PastBehavior
Fieldwork / 011Download PDF

PastBehavior Research

When Internal Implementation Scales, and When It Breaks

When should a software company keep implementation in-house, and when should it push delivery to partners? Eight companies and twenty years of evidence give a conditional answer.

The cohort spans 2005 to 2026: two companies that externalized delivery by design, one that kept delivery internal and turned it into product, two that corrected late, one low-touch company that hybridized, one that sequenced partners before product breadth, and one live disconfirmation. Every material claim carries an evidence label.

FACTREPORTEDESTIMATEINFERENCEUNRESOLVEDEvery material claim carries one of these labels, defined in the methodology.

01 / The answer

Internal implementation scales when delivery work creates reusable product. It becomes dangerous when delivery work creates only another customer-specific deployment.

Internal implementation did not fail as a function of company size. It failed as a function of where the delivery work sat in the organization.

The organizational finding

The variable that best predicted the outcome was whether delivery work sat upstream of the product roadmap or downstream of the signed contract. Palantir kept implementation internal at multi-billion-dollar scale and it worked. Zuora carried a far smaller services burden and it broke. The difference was not scale.

The economic finding

A partner ecosystem only works when there is enough services spend for independent firms to build profitable practices around the product. ServiceNow is the strongest example: roughly $2 of services spend for every $1 of software, and its high-performing partners earn 40% to 55% gross margins. That is a precondition for external delivery, not a strategy choice.

The timing finding

Forward-deployed engineering launches, first GSI hires, and certification programs were corrections, not leading indicators. Every instance found in this cohort followed the constraint rather than preceding it. By the time they appear, management usually already knows.

This is not an argument that partners are better. The cohort contains two coherent ways for implementation to scale, and one way for it to break.

02 / Three models

Three paths, one dangerous middle.

01

Keep delivery inside, because it creates product

  1. Internal delivery
  2. Deployment work
  3. Reusable product
Companies

Palantir

Core idea

The people doing implementation also feed the product roadmap. Each difficult deployment makes the product better for the next customer.

Status

Working at multi-billion-dollar scale, with qualifications.

02

Push delivery outward

  1. Standardized delivery
  2. Partner ecosystem
  3. Vendor keeps product
Companies

Workday, ServiceNow

Core idea

Implementation is necessary but it is not the differentiating product. Partners absorb delivery while the software company reduces internal services intensity.

Status

The completed end state. Both companies reached it deliberately.

03

Stuck in the middle

  1. Customer-specific work grows
  2. Neither product
  3. Nor externalized
Companies

Zuora, Sprinklr

Core idea

Delivery grows with every new customer while the work neither becomes reusable product nor moves outside fast enough.

Status

The dangerous state. Both companies corrected late.

03 / Push delivery outward

Workday and ServiceNow: implementation is necessary, but it is not the differentiating product.

Both companies decided early that partners would absorb delivery, and both kept reducing internal services intensity for more than a decade. This is the completed version of the externalized model.

Workday professional services as a share of revenue

FY2012, at IPO
~33%
FY2025
8.62%
FY2026
7.53%

FY2026: professional services revenue fell in absolute dollars, from $728 million to $719 million, while subscription revenue grew 14.45%.

01

Workday

Stated the destination in a 2013 filing, then spent a decade getting there.

Cloud HR and finance for organizations above roughly 5,000 employees. Implementation is genuinely complex, and Workday decided early that complexity belonged to partners.

What happened

FACTA 2013 10-Q stated that partners would increasingly contract directly with subscription customers, and that professional services would decline as a percentage of revenue. The direction was public years before the transition completed.

FACTProfessional services fell from roughly 33% of revenue near the 2012 IPO to 8.62% in FY2025 and 7.53% in FY2026.

FACTIn FY2026, professional services revenue declined in absolute dollars, from $728 million to $719 million, while subscription revenue grew 14.45%. The FY2026 10-K describes an explicit push to reduce professional services intensity.

Outcome

FACTMore than 11,500 customers, roughly 97% gross revenue retention, and a 29% non-GAAP operating margin in FY2026.

Mechanism
The destination was named before the pressure arrived. Certification, direct partner-to-customer contracting, and a falling internal services share were the execution of a decision taken near IPO, not a reaction to a crisis.
02

ServiceNow

The completed endpoint: partners deliver almost everything, on purpose.

