Fieldwork No. 029
Nine attempts at building outside distribution for work that has to be built one customer at a time. Six held. Three came apart in the same place.
A service that gets built differently for every customer is difficult to hand to anyone else. The reseller does not know enough about how the customer actually operates to scope the job. Teaching the reseller everything the specialist knows would take years, and if it ever worked, the specialist would no longer have a reason to exist.
Companies in that position have built outside distribution anyway. They did it by cutting the outsider's job down to something very small. The outsider notices that a customer it already serves is in a particular situation, and makes an introduction. Judging whether the buyer is a fit, working out what should be built, pricing it, closing it, and doing the work all stayed with the specialist.
This study covers nine attempts. Six of them held. Three came apart, and the three failures fall into a tight pattern: in each one, the outsider was asked to do slightly more than notice.
The cases run from 2019 to 2026 and cover tax credit work, insurance placement, cybersecurity, utility efficiency programs, cloud migration, and enterprise software. Three of the six that worked are not technology businesses at all. That is deliberate. The mechanism is clearest in industries that have been running it for decades without any of the vocabulary.
The outside party does not need to understand the solution. It only needs to recognize the situation.
When people say partnership they usually mean all of the following bundled together. They can be separated, and the whole argument here depends on separating them.
In the cases that worked, responsibility changed hands after the second item and before the third. The intermediary recognized the situation and opened the door. The specialist took it from there.
In the cases that failed, qualification or diagnosis had drifted into the intermediary's hands. That is the whole finding, and everything below is evidence for it.
Specialty tax credit firms and accountants. A research and development tax credit study is expert work. Someone interviews the engineers, reconstructs which activity qualified, builds documentation that will survive an audit, and defends the claim if it gets challenged. The accountant already holds the payroll, the project accounting, and an annual conversation about tax position, so noticing a candidate costs nothing extra. Most accountants will not do the study, partly out of audit fear. alliantgroup says on its own materials that it has worked with more than 4,000 CPA firms, and reported delivering more than $2.3 billion in credits and incentives to more than 14,000 businesses in 2021. The totals matter less than the operating detail. These firms hand accountants a checklist of things to look for. One of them, Tax Point Advisors, offers to analyze an accounting firm's existing client base and return a list of likely candidates. The specialist performs much of the recognition on the intermediary's behalf. What is left for the accountant is a phone call.
Wholesale insurance brokers. This is the oldest clean version of the model and it is enforced by licensing rather than by a contract. A retail agent runs into a risk the standard market will not write and submits it to a wholesale broker who has access to specialty carriers. The retail agent keeps the client and keeps the servicing obligations. The wholesale broker works through the retail agent and does not go to the insured directly. A 2026 industry summary citing 2025 association benchmarking puts retail compensation at roughly 10 to 12.5 percent of premium, with the wholesaler retaining roughly 2.5 to 5 percent of carrier commission, though that figure is secondary and worth holding loosely. The split is not the interesting part. Client protection is. The retail agent can introduce a specialist without any fear that the specialist will end up owning the account, because the specialist is structurally unable to.
Huntress and managed service providers. Huntress provides managed detection and response for smaller businesses. The end customer cannot run a security operations center. Most of the IT providers serving those customers cannot either, and Huntress says so in its own partner materials, telling MSPs they do not need an in-house team to behave as though they have one. The MSP already manages the customer's technology environment, sees the exposure, and adds the service into a relationship that already exists. The vendor's people do the detection and the response. Official partner pricing is not published. Third-party analysis, unverified by the company, estimates partner cost around $2.50 to $3.50 per endpoint against roughly $8.99 list, with volume tiers counted across the MSP's entire customer base. In 2026 Huntress opened a reseller program alongside the MSP channel and trade commentary immediately raised the concern that MSPs would lose some of the protected position that made the relationship worth effort in the first place. Protection is what buys the effort. It is also the thing vendors eventually want back.
