, chief emerging business officer. No head of partnerships appears anywhere.

Whatever their commercial teams contributed, both companies had something more basic: a commercial object that could be sold twice.

The enterprise-software test

If formal partnership organizations institutionalize origination, enterprise software is where it should be visible. These companies run alliances teams, certification programs, deal registration systems, marketplace listings, and co-sell motions with dedicated headcount and executive ownership. We worked five of them specifically to test this.

The result was consistent enough to be worth stating plainly. Across all five, the consequential relationships were visibly originated or sponsored at senior executive level, while the formal alliances organization was absent from the observable origination record.

Snowflake expanded its Microsoft partnership at Snowflake Summit in June 2023, quoted by chief revenue officer Chris Degnan and by Judson Althoff, then executive vice president and chief commercial officer at Microsoft. The NVIDIA partnership was announced at the same conference, with CEO Frank Slootman joined on stage by Jensen Huang.

CrowdStrike's NVIDIA collaborations carry George Kurtz's name, with Huang appearing at Fal.Con. Its Microsoft Marketplace availability in February 2026 was announced jointly by Kurtz and Althoff.

ServiceNow was selected as the strongest challenge, because alliances sit unusually deep in its operating model. It turned out to be the weakest. Every NVIDIA announcement from May 2023 through May 2026 is quoted by chairman and CEO Bill McDermott, with Huang on the keynote stage. It is also the same counterparty five times. That is one bilateral relationship expanded repeatedly, which is a different thing from repeat origination, and the alliances function is not visible in any of it.

C3.ai is the most explicit. Its Microsoft strategic alliance, signed 30 September 2024 for an initial five-and-a-half-year term, was described in contemporaneous trade coverage as having been put together between Tom Siebel and Judson Althoff directly.

Nutanix is the most useful, because the company's own filing draws the line for us. The FY2024 10-K describes the Elevate Partner Program as simplifying engagement for the partner ecosystem through a consistent set of tools, resources, and marketing platforms, covering distributors, resellers, managed service providers, telcos, and systems integrators. Nutanix's top two distributors accounted for 47 to 48 percent of total revenue. The FY24 program refresh offered partner sellers a 2 percent incentive up to $7,500 per registered deal. Then, in a separate sentence, the filing lists strategic technology partnerships with Citrix, Intel, NVIDIA, and Red Hat.

The Dell agreement announced in May 2024 was framed at Dell Technologies World by Arthur Lewis, president of Dell's Infrastructure Solutions Group, who positioned it alongside Dell's existing arrangements with VMware, Red Hat, and Microsoft.

None of this makes partner programs decorative. They perform an enormous amount of real commercial work: recruiting, certifying, enabling, registering deals, validating integrations, co-selling, and activating relationships after they exist. C3.ai closed 193 agreements through its partner network in FY25, up 68 percent, representing 73 percent of its agreements. That is genuine distribution.

The claim is narrower. Across twelve cases we did not find observable evidence of an alliances organization originating a consequential strategic relationship. That is a finding about what this evidence shows, not a proof that it never happens.

Why counterparty count was the wrong answer

For part of this research, a different explanation looked convincing. Maybe repeatability is about how many plausible counterparties exist. Marqeta had one Block. 23andMe had one GSK. GoodRx has roughly 200 manufacturers. The pattern sorted most of the early cases.

The enterprise-software work broke it. Snowflake, CrowdStrike, Nutanix, and C3.ai all repeat-originate against counterparty universes of roughly three to six. There are only so many hyperscalers, and only so many server OEMs. A small field did not stop any of them.

Running the comparison the other way finishes it. Across the case set, presence of a repeatable commercial object separates outcomes six times out of seven. Universe size separates outcomes about as often as a coin.

The abandoned hypothesis is worth keeping visible, because it clarifies what universe size actually contributes. More possible counterparties create more opportunity. They do not create repeatability. Something still has to make the same object usable by the next buyer, and if that thing is missing, a large field produces nothing.

Which brings us to the case that does not fit.