Roughly 8,100 enterprise customers, including 85% of the Fortune 500, with a reported 99% renewal rate. The most fully externalized delivery model in enterprise software.

What happened

REPORTEDPartners deliver roughly 90% to 95% of implementations, drive 80% to 90% of new bookings, and source about 20% of deals. Accenture alone reports more than 8,000 certified practitioners. One system integrator accounted for approximately 11% of revenue in Q4 2025.

FACTProfessional services revenue was $323 million in 2021, $354 million in 2022, and $291 million in 2023, against $8.68 billion of subscription revenue in 2023. Subscription gross margin runs above 81%; professional services runs at a negative gross margin.

INFERENCEThe internal services organization is kept small and deliberately unprofitable because its job is customer success and retention, not becoming a large services business. Chief Customer Officer Lara Caimi has framed partners as the delivery arm and internal services as the retention enabler.

Outcome

FACTProfessional services declining in absolute dollars while subscription compounds is the signature of the completed transition. ServiceNow has printed it more than once.

Mechanism
Services economics fund the ecosystem. Roughly $2 of services spend exists for every $1 of software spend, so independent firms can build profitable practices. The vendor keeps the product; the partners keep the delivery.

If implementation does not create product, the scalable end state is to standardize it and move more of it outside the company. Both companies said so early, then spent a decade finishing the transfer.

04 / The counter-case

Palantir: internal delivery can scale when the people doing implementation also feed the product roadmap.

This is the strongest counter-case in the cohort, and it is a genuine alternative model, not a footnote. It is also not proof that internal delivery always works: the conditions are specific, and Palantir itself is softening the model at both ends.

01

Palantir

The strongest counter-case: internal delivery that compounds into product.

Data integration and operations software for government and commercial customers. The Forward Deployed Engineer role was created in 2005 for intelligence work at the CIA, NSA, and Army. 2024 revenue was approximately $2.87 billion; Q1 2026 growth was 85% year over year, with commercial at roughly 46% of revenue.

What happened

FACTForward Deployed Engineers operate at the customer site and write production code. Recurring problems across bespoke Gotham deployments informed the development of Foundry as a reusable platform.

FACTAndreessen Horowitz, writing in January 2026, drew the distinction this study uses: FDEs operate upstream of the product roadmap; traditional consultants operate downstream of the contract.

REPORTEDServices run at roughly 18% to 20% of revenue, about 2.5 times Workday's current ratio, and the company is not trying to reduce that to Workday's level. Contribution margins reach roughly 55% at scale, with gross margin above 80% once an integration matures.

Outcome

INFERENCEPalantir has sustained a much higher services intensity than Workday or ServiceNow while continuing to grow, because delivery work feeds the roadmap. This is evidence that the model can hold, not evidence that it always will.

Mechanism
Each difficult deployment makes the product better for the next customer. The services organization is a research and development surface with a revenue line, not a cost center growing with the customer count.
Qualifications and caveats+

FACTPalantir is itself softening the model at both ends: AIP bootcamps aim to reduce the FDE hours required per customer, with data the company calls promising but not definitive, and a July 2026 alliance with DXC Technology externalizes some Fortune 2000 delivery at the edges.

REPORTEDCritics exist on the record. Anaplan CEO Charlie Gottdiener has called the FDE model a good sales tactic and a poor long-term strategy, and a Kinaxis executive who previously worked at Palantir has criticized its execution (Forbes, 10 July 2026).

Internal delivery can compound if each difficult deployment makes the product better for the next customer.

05 / Stuck in the middle

The dangerous state is not too much internal implementation. It is delivery that becomes neither product nor partner work.

Customer-specific delivery grows, but the work is neither becoming reusable product nor being externalized fast enough. Both companies below corrected late, and both corrections were visible in growth before they were visible in org charts.

01

Zuora

The commercial motion moved upmarket; the delivery capability did not.

Subscription billing and revenue software expanding from a billing base into a broader suite. Cross-sell depended on integrations and delivery capability that had not caught up to what sales was selling.

What happened

FACTOn the Q1 FY2020 earnings call, CEO Tien Tzuo said integration was taking longer than planned, that the company had slowed representations about its RevPro product, and that moving upmarket required a different sales approach and global system integrator involvement, both of which, in his words, take time to ramp (diginomica, May 2019 to March 2020).

