Ten companies that won their first customers through embedded engineers, senior operators, and custom implementation, and then tried to grow without repeating all of it. None of them made the deployment fully standard. The ones that scaled made something around it repeatable: the proof, the implementation capacity, the economics, or the path to adoption.
A company wins its first large enterprise customers the hard way. Founders in the room. Senior operators who already know the industry. Engineers embedded at the customer site. Workshops, custom integration, and a great deal of human judgment about what to build.
It works. It works again. Then somebody looks at the next twenty customers and says the sentence every implementation-heavy business eventually says: we need to make this repeatable.
The obvious reading of that sentence is that the deployment has to become standard.
The documented histories point somewhere else.
Across ten companies with traceable public records, not one eliminated the human deployment layer. Guidewire is roughly two decades into a deliberately partner-led implementation strategy with a systems integrator ecosystem numbering hundreds of certified consultants, and its own services revenue grew 21% last fiscal year. C3.ai signed an exclusive channel agreement and then contractually committed to grow its own forward deployed engineering headcount in fixed proportion to the partner's sales headcount.
What the successful companies did was narrower and more interesting. Each found one bottleneck adjacent to the deployment and made that cheap. Three mechanisms show up clearly.
Palantir standardized proof. Guidewire externalized implementation capacity while keeping accountability. nCino built an ecosystem where another company earns money automatically every time nCino sells.
The failures are more uniform. They acquired introductions, credibility, partner logos, or guaranteed revenue, and modeled those things as a distribution channel.
A partner, investor, or customer creates introductions, leads, credibility, or access. The vendor still qualifies, closes, and deploys. Cheapest to acquire. Easiest to mistake for the others.
Implementation becomes faster, cheaper, or gets performed by someone else. The vendor's own cost per customer falls, or the customer's total cost falls, or both.
One relationship reliably produces access to or adoption by many customers. The word doing the work is reliably.
A company can hold one without the others. Several of the cases below acquired the first, described it publicly as the third, and found out the difference when the intermediary's priorities changed.
Palantir created the forward deployed engineer because intelligence customers could not be served by consultants who could not write production code or by solutions engineers who could not change the product. Two decades later the company still describes embedded engineering as its central mechanism. On the Q2 2026 earnings call the CTO noted that it has become fashionable for other companies to hire FDEs and that most of them do not understand what they are copying.
The commercial business was the constraint. US commercial customer count stood at 80 at the end of 2021 and 143 at the end of 2022. Those are small numbers for a company that had been selling commercially for years, and they are what a motion looks like when every new logo consumes scarce senior engineering time.
In February 2021, IBM and Palantir announced a joint offering built on Palantir Foundry and IBM Cloud Pak for Data, targeting retail, financial services, manufacturing, healthcare, and telecommunications, with general availability planned for that March. IBM named the partnership in its Q1 2021 prepared remarks.
Unresolved. Neither company appears to have disclosed revenue, customer count, or deployments from it since. Neither announced its end. A partnership presented as a route into five industries left nothing measurable in either company's public reporting. That is not proof of failure, and it is worth sitting with.
With the launch of AIP in the second quarter of 2023, Palantir introduced bootcamps: one to five days, hands on keyboard, on the customer's own data, ending with a working application. The company ran 92 in 2022 and more than 500 in 2023 across more than 465 organizations. On the Q3 2023 call, President Ryan Taylor said the company was on track to run bootcamps for more than 140 organizations by the end of November, with nearly half falling in that month alone, which exceeded the number of US commercial pilots run in the entire prior year.
| US commercial customers | Count |
|---|---|
| Q4 2021 | 80 |
| Q4 2022 | 143 |
| Q1 2023 · AIP launch quarter | 150 |
| Q1 2024 | 262 |
| Q1 2026 | 615 |
US commercial revenue grew 149% year over year to $764 million in the second quarter of 2026. Bootcamps are run by Palantir and by partners including Accenture.
Evidence-backed inference. Palantir did not make deployment cheap. It made the first three days of the engagement cheap, repeatable, and self-funded. The expensive human judgment moved out of an open-ended sales cycle and into a fixed format that can be run several times a day and handed to partners. Behind a converted bootcamp, the real deployment still involves ontology work and embedded engineers, and it can stay complicated for years.
The bet is the same one the FDE model always made. Palantir spends its own engineering time before any revenue exists. What changed is that the bet now takes days instead of months, and it can be placed hundreds of times a year.
On 6 June 2019, C3.ai appointed Baker Hughes as its exclusive reseller into the worldwide oil and gas field, defined to cover upstream, midstream, downstream, distribution, petrochemical, LNG, and fertilizer. The Joint Venture Agreement was filed as an exhibit to C3.ai's S-1, which is why this case can be examined at a level of detail the others cannot.
