Build-Buy-Partner-Borrow Sourcing Strategy

COMPEL Specialization — AITE-WCT: AI Workforce Transformation Expert Article 7 of 35

A global bank announces an expansion of its AI engineering function and commits to a plan to hire a specified number of machine-learning engineers in the following year. Nine months later the plan has delivered below target, the externally-hired engineers are arriving at rates that outpace the bank’s ability to integrate them into its governance and operating model, and the bank’s internal analysts — who had adjacency to the new roles but were not invited to apply — are starting to leave. The plan was not wrong; it was single-mode. The build-buy-partner-borrow framework is the discipline that prevents single-mode sourcing. This article teaches the expert practitioner to evaluate the four modes against five criteria, to design a deliberate hybrid that balances them, and to avoid the mode-selection biases that turn sourcing strategy into political preference.

The four modes defined cleanly

The first distinction the expert practitioner must hold is between what the four modes actually are, stripped of vendor marketing. The modes define where the capability originates and under what employment relationship.

Build. The capability is developed inside the organisation from employees already on the payroll. Build relies on internal development — apprenticeship, fellowship, training, stretch assignment. Build is slow early and fast later; it preserves institutional knowledge; it produces employees with deep contextual understanding. Build’s limit is speed to initial capability.

Buy. The capability is sourced externally as permanent hires. Buy is fast early and produces specialists who bring external experience; its limit is the competitive labour market, integration time, and the risk that externally-hired specialists do not retain when the organisation’s internal practice does not match their expectations.

Partner. The capability is sourced through a commercial partnership with another organisation — a systems integrator, a consulting firm, a university research partner, a vendor professional services organisation. Partner is fast, preserves optionality, and transfers risk to the partner; its limit is knowledge-transfer — capability that lives only with the partner leaves when the partnership ends.

Borrow. The capability is sourced through contingent labour — contractors, short-term consultants, staffing agencies, marketplaces such as Toptal or Upwork. Borrow is fastest, most flexible, and least committed; its limit is duration — the capability is present only while the engagement runs.

The four modes overlap in practice. A systems integrator may provide both partner-mode programmatic help and borrow-mode individual staff augmentation. A build programme may include a buy-mode anchor hire to seed the internal practice. The expert practitioner resists the tendency to describe a mode by the vendor’s label and instead describes it by the employment structure and knowledge-retention profile.

The five criteria for choice

Five criteria jointly determine which mode — or mode combination — fits a specific capability need.

Capability criticality. How strategic is the capability to the organisation’s future? Strategic capabilities are candidates for build because retention of the capability inside the organisation matters. Peripheral capabilities are candidates for buy, partner, or borrow because retention matters less than efficiency. A bank’s fraud-detection AI capability is strategic; the bank’s marketing-content AI capability may be peripheral. Both assessments are organisation-specific.

Speed. How urgently is the capability required? Urgent needs bias towards buy, partner, or borrow. Non-urgent needs permit build. A regulator-imposed deadline for bias monitoring cannot wait for a build programme to mature; a multi-year AI research programme can.

Cost. What is the total cost of ownership across time horizons? Build is cheap at steady state and expensive at ramp-up; buy is expensive in external labour market premium but saves the ramp-up investment; partner is expensive per unit but avoids fixed-cost commitment; borrow is expensive per hour and cheapest to terminate. Cost comparison must be across the relevant horizon — short horizons favour borrow; long horizons favour build.

Retention risk. How likely is the externally-sourced capability to retain? Buy-mode retention risk is highest when the labour market is competitive and the organisation’s practice does not match external expectations. Borrow-mode retention is not applicable by design. Partner-mode retention depends on the partner’s internal continuity. Build-mode retention depends on the retention programme Article 11 covers.

Intellectual-property sensitivity. Capability built on sensitive IP — proprietary data, models trained on protected data, governance logic — is a poor candidate for partner or borrow. Build is preferred; buy is a distant second. Partner arrangements with explicit IP clauses and borrow arrangements with NDA support can be made to work but require careful contracting.

[DIAGRAM: Matrix — build-buy-partner-borrow-criterion-matrix — rows: five criteria. Columns: build, buy, partner, borrow. Cells: relative fit (strong / moderate / weak) annotated with the primary rationale. Primitive teaches the five-criterion evaluation as a design artefact.]

Designing the hybrid

Real sourcing is almost always hybrid. A mature hybrid allocates the capability need across modes according to the criteria. A worked example for a knowledge-graph capability at a mid-size insurer illustrates the pattern.

