Skip to content

2026 04 22 historical technology adoption enterprise ai capability building

Jump to bottom
github-actions[bot] edited this page Apr 30, 2026 · 2 revisions

Historical technology adoption patterns as analogues for enterprise Artificial Intelligence capability building

Research Question

What can organisations learn from retrospectives of prior technology introductions, specifically personal computing, Enterprise Resource Planning (ERP), cloud computing, Robotic Process Automation (RPA), and electronic trading systems in financial services, about the organisational, governance, and operating model failures that prevented individual productivity gains from translating to enterprise-level value, and what patterns of successful capability building are observable in hindsight for enterprise Artificial Intelligence (AI)?

Scope

In scope:

  • Retrospective evidence on organisational outcomes from personal computing, ERP, cloud computing, RPA, and electronic trading systems in financial services.
  • Failure modes where local productivity gains did not scale to enterprise value, for example governance fragmentation, capability bottlenecks, and weak operating model redesign.
  • Patterns of successful capability building, leadership model, platform model, standards, funding, talent, and change management, that are reusable for enterprise AI.
  • Longitudinal or multi-wave studies that assess whether repeat failure modes are structural versus situational.

Out of scope:

  • Vendor product comparisons or tool feature benchmarking for current AI platforms.
  • Small-team tactical implementation guides that do not address enterprise-level operating model design.
  • Country-specific legal deep dives unless they materially explain success or failure patterns across waves.

Constraints: (time, source types, access)

  • Prioritise sources published between 2019 and 2026, while including foundational historical studies where needed for context.
  • Include both practitioner evidence and peer-reviewed organisational change literature.
  • Note paywalled or limited-access evidence explicitly and triangulate with publicly accessible corroboration.

Context

[fact; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] Enterprise AI adoption is already broad and governance pressure is rising, which means capability-building choices now determine whether local gains scale safely into enterprise value.

[inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] The relevant design problem is therefore not whether AI can raise local productivity, but whether organisations build the shared standards, platforms, training, and control loops that convert faster local output into durable enterprise outcomes.

Approach

  1. Map each technology wave, personal computing, ERP, cloud, RPA, and electronic trading, to its common failure mechanisms and identify where the same organisations repeated or avoided those patterns.
  2. Extract and compare successful capability-building patterns across waves, governance, operating model, platform ownership, funding model, and talent strategy, then assess transferability to enterprise AI.
  3. Synthesize structural versus situational drivers using longitudinal studies and financial-services retrospectives, and produce a practical pattern library for enterprise AI operating model design.

Sources

Research Skill Output

§0 Initialise

  • [fact; source: https://github.com/davidamitchell/Research/blob/b1dc13aaa535079319294329f03d3f37a927514d/Research/in-progress/2026-04-22-historical-technology-adoption-enterprise-ai-capability-building.md] The research question asks which recurring organisational, governance, and operating-model failures caused prior technology waves to stall at local productivity gains, and which hindsight capability-building patterns are reusable for enterprise AI.
  • [fact; source: https://github.com/davidamitchell/Research/blob/b1dc13aaa535079319294329f03d3f37a927514d/Research/in-progress/2026-04-22-historical-technology-adoption-enterprise-ai-capability-building.md] The scope includes five technology waves, personal computing, ERP, cloud, RPA, and electronic trading in financial services, and excludes vendor product comparison, tactical small-team implementation advice, and country-specific legal deep dives that do not materially change the cross-wave pattern.
  • [fact; source: https://github.com/davidamitchell/Research/blob/b1dc13aaa535079319294329f03d3f37a927514d/Research/in-progress/2026-04-22-historical-technology-adoption-enterprise-ai-capability-building.md] The output format is a structured research record with labelled facts, inferences, and assumptions, followed by Findings that answer the question directly and translate the evidence into a reusable enterprise AI pattern library.
  • [fact; source: https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-platform-operating-models.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-03-12-ai-force-multiplier-ambition-expansion.md] Prior completed repository work already established that enterprise AI value depends on shared foundational capabilities, that AI operating models work best when the shared enterprise layer for policy, access, evaluation, observability, and configuration is centralised, and that personal-computing-style productivity gains depend on organisational restructuring rather than tool rollout alone.
  • [inference; source: https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-platform-operating-models.md; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] This item therefore extends prior work from present-tense AI capability design into historical pattern recognition: if the same organisational failures repeat across earlier waves, then enterprise AI capability building should be designed as a structural response, not as a one-off maturity exercise.