Four provisions matter.
| Provision | What it says |
|---|---|
| Exhibit B-1 | Minimum Annual Revenue Commitment of $50 million in Year 1, $100 million in Year 2, $170 million in Year 3. $320 million total, described as irrevocable and non-cancellable. |
| Shortfall | If actual revenue falls below the commitment, Baker Hughes pays the difference in cash. The contract works the example: $45 million actual against a $50 million commitment produces a $5 million shortfall payment. |
| Section 3.2(b) | C3.ai must maintain forward deployed engineering personnel to support the Baker Hughes sales team, at a fixed ratio of C3 FDEs to Baker Hughes sales FTEs. The ratio is redacted. The structure is not. |
| Exhibit E, 2.2 | Custom workflows built for customers are to be reviewed periodically with the partner "for potential generalization and incorporation into the C3.ai product roadmap." |
A Third Amendment filed in 2021 records that Baker Hughes had secured the Year 3 commitment and that C3.ai would pay Baker Hughes a sales commission of $16,000,000 for that year.
Baker Hughes accounted for roughly 10% of C3.ai revenue at IPO and as much as 45% two years later, per CNBC reporting. The same reporting states that the agreement was amended repeatedly, with Baker Hughes negotiating the annual commitments downward. In May 2025 the parties announced a renewal through June 2028; Tom Siebel's quote in that release attributes more than half a billion dollars of oil, gas, and chemical revenue to the alliance. That is an interested-party statement in a company press release.
Unresolved. How much of that revenue was originated by Baker Hughes sellers at end customers, how much was Baker Hughes buying for itself, and how much was shortfall payments, cannot be determined from public sources. That separation is the entire question and neither party has disclosed it.
Evidence-backed inference. A minimum revenue commitment with a cash shortfall clause turns the partner into a purchaser of last resort. When such an arrangement produces revenue, the revenue is evidence about the partner's balance sheet and negotiating position. It is not evidence that the channel works.
Section 3.2(b) is the detail an operator should sit with longest. Inside an exclusive channel arrangement, with a partner supplying its own dedicated sellers, C3.ai still committed by contract to scale its own forward deployed engineering headcount in proportion to that partner's hiring. The channel did not remove the vendor's human deployment layer. It indexed that layer to somebody else's headcount plan.
Core policy, billing, and claims replacement at a property and casualty insurer is among the most implementation-heavy categories in enterprise software. Guidewire built PartnerConnect, a tiered consulting alliance with Select, Advantage, and Global levels, the top tier split into Premier and Strategic. Guidewire's own description of the Global Strategic tier is that these partners drive market expansion and sales globally.
Accenture joined as a Select consulting partner for continental Europe and Latin America in June 2016. Accenture's own materials claim more than 500,000 hours of Guidewire services delivered, and note that it serves 27 of the world's top 30 P&C carriers. That is a partner's marketing document, and it establishes that the implementation capacity is real and durable.
Then the question the case exists to answer. Did the vendor's own human layer shrink?
| Guidewire, fiscal year 2025 | Revenue | Growth |
|---|---|---|
| Total | $1,202.5m | +23% |
| Subscription and support | $731.3m | +33% |
| License | $251.9m | +1% |
| Services | $219.2m | +21% |
Services were 18.2% of revenue and grew at 21% against total growth of 23%. Roughly two decades into a deliberate partner-led implementation strategy, at a company whose largest integrator has logged half a million hours, the vendor's own services line did not shrink toward zero. It stabilized as a fraction and kept growing in absolute terms. A small part of that line is an accounting artifact, since Guidewire discloses that $9.5 million of recurring contract value was recognized as services revenue in the year. It does not change the shape.
Hypothesis. The residual is not implementation labor that failed to move. It is the part of implementation that cannot be delegated: platform architecture decisions, escalations partners cannot resolve, and the work of holding partner delivery quality inside a range the vendor can defend to a customer.
This is what success looks like. It is worth having. It is also not what most people picture when they say the partners will do the implementations.
nCino began around 2011 as a majority-owned subsidiary of Live Oak Bancshares, built to solve the bank's own commercial lending workflow, then spun out. Live Oak's own filing describes organizing nCino as a subsidiary, spinning it off, and licensing back the rights to use the system nCino owns. By the fiscal year ending January 2020 the company reported $138.2 million of revenue, up 51%, with more than 1,180 customers across ten countries.
The mechanism that carried it sits in one sentence of the 10-K.
Evidence-backed inference. Salesforce earns platform subscription revenue on every nCino sale, held equity through Salesforce Ventures, and faced no channel conflict because it was never going to build banking core workflows itself. The incentive is per transaction and it is automatic. It does not require anyone at Salesforce to notice, care, or act.