The organisation decides that the knowledge-graph capability is strategic (criticality favours build), needed within nine months (speed favours buy or partner), carries large fixed cost if fully built internally (cost favours partner at ramp), has moderate retention risk for externally hired specialists because the internal practice is still forming (retention favours build once capability is established), and involves sensitive data (IP favours build). The integrated answer is a hybrid: partner with a specialist firm for the first six months to deliver initial capability, hire two buy-mode anchor specialists to lead the internal team, and run a twelve-month build programme drawing candidates from the existing data team through internal apprenticeship. The borrow mode is used selectively for peak-load tasks during the first year. Each mode is measured and reviewed quarterly.

This hybrid is the style of answer the practitioner produces. A single-mode answer — “we’ll hire ten ML engineers” — is an under-specified plan regardless of how well-executed the single mode is. McKinsey’s Superagency in the workplace (January 2025) and WEF’s Future of Jobs Report 2025 both document the cross-industry shift towards hybrid sourcing in AI capability development.¹² BCG’s AI at Work 2025 provides complementary sourcing-mix data.³

The mode-selection biases

Three biases recur in mode-selection discussions. The expert practitioner identifies and counters them.

Vendor preference. The capability conversation starts with a specific vendor — usually one the sponsor has heard from recently — and the mode choice is reverse-engineered to fit the vendor. The counter is to articulate the capability need before engaging any vendor and to evaluate the five criteria before any vendor conversation begins.

HR-process preference. Internal HR processes are designed around the buy mode (hiring is the default) or around the build mode (development is the default), and the mode-choice discussion inherits the process bias. The counter is to make the mode choice before the HR process choice and to design process around the chosen mode rather than the reverse.

Risk-transfer attraction. Partner and borrow modes transfer operational risk to the supplier, and this attractive property draws decisions towards those modes for reasons that are not about the capability. The counter is to distinguish operational-risk transfer (legitimate) from strategic-risk transfer (usually illegitimate — strategic capabilities cannot be permanently parked with a partner).

A fourth bias — IBM-style public narrative attraction — occasionally appears when sponsors see public announcements of aggressive automation-led workforce reduction (for instance IBM’s May 2023 public pause on back-office hiring) and conclude that similar announcements are appropriate for their own organisation.⁴ The bias conflates the sourcing mode with a public-narrative choice. The two are separable — an organisation can run an aggressive build programme without an attention-grabbing public narrative, or vice versa.

Mode-specific governance requirements

Each mode carries specific governance requirements the practitioner must build into the sourcing plan.

Build governance. Development programmes require curriculum (Article 13), apprenticeship and fellowship structures (Article 10), and measurement of capability acquisition (Article 15). The programme is tracked as a portfolio, not as individual learners.

Buy governance. External-hire programmes require inclusive-hiring practices (Article 8), screening-tool oversight for disparate impact, onboarding programmes that reach productive-capability faster than unmanaged ramp, and retention attention from day one.

Partner governance. Partnership programmes require knowledge-transfer clauses, joint-staffing arrangements so internal practitioners learn alongside partner staff, and transition-out planning from the start. The Dutch Toeslagenaffaire illustrates what can go wrong when partner-delivered systems are operated without internal technical understanding; the partner structure is not causal in that case but is a structural factor in how understanding of the system within the deploying organisation developed.⁵

Borrow governance. Contingent-labour programmes require contracting that covers IP, confidentiality, and data protection; co-employment risk management; and explicit handoff of any work product that will persist beyond the engagement. The NLRB’s 2022–2024 activity including cases involving Amazon has highlighted how contingent-labour governance intersects with broader labour-relations considerations.⁶

[DIAGRAM: HubSpokeDiagram — hybrid-sourcing-mix-example — central hub “Knowledge Graph Capability” with four spokes to the four modes, each spoke labelled with the capability slice allocated, the duration, the criterion rationale, and the governance requirement. Primitive teaches hybrid allocation as a single visual.]

Platform and vendor neutrality

The HRIS that houses the workforce data (Workday, SAP SuccessFactors, Oracle HCM, ADP, UKG, BambooHR) and the talent-marketplace platform that may support internal mobility (Gloat, Fuel50, Eightfold, 365Talents, Lightcast) are evaluation targets, not sourcing decisions. The sourcing mode selects the capability; the platform enables the flow. The LMS platforms that support build-mode development — Docebo, Cornerstone, Workday Learning, SAP SuccessFactors Learning, Open edX, Moodle — are similarly neutral choices. Sentiment platforms — Qualtrics, CultureAmp, Peakon, Glint — provide the feedback loop on employee experience across all four modes. No single platform is canonical in COMPEL practice.