FACTGrowth decelerated from 47% year over year in Q2 FY2019 to 17% in Q3 FY2020 while the correction was underway.

Outcome

INFERENCEThe GSI ramp began during the deceleration, not before it. Partner delivery capacity is a lagging asset with build time measured in years, so starting it after growth slows guarantees a gap.

Mechanism
Selling ahead of delivery capability. The deals arrived on the new product story; the integrations, the implementation playbooks, and the partner bench arrived later.
02

Sprinklr

Won the enterprise deals, then discovered delivery and renewal value lagged.

Customer experience management moving into large enterprise accounts, including Deutsche Telekom and BT, while implementation investment and renewal value realization lagged the wins.

What happened

FACTOn the Q2 FY2025 call, management said the renewal process was not demonstrating value and that newer products were requiring more product and implementation investment. Growth fell to 11% from 18% despite the flagship enterprise wins.

FACTSales and services were merged under a single Chief Customer Officer in February 2024, and the CRO departed. Rory Read was appointed CEO on 5 November 2024, with founder Ragy Thomas moving to Chairman.

Outcome

INFERENCEThe organizational correction came after the deceleration was visible, not before it. Merging sales and services was an admission that winning the deal and delivering it had become separate problems.

Mechanism
Delivery capacity lagged the commercial motion, and the gap showed up first in renewals. The reorganization followed the constraint by more than a year.

The risky state is a growing services organization that sits downstream of the contract and grows with every new customer.

06 / The framework

Upstream of the roadmap, or downstream of the contract.

The intellectual center of this study is a single distinction: what the delivery work produces after the customer signs.

Upstream of the roadmap

Implementation work creates

  • Reusable product capability
  • Better primitives
  • Standardized workflows
  • Product learning
  • Lower delivery intensity for the next customer

Can remain internal. Palantir at 18% to 20% of revenue is the proof the model can hold.

Downstream of the contract

Implementation work creates

  • One customer's configuration
  • One customer's custom workflow
  • One customer's migration
  • More services headcount
  • Work repeated for the next customer

Eventually needs to be standardized, automated, reduced, or externalized. Workday and ServiceNow are the completed versions.

The upstream/downstream formulation is from Andreessen Horowitz, January 2026. This cohort is the evidence base for it. The a16z warning that accompanies the distinction: the problem is pretending you are SaaS when you are really services with a platform.

What determines the model, ranked by evidence

  1. 01Organizational position of deliveryStrongUpstream of the roadmap or downstream of the contract. The best-supported variable in the cohort.
  2. 02Services-to-software spend ratioStrongDetermines whether a partner ecosystem is economically possible at all.
  3. 03Customer-specificity of the workStrongRepeatable deployments can be certified; bespoke ones compound headcount.
  4. 04Contract size and partner economicsModerate to strongPartners need engagements large enough to build a practice around.
  5. 05Product architectureModerate, weaker than expectedWorkday ran a unified architecture and externalized anyway. Architecture did not decide the model.
  6. 06Speed of expansion versus enablementModerateFast commercial motion with slow delivery ramp is the recurring break pattern.
  7. 07Strategic philosophyModerateWorkday's 2013 filing shows stated intent shaping outcomes years later.
  8. 08Geographic expansionWeakLooked like a driver until Deel refuted it by owning 55-plus jurisdictions internally.
  9. 09RegulationNot establishedThe cohort does not contain enough regulated-category variation to test it.

07 / False signals

What looks like an early warning but usually is not.

These moves look like early warnings. In this cohort they were corrections, visible only after management already knew there was a delivery problem.

  • Launching a forward-deployed engineering organization
  • Hiring the first GSI or SI leader
  • Launching partner certification
  • Creating an implementation partner directory
  • Rapid implementation hiring
  • Adding more solution architects
  • Merging sales and services leadership

What the record shows

  • Sprinklr merged sales and services in February 2024, after growth had fallen from 18% to 11%, and ten months before the founder was replaced as CEO.
  • Zuora began its GSI ramp during the deceleration, with the CEO describing on earnings calls why it would take time to ramp.

Every instance found in this cohort followed the constraint rather than preceding it. Treat these moves as confirmation that a problem is being addressed, not as the first sign it exists.