That is the distinguishing feature of every durable distribution relationship in this study, and its absence is the distinguishing feature of the ones that stalled.
A second mechanism runs alongside it and deserves separating. Building the product inside a working bank produced software fitted to bank workflow before any external customer existed, and a founding team of bankers rather than engineers who had interviewed bankers. Live Oak then divested. Hypothesis. The anchor-customer-as-parent pattern works when the parent exits and fails when the parent stays, because a controlling parent has every reason to keep the roadmap fitted to itself, and every prospective customer knows it.
Veeva launched QualityOne and entered consumer products in 2016 and chemicals in 2017. Its FY2019 10-K states that it is bringing its content and data management solutions to customers outside life sciences in regulated industries including consumer goods, chemicals, and cosmetics, and argues that solutions meeting life sciences compliance requirements translate well into many other regulated industries. The FY2020 filing repeats the language almost verbatim.
The adjacency was well chosen. Cosmetics, chemicals, and consumer goods share the exact structural feature the product addresses, which is regulated product development with document control and quality management obligations. The technology transferred.
| Marker | What the record shows |
|---|---|
| 2017 | Around 10 of Veeva's 520-plus customers were outside life sciences. |
| Late 2019 | More than three years after entry: over 40 early adopters across three offerings, with segment revenue described as not yet material and the addressable market estimated at $1 billion or more annually. |
| Today | Veeva does not separately disclose non-life-sciences revenue. For a company that reports R&D and Commercial segments, the absence of a third line after roughly a decade carries information. |
Unresolved. Whether Veeva starved the adjacency by choice because the life sciences core kept compounding faster, or whether the adjacency resisted entry, cannot be separated from public sources. Both readings fit, and a company growing above 20% in its core has an entirely rational reason to underfund a second market. That is a different finding from the second market being hard.
What can be said is narrower and still useful. Shared product infrastructure lowered the cost of building for a new vertical. Nothing in the record suggests it lowered the cost of being credible in one. In life sciences, Veeva had a decade of reference network, regulatory fluency, and the fact that a Veeva decision was a defensible decision. None of that came in the platform.
Spun out of Ping An in 2015, IPO'd on the NYSE in December 2019 with an explicit strategy of reducing revenue concentration from the parent. Dependence rose instead, from 56% of revenue in 2020 to 65% in 2023. In May 2024 Ping An terminated the cloud services contract in favor of an in-house replacement; the parent generated roughly 60% of that segment, the segment closed, and about a third of revenue went with it. The company was taken private by a Ping An subsidiary and delisted in November 2025, six years after listing. Anchor dependence can become concentration rather than distribution, and an anchor who is also the parent can build its own replacement faster than any third party could switch.
The SS&C scheme document filed in 2021 records that Blue Prism had approximately 100 technology partners and, in 2020, 167 distribution partners who implemented its product and earned associated fees. The company was acquired anyway, while its principal competitor scaled far larger on a mixed direct and partner motion. Partner count is not evidence of channel quality. Unresolved: Blue Prism's split between partner-originated and vendor-originated revenue is not public, so this rests on the outcome rather than the channel's own numbers.
Blackstone runs a group purchasing platform leveraging more than $30 billion of spend across its portfolio in more than 75 indirect spend categories. Portfolio participation is governed by clauses in Support and Services Agreements filed with the SEC, and the language is precise: participation is an opportunity, "on terms mutually agreed," and the firm "may receive commissions, payments or fees from vendors" on spending through the program. A rebate on spend volume rewards aggregating high-volume, low-variance, commoditized purchases. Shipping rates, laptops, telecom, employee health benefits. Nothing in that structure rewards anyone for getting twelve portfolio companies to each run a nine-month bespoke implementation.
More than 85 enterprise software businesses, benchmarked against Vista standards, with functional calls, formal certification of portfolio employees in Vista methodologies, and real-time tracking of recurring revenue and churn across the portfolio. Where Vista turned portfolio scale into commercial scale, the documented mechanism was merger: Misys and D+H, both portfolio companies, combined into Finastra in 2017. Portfolio networks clearly diffuse operating practices. Evidence that they diffuse an implementation-heavy product is weak.
The only intermediary in the study that can compel. Franchise agreements routinely reserve the right to require franchisees to adopt specified technology at their own expense. Item 8 of the Franchise Disclosure Document requires disclosure of designated suppliers and any franchisor financial gain, including percentage-of-sales payments. In JDS Group v. Metal Supermarkets (W.D.N.Y. 2017) a court declined to enjoin a mandated POS rollout despite six franchisees documenting that the software did not work. Contractual mandate is genuine distribution power. It is also structurally unusual, bounded by antitrust and state franchise-fairness law, and it produces compelled buyers rather than convinced ones.
