How organisations practically implement IT RUN vs BUILD cost allocation

Research Question

How have organisations actually implemented a working RUN vs BUILD IT cost allocation — specifically: how did they agree on what counts as an "application", how did they get consistent work-item tagging across teams, how did they establish a shared team taxonomy, who drove the change, and how did they make the business case for the investment required?

Findings

(Populated from §6 Synthesis above.)

Executive Summary

Organisations that achieve a working RUN vs BUILD cost allocation must build three artefacts in strict sequence — application register, team taxonomy, and work-item tagging — and each requires sustained governance that the majority of organisations underestimate. Joint CIO and CFO sponsorship is necessary, not preferable: the CIO controls data access and the CFO controls budget authority, and CIO-only or CFO-only programmes stall at the boundary of the other's domain. The primary failure mode is governance failure, not tooling failure: the GAO found that 18 of 26 US federal agencies (69%) failed to achieve reliable cost allocation after 8 years of TBM implementation under a regulatory mandate. Programmes that complete the full implementation deliver strong returns — Forrester documents 270% ROI over three years — but completing the implementation is the hard part, and the majority do not.

Key Findings

1. The application register is a non-negotiable prerequisite for both team taxonomy and work-item tagging; building either before the register is stable produces data that must be retroactively reclassified, causing model rebuilds and sponsor attrition. (high confidence)
2. An application is defined by four attributes — business capability delivery, independent ownership, defined support model, and independent lifecycle — with boundary disputes (middleware, vendor bundles) resolved in governance workshops and documented in the register with a rationale note. (high confidence)
3. KPMG's documented 18-month phased model sequences implementation as governance/application register (months 1–3), data quality remediation (months 4–6), taxonomy deployment (months 7–9), model building (months 10–12), and expansion (months 13–18); compressing or parallelising phases is a documented failure cause. (high confidence)
4. Joint CIO and CFO sponsorship is the consistent success pattern across three independent sources (KPMG, EY, Serviceware); CIO-only programmes stall at budget authority boundaries, CFO-only programmes stall at data access boundaries. (high confidence)
5. Work-item tagging fails because the individual bearing the tagging cost receives none of the benefit; the two evidence-supported mitigations are reducing friction through AI pre-fill and simplified categories, and making team-level cost output visible to the tagging teams so they see their own spend. (medium confidence)
6. Degraded tagging does not produce obvious errors; it produces a systematic upward drift in RUN percentage and suspiciously stable category distributions that only become visible through scheduled audits comparing tagging rates and distributions against baselines. (medium confidence)
7. The GAO-25-106488 audit (July 2025) found that 18 of 26 US federal agencies failed to achieve reliable cost allocation after 8 years of TBM implementation under an OMB mandate, with federal investment ranging from $1.5M to $28.9M per agency; this establishes that mandate without funded enforcement and sustained sponsorship does not produce compliance. (high confidence)
8. ISG identifies People (siloed mindsets and turf protection) as the top failure dimension ahead of Data and Technology; leadership turnover after programme initiation is the most cited single root cause of stalled programmes, as replacement sponsors rarely inherit the original governance commitment. (high confidence)
9. Programmes that complete the implementation deliver Forrester-documented 270% ROI over three years with payback under six months, primarily from redundant licence elimination, contract renegotiation, and reduced finance reconciliation overhead. (medium confidence — Forrester TEI accessed via secondary citation)
10. Cost transparency alone does not produce optimisation; the "now what?" stall is escaped by wiring the model output into a specific governance decision process — annual IT portfolio review, vendor renegotiation cycle, or application rationalisation programme — before the model is declared complete. (medium confidence)
11. The board argument must lead with cost visibility deficits and benchmark-identified inefficiency, with the RUN/BUILD ratio as a measurement instrument rather than the goal; boards respond to savings and strategic reinvestment, not to abstract ratio improvement. (medium confidence)
12. The People-dimension failures (resistance, siloed mindsets) rank above Data and Technology failures in ISG's framework, meaning that change management investment — not additional tooling capability — is typically what determines whether the programme completes. (high confidence)

Assumptions

• Assumption: Private-sector TBM failure rates are lower than the 69% federal rate. Justification: Clearer commercial incentives and faster decision cycles; no equivalent private-sector audit data exists for comparison. Federal rate is used as a lower bound, not a direct estimate.
• Assumption: Automated NLP classification of work items achieves 70–85% accuracy at category level. Justification: Extrapolated from general NLP text classification benchmarks and practitioner accounts; no primary source measuring RUN/BUILD classification accuracy directly was found.
• Assumption: KPMG's 18-month phased model is broadly representative of industry practice rather than KPMG-specific. Justification: The phase logic is independently consistent with TBM Council guidance and Praecipio practitioner steps developed by different organisations.
• Assumption: Private-sector implementation costs are $2M–$10M for mid-to-large organisations. Justification: Derived from federal data ($1.5M–$28.9M per GAO) with a discount for federal procurement overhead; no independent private-sector cost benchmarks found.

Analysis

RUN/BUILD allocation programmes fail at the rate they do because they are governance programmes that are typically scoped, funded, and executed as technology programmes. The tooling is mature; Apptio, Serviceware, and comparable platforms can process the required data once it is available in the required form. The failure is in sustaining the governance structures — consistent application register, team taxonomy, and tagging compliance — that give the tooling something accurate to compute over.