Signals that did not hold at all

  • Customer complaints: present at every company in the cohort, including the healthy ones. Non-discriminating.
  • Geographic expansion: looked like a stress signal until Deel refuted it.
  • Employee reviews and delivery turnover: no usable public signal.

08 / The real signal

The strongest leading indicator in the cohort.

The strongest observable warning in this cohort was sustained professional-services growth outpacing subscription growth, when the services organization was not producing reusable product.

Both clauses matter. The ratio alone is ambiguous: Palantir carries 18% to 20% services intensity and it is not a warning, because the work feeds the roadmap. The organizational qualifier is what turns a ratio into a signal. This is the strongest signal found in the cohort, not a universal law.

The signature of a completed transition

The mature externalized state is visible when professional services revenue can decline in absolute dollars while subscription revenue continues to grow.

Workday

Professional services fell from $728 million in FY2025 to $719 million in FY2026 while subscription revenue grew 14.45%.

ServiceNow

Professional services went from $323 million in 2021 to $354 million in 2022 to $291 million in 2023, against $8.68 billion of subscription revenue.

No threshold survived. No revenue, headcount, or customer-count threshold survived the cohort. Workday externalized before its IPO at low hundreds of millions of revenue. Palantir has not externalized at multi-billion-dollar scale. Zuora broke at a few hundred million. Size did not decide any of it.

09 / Partner economics

External delivery only works if an independent firm can make money doing it.

External delivery only works if an independent firm can build a profitable practice around the work.

FACTThe ServiceNow ecosystem runs at roughly a 2:1 services-to-software spend ratio, with the European ecosystem alone estimated at 5.3 billion euros (Volpi Capital, November 2025).

FACTTop-quartile ServiceNow partners grow at roughly 25% annually with 40% to 55% gross margins, 15% to 20% EBITDA margins, 70% chargeability, and day rates above 950 euros (Volpi Capital, November 2025).

INFERENCEThe 2:1 figure is one measured case, not a universal threshold. The general claim is narrower: where services spend is only a fraction of software spend, no partner practice can be built at any level of vendor effort.

INFERENCEThe check is observable from outside: do independent firms advertise paid implementation work for the product at all? Below viable engagement sizes, partner recruitment produces signed partners who never activate. PastBehavior's earlier partnership research put the base rate at roughly 20% to 30% of recruited partners ever producing.

REPORTEDThe disconfirming evidence: K2 Partnering Solutions' 2026 research found a persistent gap between what ServiceNow customers need and the certified talent supply, and Volpi Capital notes that many partners underperform. Externalizing relocates the constraint into a certification labor market. It does not remove it.

10 / The rest of the cohort

Atlassian hybridized, HubSpot sequenced, and one case refuses to resolve.

01

Atlassian

Even the strongest low-touch model hybridized when it moved upmarket.

Famous for low-touch distribution: more than 300,000 customers and 84% of the Fortune 500, with 524 customers above $1 million in annual recurring revenue.

What happened

FACTPresident Jay Simons: low touch does not mean no touch. The company built an Enterprise Advocate organization with quota-carrying representatives across three regions, growing for roughly four years as of 2019 to 2020 (Stanford GSB case, 2017, and subsequent coverage).

FACTSolution Partners handle complex enterprise rollouts. FastShift, launched to productize migrations, compresses timelines from 12 to 16 months down to 2 to 6 months (October 2025).

REPORTEDIn March 2026 the company cut roughly 10% of staff to pivot investment toward AI and enterprise sales.

Outcome

INFERENCEProductization can reduce the implementation burden, but enterprise complexity rarely disappears entirely. The lowest-touch company in the cohort still built enterprise sales, partners, and migration tooling.

Mechanism
Product-led distribution lowered the cost of acquiring customers; it did not eliminate the cost of deploying into large enterprises. The company absorbed that complexity with a hybrid of internal touch, partners, and productized migration.
02

HubSpot

The partner channel existed before the product breadth that needed it.

Marketing software origin that expanded into a multi-hub platform and moved upmarket gradually over roughly seven years, with a solutions partner ecosystem already in place.

What happened

FACTQ1 2026: 18.2% constant-currency growth, deals above $60,000 in annual recurring revenue up 37%, deals above $120,000 up 64%, and 42% of the Pro-plus install base by ARR owning four or more hubs.