Utility trade ally programs. Utilities want efficiency improvements installed in small commercial buildings, and every building differs enough that nothing can be sold from a catalogue. The contractor already servicing the heating or the lighting is the party most likely to notice. In the Consumers Energy small business program, the contractor conducts the initial on-site assessment at no cost to the customer, and the program document states plainly that the assessment does not obligate the customer to participate in anything. The contractor then explains what could be changed. A comparable Oregon program shows the quality mechanism: incentive payments can be held until randomly selected work passes verification against published technical standards. Look at what flows back toward the contractor. Programs supply inbound customer leads, free analysis of the contractor's own customers, technical support, training, co-marketing, and public directory placement. The contractor does a narrow noticing job inside work it was doing anyway, and the program carries almost everything around it.
The AWS Migration Acceleration Program. Cloud migration is specific to whatever the customer already has running, which means it cannot be scoped without discovery. AWS built a three-phase structure where the first phase, Assess, is a discovery engagement that partner materials commonly describe as four to eight weeks, producing a workload inventory, a cost model, and a readiness analysis. It is delivered by qualified partners rather than by AWS. One of them, Loka, lists the Assess engagement on AWS Marketplace and states it is 75 to 100 percent funded by AWS. AWS is paying an outside firm to do the customer's homework, because the homework is the thing sitting between the vendor and future consumption. The partner earns revenue on the assessment and can then win the migration it just scoped. Microsoft and Google run migration programs with the same shape, which is worth more as corroboration than any single program's results.
Palantir bootcamps. Palantir's deployments depend on engineers working inside the customer, which is about as far from resellable as software gets. In 2023 the company shifted toward short bootcamps where the customer's own people work on a real use case using their own data. On the third quarter 2023 earnings call an executive described these as producing real workflows in five days or less, against traditional pilots that ran one to three months. There were 92 pilot sessions in 2022, then more than 500 bootcamps across more than 465 organizations in 2023. Partners including Accenture ran them. J.D. Power later ran bootcamps for its own clients. The thing a partner could distribute was no longer the full customized deployment. It was a short structured event that let the specialist and the customer find out together whether anything was worth building.
The Employee Retention Credit. This is the tax credit model from above with one thing changed, which is what makes it the most useful case in the study. The intermediary often moved past recognizing the opportunity into determining eligibility, and compensation could depend directly on the size of the claim. The IRS warned that employers remained responsible for improper claims even where promoters had taken contingency fees as high as 25 percent. On September 14, 2023 the agency announced an immediate moratorium on processing new claims. In August 2024 it reported disallowing roughly 28,000 claims worth about $5 billion. By October 2024 the backlog had reached approximately 1.2 million claims. Voluntary disclosure programs opened, including one that let employers repay 85 percent of what they had received, with the missing 15 percent reflecting the fees already paid to promoters. Penn Wharton later projected total program cost around $302 billion. The problem was not that promoters were paid. The person making the eligibility judgment had a direct financial interest in the answer, and recognition had quietly expanded into qualification.
Palantir and IBM. In February 2021 the two companies announced a joint product combining Palantir Foundry with IBM Cloud Pak for Data, Watson, and Red Hat OpenShift. An IBM executive described an ambition to raise AI use among IBM customers from 20 percent to 80 percent and to sell it across 180 countries. What the announcement did not define publicly was what an IBM seller should be looking for, how those sellers would be paid, who owned delivery, or what the handoff actually looked like in practice. No material revenue attributed to the arrangement was ever disclosed and the joint product did not become a durable line of business, which is an inference drawn from the absence of later evidence rather than from any announced shutdown. Set it beside the joint venture Palantir formed with the Japanese insurer Sompo in November 2019, which had $150 million behind it, fifty-fifty ownership, and an anchor customer already running the software. That relationship produced a $22.5 million contract in 2020, a $50 million expansion in 2023, and a further expansion in 2025 with more than 8,000 Sompo users reported. Reach alone does not explain which relationships become distribution.
The early RPA consultancy channel. Automation software vendors put their tools into the hands of the large consultancies, which already held enterprise efficiency mandates and could bill for implementation. Those firms were expected to work out which business processes should be automated, redesign them, scope the work, and deliver it. EY, which had run these projects across 20 countries and was often called in after a failed first attempt, reported seeing 30 to 50 percent of initial projects fail. A Deloitte survey of 400 organizations found 63 percent missed delivery deadlines and only 3 percent scaled beyond fifty bots. Deloitte's own ranking of what blocked them put integration difficulty, missing skills, and inability to change business processes at the top. A substantial part of the failure sat upstream of implementation, where someone had to determine what should actually be automated. In October 2019 UiPath bought ProcessGold and StepShot on the same day, and coverage at the time described the resulting process discovery output as a specification telling the developer what to build. The missing thing had been a reliable diagnosis artifact.