Beyond Meat

In early 2021 Beyond Meat announced a three-year global strategic agreement with McDonald's as preferred patty supplier for the McPlant, and a global strategic partnership with Yum! Brands spanning KFC, Pizza Hut, and Taco Bell. Founder and CEO Ethan Brown was in the seat before, during, and after.

No comparable US foodservice partnership followed. International foodservice revenue fell 25.8 percent year over year in the second quarter of 2025 to $15.1 million, which the company attributed to lower burger sales to certain QSR customers. First-quarter 2026 revenue fell 15.3 percent to $58.2 million, attributed to a combination of QSR losses and weak category demand.

The obvious explanation is that the category collapsed and no counterparty was available. The competitor control does not support that. Impossible Foods originated an IHOP partnership in September 2023 and held Burger King, Starbucks, Disney, Wendy's, and White Castle, several of them running four years or longer. Trade coverage in the same period described Beyond Meat's US fast-food record as spotty by comparison, while noting that Beyond Meat's European McDonald's business went nationwide in the UK and Germany.

So the counterparty universe was open. Chains were still signing. The commercial object, a plant-based patty on a chain menu, is about as repeatable as objects get, and Beyond Meat proved it was repeatable by executing it in Europe.

What the exception does establish is the boundary of the finding. A repeatable object creates the possibility of repetition. It does not create demand. Something still has to make the next counterparty want to buy it from you specifically, and Beyond Meat is what it looks like when that condition fails while everything else holds.

Partnership announcements are not evidence the business is working

This is the most portable finding in the study and it runs across industries.

GoodRx originated roughly 200 manufacturer partnerships while total revenue went flat at $796.9 million in 2025, up 1 percent, with 2026 guided down to between $750 and $780 million. Prescription transaction revenue, the core, declined 6 percent.

Peloton originated five distribution partnerships while the stock fell roughly 90 percent under the CEO who signed them. He left in May 2024 alongside a 15 percent workforce reduction.

Impossible Foods held and extended restaurant relationships while its then-CEO said publicly that the category was not where it needed to be, and while secondary-market pricing implied a valuation around $427 million by mid-2026.

C3.ai is the sharpest version. Alliances with Microsoft, AWS, Google Cloud, and McKinsey QuantumBlack. Partner-network agreements up 68 percent in FY25. Ninety percent of revenue running through alliances in the first quarter of FY26. Then revenue down 46 percent to $53.3 million and a 26 percent workforce reduction.

None of this means partnerships do not matter. GoodRx's Pharma Direct is a real and growing revenue line. CrowdStrike is originating repeatedly while growing 26 percent. The point is diagnostic. A sequence of partnership announcements tells you that a company can produce partnerships. It does not tell you whether the business underneath is improving, and in four of our twelve cases it was moving in the opposite direction at the same time.

A revised taxonomy

PastBehavior 013 currently distinguishes a recruited network from a channel. This research suggests a recommendation, not yet doctrine: the distinguishing dimension should be whether the commercial object is reusable, and if it is, whether entry into it has been standardized.

The version worth flagging is that the old binary treats Marqeta and Block identically to Nutanix and Dell. Both are deeply integrated bilateral relationships that are not channels. They behave nothing alike, and the difference sits in the object. Marqeta's was built to the specifications of one counterparty. Nutanix's was built to be installed by whichever OEM wanted it.

Counterparty count is an observable characteristic of these states and is often correlated with them. It is not what defines them, and the research is explicit that it failed as a predictor on its own.

One case fits badly. ServiceNow's NVIDIA relationship is a single counterparty expanded five times, and whether its object is reusable was never tested against a second buyer. That misfit belongs in the record.

The diagnostic

When assessing a partnership strategy, the first question is not how many partners exist or who owns the function. It is: what exactly did the last partner buy, and can the next partner buy substantially the same thing?

Then test it against the last two relationships:

If the answers change with every partner, the company has a collection of partnerships. That may be perfectly good business. It is not a repeatable motion, and it should not be forecast as one.