The dependency chain (register → taxonomy → tagging) is the central practical insight. Each artefact requires the prior one as its scope definition. Organisations that compress or parallelise the sequence consistently encounter rework — tagging data mapped to an unstable application list, team taxonomies that predate the final application scope — that erodes sponsor confidence and causes programmes to stall. The 18-month timeline reflects the minimum calendar time needed to complete the prerequisites, not the complexity of the tooling itself.

The GAO data provides the strongest empirical grounding. Federal agencies had what private organisations typically lack: a regulatory mandate, OMB oversight, and multi-year budget commitments. They still failed at 69%. The delta between the federal failure rate and the implied private-sector rate represents the value of commercial incentives — but the federal evidence establishes that even strong external pressure does not substitute for internal governance commitment.

The sponsorship finding has a direct structural explanation rooted in organisational authority: TBM programmes require authority in two separate organisational domains. The CIO controls the data systems and the data access permissions that make the model possible; the CFO controls the budget processes and the financial governance that make the model actionable. A programme with single-domain sponsorship will encounter the boundary of the sponsor's authority and stall there — a structural constraint of how IT and finance authority are divided in most organisations, not a cultural or political problem.

The vendor case studies (Praecipio retail client: $2M/year savings; financial services client: 90% forecasting accuracy improvement) represent the outcome distribution for programmes that complete. The GAO data represents the full distribution including non-completions. A programme that completes the implementation can expect strong returns; the primary risk is not-completing, and the mitigant for that risk is governance design, not tooling selection.

Risks, Gaps, and Uncertainties

• PDF inaccessibility: GAO-25-106488, KPMG Empowering TBM Program, and ISG Multi-Dimensional TBM Framework were not directly readable during this investigation (binary/encrypted content returned by fetch tools). Claims derived from these sources are based on web search summaries. Quantitative claims (GAO failure counts, investment ranges) cited in multiple independent summaries are rated high confidence; interpretive claims (ISG dimension rankings) are rated medium confidence.
• Private-sector failure rate unknown: No audit comparable to GAO-25-106488 exists for private organisations. The 69% figure cannot be applied directly. The actual private-sector failure distribution is unknown; vendor case study publication bias inflates the apparent success rate.
• Automated classification accuracy unverified: The 70–85% NLP accuracy figure is an extrapolation from general classification benchmarks. Actual accuracy on short ticket descriptions with ambiguous RUN/BUILD categorisation is not empirically documented and may be lower.
• Recovery from stalled programmes: No documented case studies were found of organisations that stalled a TBM programme and successfully restarted it. The failure-to-recovery pathway is a genuine gap in the evidence.
• CapEx/OpEx interaction: How BUILD work that is capitalised under GAAP/IFRS interacts with the TBM model's RUN/BUILD classification was not investigated in depth; this interaction may create accounting constraints that affect how the model is structured in practice.

Open Questions

• What is the minimum viable TBM scope — the subset of the full taxonomy that delivers enough value in under six months to justify continued investment and can be completed before sponsor tenure risk materialises? (Potential backlog item; medium priority)
• How do organisations with >50% cloud workloads adapt the application register and team taxonomy when infrastructure costs are usage-based rather than fixed allocation? (Potential backlog item; medium priority)
• Are there documented private-sector cases of stalled TBM programmes that were successfully restarted, and what governance or structural changes enabled the restart? (Potential backlog item; low priority)
• How does the RUN/BUILD allocation interact with CapEx/OpEx capitalisation accounting — when BUILD work is capitalised, does the TBM model track the asset separately from ongoing run costs, and how is the amortisation stream handled? (Potential backlog item; medium priority)

sources

• [x] TBM Council practitioner community — implementation experience, not vendor materials (tbmcouncil.org forums/publications)
• [ ] Apptio case studies for National Grid, Liberty Mutual, UPMC, Hermes Parcel — read specifically for implementation mechanics and challenges, not outcomes
• [ ] Gartner "Run-Grow-Transform" framework (G00308477) — application definition criteria and governance guidance
• [ ] Mik Kersten "Project to Product" (2018) — Flow Framework® value stream and team definition, and the practical challenge of mapping work to value streams
• [ ] IT Financial Management (ITFM) practitioner blogs, conference talks (HDI, itSMF, FinOps Foundation) — implementation war stories
• [x] "Application portfolio management" literature — how APM programmes define and govern the application register (LeanIX, SAP Help Portal)
• [ ] Google Scholar: "IT cost allocation implementation challenges", "application portfolio rationalization governance", "TBM implementation lessons learned" — filtered post-2018
• [ ] LinkedIn / practitioner accounts of TBM, FinOps, or ITFM programme implementations — specifically seeking accounts of what did not work as expected
• [x] GAO-25-106488 (2025) — US federal agency TBM implementation audit
• [x] KPMG Empowering TBM Program — 18-month phased implementation guidance
• [x] Praecipio Consulting — practical Apptio TBM implementation steps and case data
• [x] EY — ITFM business case and sponsorship patterns
• [x] Serviceware — ITFM vs TBM sponsor alignment patterns
• [x] Rego Consulting — TBM business case methodology
• [x] ISG Multi-Dimensional TBM Framework — failure mode taxonomy