Vendors routinely build private equity go-to-market motions on the assumption that the mechanism is established. In filings, earnings material, and credible reporting, this study found little documented evidence that PE ownership alone turns implementation-heavy enterprise software into a repeatable portfolio-wide distribution channel.
The precise form of that statement matters. This is a null result rather than a demonstration that the channel fails. Portfolio-wide deployments would be commercially sensitive and rarely disclosed, and the absence of a public record is not the absence of the thing. But it is relevant that a vendor building a PE portfolio motion is acting on a proposition the public record does not support, and most do not know that.
What is well documented is the adjacent mechanism, and the contrast is instructive. Group purchasing works, at very large scale, with contractual formalization, for standardized indirect spend. It runs on vendor-side commissions, participation is optional by contract, and the categories are commodities. It is a procurement aggregator. It has been repeatedly mistaken for a distribution channel because both are described in the same sentence: the PE firm gets its portfolio companies to buy your thing.
Evidence-backed inference. The two are separated by variance. Procurement aggregation rewards the intermediary for volume in low-variance categories. Bespoke software implementation is the highest-variance purchase a portfolio company makes, the rebate is small, and the reputational exposure if implementations go badly sits with the firm that recommended the vendor.
Across the cases, the outcome separates cleanly by one variable: what the investor earns when the vendor wins.
| Investor and vendor | Economic interest when the vendor wins | Result |
|---|---|---|
| Salesforce Ventures / nCino | Platform subscription fee on every sale, plus equity | Distribution leverage. Automatic, per transaction, requires no decision. |
| In-Q-Tel / Palantir | Mission capability for the investing agency | An extraordinary anchor customer. Never a channel to anyone else. |
| Baker Hughes / C3.ai | A revenue commitment it owed, plus a $16m sales commission paid by the vendor | Sales leverage. Guaranteed revenue obscured whether distribution existed at all. |
| Ping An / OneConnect | Consuming its own software | Anchor that became concentration risk, then built a replacement. |
| Blackstone / portfolio | Vendor-side commission on aggregate spend | Procurement leverage, in commodity categories only. |
The one case that produced durable distribution is the one where the investor earned money on every individual transaction automatically. Where the benefit was strategic, reputational, or negotiated case by case, the mechanism produced introductions and stopped there.
When a new customer buys, who at the intermediary had to decide to help us, and what do they earn if they do?
If nobody has to decide and the intermediary benefits automatically, you have one of the strongest forms of distribution leverage in the cohort. The nCino and Salesforce structure and the franchise Item 8 structure are the clearest examples, because the incentive or the mandate operates without a person choosing to champion each individual deployment.
If someone had to decide to make the introduction, champion the vendor, or push the opportunity forward, that is sales leverage. It can be extremely valuable. It is what opened oil and gas to C3.ai and what opened the first doors for several others in this study. It should not be modeled as a compounding channel, because it reverses when the intermediary's priorities move.
How many hours from the people we cannot easily replace does customer twenty require, compared with customer five?
Not total headcount. Every successful company in this study added people, and Guidewire's services revenue is still growing at roughly the rate of the business. The question is whether scarce senior judgment per incremental customer declined.
In the cases that worked, that number fell because the proof stage was compressed. In the cases that did not, it stayed flat, or, in one contractually explicit instance, it was indexed to a partner's hiring plan.
Not the deployment. Depending on the case, it was one of four things.
Each of these appeared in the cases, each was real, and none of them on its own established that one relationship reliably produces many customers.
One more observation, offered without a neat resolution. Every successful case took years. nCino from a bank subsidiary in 2011 to more than a thousand customers by 2020. Guidewire's partner ecosystem built over roughly two decades. Palantir's bootcamp motion arrived eighteen years after the FDE role was invented. Nothing in the record suggests the transition can be accelerated by picking a better mechanism. The mechanisms that worked look less like decisions and more like the accumulated result of doing the expensive human version until enough of it had been generalized to be worth packaging.
The question is not whether the deployment can become standard. It is whether the scarce human judgment surrounding it becomes cheaper with every customer.
Ten cases, selected because credible public evidence exists about what happened after the anchor-customer stage, which excludes most companies people reach for in this conversation. Claims are drawn from SEC filings and exhibits, UK scheme documents, earnings releases and calls, court records, executive interviews, and contemporaneous reporting. Interested-party sources are identified where they are used.
The cohort over-weights companies that went public or were acquired, because those are the ones with filing obligations. Private companies that solved this quietly are invisible to the method, and there is no reason to think there are none. Failures also generate more disclosure than smooth successes do, which biases the study toward caution. Read the findings as what the documented record shows rather than as a base rate.
Where the record conflicts, the conflict is reported rather than resolved. Where a pattern weakened under scrutiny, it is stated in the weaker form.