REPORTEDManagement acknowledged enablement and packaging disruption during the same quarter. The expansion has not been frictionless; it has been manageable.

Outcome

INFERENCEDelivery and enablement capacity had time to develop alongside product breadth because the partner ecosystem predated the multi-hub expansion. The companies that handled the transition best built delivery capacity before they desperately needed it.

Mechanism
Sequencing. Partners, enablement, and packaging matured in front of the product surface area, so each new hub landed on existing delivery capacity rather than creating a new gap.
03

Deel

The live disconfirmation: vertical integration is still working.

Expanded horizontally from contractor payments and employer-of-record into a full HRIS, payroll, equity, equipment, and immigration, while owning entities and payroll engines in more than 55 jurisdictions.

What happened

REPORTEDApproximately $1.4 billion in annual recurring revenue as of March 2026, with an October 2025 raise at a $17.3 billion valuation.

ESTIMATEThird-party modelling puts gross margin near 85% and EBITDA margin near 16% for 2025. The figures are unaudited and not company-disclosed.

Outcome

UNRESOLVEDA large, horizontally expanding company is still making an internal, vertically integrated model work. The evidence does not settle whether this is durable or simply early, and this study does not force a conclusion from it.

Mechanism
Where delivery requires owning regulated infrastructure, internal delivery may be the product. Whether that generalizes beyond Deel's category is an open question.

The cohort, side by side

Workday

Delivery model
Internal at IPO, deliberately externalized
Delivery creates product
No. Treated as a transferable cost
Partner ecosystem
Partners contract directly with customers
Outcome
Scaled. Services 33% of revenue near IPO, 7.53% in FY2026

ServiceNow

Delivery model
Almost entirely external
Delivery creates product
No. Internal services run at negative margin for retention
Partner ecosystem
90% to 95% of implementations
Outcome
Scaled. The completed endpoint

Palantir

Delivery model
Internal by design
Delivery creates product
Yes. Gotham deployments produced Foundry
Partner ecosystem
Beginning at the edges (DXC, 2026)
Outcome
Holding at 18% to 20% services intensity while growing

Zuora

Delivery model
Internal, GSI ramp started late
Delivery creates product
No
Partner ecosystem
Began during the deceleration
Outcome
Corrected late. Growth 47% to 17%

Sprinklr

Delivery model
Internal, capacity lagged the wins
Delivery creates product
No
Partner ecosystem
Not central to the model
Outcome
Corrected late. Growth 18% to 11%, then reorganization

Atlassian

Delivery model
Product-led origin, hybridized
Delivery creates product
Partly. FastShift productizes migrations
Partner ecosystem
Solution Partners for complex rollouts
Outcome
Hybridized. Enterprise complexity absorbed, not eliminated

HubSpot

Delivery model
Hybrid, partner channel first
Delivery creates product
Partly
Partner ecosystem
Ecosystem predated the multi-hub expansion
Outcome
Scaled with manageable friction

Deel

Delivery model
Internal by vertical integration
Delivery creates product
Owns delivery infrastructure in 55-plus jurisdictions
Partner ecosystem
No
Outcome
Unresolved. Currently working

11 / What worked

What tended to work, and what tended not to.

What tended to work

  1. 01

    Decide the model early

    Workday stated its direction in a 2013 filing, years before the transition completed. Early intent made the long transfer look like execution rather than retreat.

  2. 02

    If delivery stays internal, keep it upstream of product

    Palantir's FDEs write production code and feed the roadmap. The services organization is a product surface, not a delivery queue.

  3. 03

    If delivery does not create product, reduce internal services intensity over time

    Workday and ServiceNow both reached the state where services revenue can fall in absolute dollars while subscription grows.

  4. 04

    Build partner delivery capacity before the deals that require it

    Zuora shows what happens when the ramp starts late: partner capacity has a build time measured in years, and the gap lands on growth.

  5. 05

    Sequence the partner channel ahead of product breadth

    HubSpot's ecosystem predated the multi-hub expansion, so each new product landed on existing delivery capacity.

  6. 06

    Use productization to reduce services intensity regardless of model

    Palantir's AIP bootcamps, Atlassian's FastShift, and standardized deployment all shrink delivery effort per customer without changing the underlying model.