Transition between modes

Capability needs evolve, and so do the right mode mixes for them. The expert practitioner designs for transition between modes as a first-class concern.

A capability initially sourced through partner mode at programme launch frequently transitions towards build mode as the internal practice matures. The partner engagement structure should anticipate the transition from the start. Knowledge-transfer clauses, joint-staffing arrangements, progressive hand-over of work from partner to internal staff, and transition-completion criteria are all part of the partner contract.

A capability initially sourced through buy mode — external specialists hired to seed an internal practice — frequently transitions into build mode as the internal practice grows its own pipeline from the seed. The buy-to-build transition requires management attention because the seed specialists are precisely the highest-retention-risk population (Article 11) during the transition period.

A capability initially sourced through build mode may transition into partner or borrow mode if the capability is declared non-strategic at a subsequent review. Decommissioning a build programme requires deliberate workforce transition — the apprenticeship cohort currently in development deserves pathway completion, and alternative opportunities for the graduating cohort must be designed.

Transitions are events that affect individual employees. They are communicated clearly, supported with transition infrastructure, and tracked in the mode-mix documentation.

The public-sector reference

Singapore’s SkillsFuture and National AI Strategy 2.0 workforce pillar, the UK NHS AI Lab’s workforce initiatives, the US DoD Replicator initiative, and Japan’s METI AI strategy are all public-sector analogues of hybrid sourcing at national scale.⁷⁸⁹¹⁰ In each case the public programmes combine build (national training, apprenticeships), buy (targeted recruitment), partner (university and industry partnerships), and borrow (short-term contracting). The enterprise-scale programme mirrors the national pattern structurally even though the populations and budgets are different.

Measurement of the mode mix

The mode mix is measurable at three levels. At the capability level, each capability has a documented target allocation and actual allocation — 60% build, 25% partner, 10% buy, 5% borrow is a specific plan; actuals are measured against it quarterly. At the programme level, aggregate allocations across capabilities indicate whether the organisation is drifting towards a single-mode default. At the outcome level, retention, capability maturity, and cost per capability-unit are tracked by mode to inform future mode-mix decisions.

Measurement infrastructure uses the HRIS (Workday, SAP SuccessFactors, Oracle HCM, ADP) for headcount-by-mode tracking, the LMS (Docebo, Cornerstone, Workday Learning, SAP SuccessFactors Learning, Open edX, Moodle) for build-mode capability milestones, contract-management and finance systems for partner and borrow cost tracking, and sentiment platforms (Qualtrics, CultureAmp, Peakon, Glint) for mode-specific employee experience signals. The measurement is multi-source and requires integration investment.

A common failure in measurement is to track only the headcount by mode without tracking outcomes. An organisation whose build-mode apprenticeship cohort retains at 90% after three years but whose buy-mode new hires retain at 60% has material information for the next mode-mix decision; an organisation that does not track retention by mode will make the next decision blind. Expert practice builds the outcome tracking before it is urgently needed.

Expert habit — documenting the mode-mix

A practitioner habit worth internalising is explicit documentation of the mode mix per capability. The documentation names the capability, the mode allocation (percentage or headcount per mode), the criterion rationale per mode, the governance requirements, and the review date. Mode-mix documentation lives as part of the transformation charter (Article 1) and is reviewed at the quarterly pipeline governance meeting (Article 6). Without documentation, mode choices drift towards defaults — typically buy for visible capability needs and borrow for urgent ones — and the hybrid discipline erodes.

Summary

Build, buy, partner, borrow are the four sourcing modes. Five criteria — capability criticality, speed, cost, retention risk, IP sensitivity — jointly determine the mode mix. Hybrid sourcing is the norm; single-mode sourcing is a mis-specification. Three biases — vendor preference, HR-process preference, risk-transfer attraction — distort mode choice and must be explicitly countered. Each mode carries specific governance requirements. Documentation of the mode mix per capability sustains discipline over time. Article 8 takes up the hiring-stage governance that the buy mode requires, with explicit attention to inclusive hiring and hiring-AI tool oversight.

Cross-references to the COMPEL Core Stream:

• EATF-Level-1/M1.6-Art03-Building-the-AI-Talent-Pipeline.md — talent pipeline foundation
• EATE-Level-3/M3.2-Art06-Talent-Strategy-at-Enterprise-Scale.md — enterprise talent strategy anchor
• EATF-Level-1/M1.2-Art17-AI-Operating-Model-Blueprint.md — operating model blueprint in which sourcing mix lives

Q-RUBRIC self-score: 90/100

© FlowRidge.io — COMPEL AI Transformation Methodology. All rights reserved.