1. Their own delivery gets better. This is the strongest pattern in the whole study. Gusto's materials for accounting firms lead with what the firm gets operationally, including multi-client tools, compliance automation, and a company-reported claim of roughly 200 hours saved annually based on a survey of 538 of its customers. Huntress lets an MSP add security coverage without building a security team. Utility programs help contractors find and complete work they were already qualified to do. The question worth asking about any candidate partner is whether they would still want the relationship if the referral payment disappeared tomorrow.
2. They know the client stays theirs. Wholesale insurance makes this structural, since the specialist is not permitted to approach the insured. Security vendors market the same promise in plain language. When a channel structure starts to make the partner wonder whether the specialist might end up owning the account, willingness to introduce customers drops, and the 2026 reaction to the Huntress reseller program is a live example of that happening.
3. Something useful comes back. Inbound leads, directory placement, analysis of the partner's own client list, technical support. The arrangements that ran for years all send something in the other direction rather than treating the intermediary purely as a source of names.
4. Their own work expands. AWS migration partners earn the assessment and then can win the migration. The RPA consultancies earned implementation hours. This incentive produces behavior faster than any of the others, and it is present in both of the worst failures here, because it drives volume and volume is dangerous once the partner is making judgments beyond its competence.
5. Money. Compensation shows up in every case, the ones that worked and the ones that did not, so it does not distinguish between them. For certain attest clients, accounting ethics rules prohibit CPAs from accepting a referral commission at all. Lawyers generally cannot share fees with non-lawyers. Referrals in both professions happen constantly, on reciprocity and on client service. Meanwhile some of the largest percentage incentives anywhere in this research turned up in the promoter market that produced the ERC collapse. Money can support a reason to act that already exists. It does not reliably create one.
Owner: the outside party
Recognition should run on information the intermediary already sees during ordinary work. Standardize the situation they should notice and stop there. Deep product training tends to backfire. Research on one regional utility contractor network found that first-year training built around program mechanics and efficiency fundamentals had only an indirect relationship with actual projects, and that delivery improved once the message was cut down to a simple proposition about growing the contractor's own business.
Owner: the intermediary starts it, the specialist owns the receiving process
This fits on one page. What situation triggered the referral, who the buyer is, what the intermediary has already said, what it promised, and who contacts the buyer next. Keep attribution unambiguous. Response speed matters more than people expect, because the intermediary spent relationship capital to make the call and a slow or clumsy first contact makes them look careless in front of their own client.
Owner: the specialist
Every failure in this research clusters here. The outsider should not be determining eligibility, designing the solution, or deciding what gets built, unless the outsider is genuinely expert in that domain. A standardized assessment, workshop, or discovery process makes diagnosis easier to distribute without handing over the judgment inside it.
Owner: the specialist, until the method is genuinely repeatable
Externalize delivery only once the process can be written down clearly enough for someone else to follow and quality can be measured. The utility programs are instructive because payment there depends on verified work against published specifications, which is a different arrangement from simply certifying a partner and hoping.
No portal. No tiers. No certification framework. No partner marketing program. Those are things a working system produces later, and building them first is the most common way to spend a year on this and get nothing.
If every introduction has to be prompted, there is no channel yet, and what you have is a slower and more expensive version of selling directly. If the introductions are good but delivery cannot absorb them, stop expanding the test until delivery can handle the additional customers.
Public evidence overstates how well this works, because companies that launch a partner program and get nothing rarely publish that outcome. Several figures above come from companies describing their own programs, and are labeled that way where it matters. The conclusion about Palantir and IBM rests partly on the absence of later evidence rather than on any announced failure. There is also a real possibility that good outside distribution forms around companies that were already succeeding rather than causing them to succeed, since intermediaries protect their own client relationships and steer clear of vendors who might embarrass them.