What tended not to work

  1. 01

    Selling ahead of delivery capability

    Zuora slowed representations about a product it had already sold. The gap between the commercial motion and delivery showed up directly in growth.

  2. 02

    Starting the partner ramp only after growth slows

    In this cohort, late ramps guaranteed a multi-year gap between the deals and the capacity to deliver them.

  3. 03

    Treating referral partners as implementation capacity

    Signed partners are not delivering partners. Prior research puts the share of recruited partners who ever produce at roughly 20% to 30%.

  4. 04

    Assuming product-led means implementation complexity disappears

    Atlassian, the lowest-touch company in the cohort, still built enterprise sales, solution partners, and migration tooling.

  5. 05

    Letting internal services become a large downstream organization that grows with each customer

    That is the state that broke Zuora and Sprinklr. Headcount that scales with customer count, producing nothing reusable, is the failure mode itself.

12 / The decision framework

Three coherent positions. The test is which one the delivery work actually supports.

Three coherent positions emerged from the cohort. The test is not which one sounds right; it is which one the delivery work actually supports.

Keep implementation internal if

  • Deployment work consistently produces reusable product
  • Customer learning materially shapes the roadmap
  • Contract values support the internal delivery model
  • Delivery intensity per customer is falling over time

The Palantir position.

Push more implementation outward if

  • Deployments are increasingly customer-specific
  • The same work gets repeated across customers
  • Services headcount rises with customer count
  • Professional services grows faster than subscription
  • Partners can build economically viable practices around the work

The Workday and ServiceNow position.

Hybridize if

  • Strategic or unusual implementations generate product learning
  • Repeatable deployments can be standardized and certified
  • Enterprise or regional complexity exceeds what internal teams should absorb

The Atlassian and HubSpot position.

The question is not whether implementation is internal or external. The question is what happens to the work after the customer goes live.

PastBehavior

Methodology and limitations

Case selection

Eight companies, 2005 to 2026, assembled on structural position: two completed externalizations, one sustained internal model, two late corrections, one low-touch hybridizer, one partner-first sequencer, and one live disconfirmation.

Salesforce was considered and excluded because dated figures on partner-delivered share at the point of horizontal expansion could not be obtained. Snowflake, Datadog, Twilio, UiPath, Toast, and Shopify were not researched for this study and are excluded rather than asserted about.

Evidence labels
FACT: verifiable public record: a filing, a dated statement, a documented event. REPORTED: contemporaneous secondary source, named where used. ESTIMATE: modelled or approximate figure, not audited or company-disclosed. INFERENCE: the study's reading of the evidence. UNRESOLVED: the public record does not settle the question.
Confidence

High confidence: the Workday arc, the ServiceNow partner-delivered share, the absence of a size threshold, and corrections being misread as signals.

Medium confidence: upstream versus downstream as the primary variable, generalizing the 2:1 ratio from one measured case, and the durability of Palantir's model.

Unresolved: Palantir's commercial scaling without delivery erosion, Deel's audited economics, a dollar floor for partner viability, and whether the services ratio is cause or symptom.

Known weaknesses

  • Selection bias runs against internal delivery. A successful internal model generates no public artifact: no reorganization, no earnings-call language, no press coverage. This cohort likely overstates how reliably internal delivery breaks, because the internal models that worked are invisible.
  • Private-company services economics are not disclosable, which is why the Deel evidence is thinner than the public-company evidence.
  • Several load-bearing figures come from 2026-dated secondary sources rather than filings. They are marked REPORTED or ESTIMATE where they do material work.
  • The 2:1 services-to-software ratio rests on one measured case. It is presented as evidence that partner ecosystems need real economic room, not as a threshold.

What would change the answer

  • A company running downstream, customer-specific delivery at scale, profitably and permanently, with no externalization and no product absorption. The study searched for this case and did not find it.
  • Evidence that ServiceNow's 2:1 ratio is idiosyncratic to IT service management rather than a general precondition.
  • Palantir launching a formal implementation partner program, which would move the strongest counter-case toward the externalized end state.

Deciding what to do with implementation?

PastBehavior studies what has already happened so operators can see their position clearly before the correction is forced. If you are weighing internal delivery against a partner motion, we can map the historical cases against your situation.

Discuss a delivery model decision