§1 Question Decomposition

  • Branch A - personal computing and Information Technology (IT)
    • A1. Did personal-computing and broader IT productivity gains require organisational complements rather than hardware or software adoption alone?
    • A2. Which complements, process redesign, training, decentralisation, and management redesign, recur in the evidence?
  • Branch B - ERP
    • B1. Which failure factors repeatedly prevented ERP implementations from translating local system use into enterprise value?
    • B2. Which capability-building patterns, executive sponsorship, process standardisation, training, and change management, repeatedly improved outcomes?
  • Branch C - cloud
    • C1. Why did many enterprises fail to realise cloud value after migration?
    • C2. Which operating-model, governance, talent, and platform changes separated better outcomes from weaker ones?
  • Branch D - RPA
    • D1. Why did RPA produce many pilots but limited scaled enterprise value?
    • D2. Which governance, process-selection, and support-model capabilities mattered for scale?
  • Branch E - electronic trading in financial services
    • E1. Which central controls were required once trading automation became faster and more consequential?
    • E2. What does this imply about shared-enterprise-rail design for enterprise AI in regulated settings?
  • Branch F - cross-wave synthesis
    • F1. Which recurring failure modes appear in at least three waves and therefore look structural rather than situational?
    • F2. Which recurring success patterns appear transferable to enterprise AI capability building?
  • Branch G - enterprise AI
    • G1. Which capabilities should be shared enterprise rails?
    • G2. Which responsibilities should stay federated near business domains and use cases?

§2 Investigation

Source audit and accessibility notes

  • [fact; source: https://www.forrester.com/blogs/robotic-process-automation-rpa-is-at-a-tipping-point/; https://www.businesswire.com/news/home/20200212005226/en/RPA-Reality-Check-New-Forrester-Research-Identifies-Barriers-to-RPA-Scalability] Search query Forrester RPA Is At A Tipping Point produced no accessible official Forrester page in this environment because the seeded Forrester URL returned 404, so replacement evidence uses the accessible BusinessWire summary and independent RPA scaling studies.
  • [fact; source: https://www.gartner.com/en/newsroom/press-releases/2022-05-24-gartner-says-worldwide-robotic-process-automation-software-revenue-grew-19-5-percent-in-2021; https://www.itvoice.in/gartner-says-worldwide-rpa-software-revenue-to-reach-2-9-billion-in-2022] Search query Gartner 2022 RPA revenue 19.5 percent found the seeded Gartner page blocked with 403, so the headline revenue figures are treated as accessible secondary context, not as sole support for any core conclusion.
  • [fact; source: https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/clouds-trillion-dollar-prize-is-up-for-grabs; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.pwc.com/gx/en/services/consulting/cloud-transformation/reaching-full-cloud-potential.html] Search query cloud trillion-dollar prize up for grabs McKinsey found the seeded McKinsey page inaccessible from this environment, so cloud claims rely on accessible AWS and PwC material plus search-discovered summaries only where they align.
  • [fact; source: https://www.standishgroup.com/sample_research_files/CHAOSReport2015-Final.pdf] The seeded Standish sample PDF returned 404 and is not used for downstream support.
  • [fact; source: https://doi.org/10.1145/103162.103188; https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download] Search query Brynjolfsson productivity paradox information technology 1993 showed the seeded Digital Object Identifier (DOI) returning 404 in this environment, so the investigation uses the accessible Brynjolfsson and Hitt working-paper version and related open-access complements literature.
  • [fact; source: https://www.bankofengland.co.uk/quarterly-bulletin/2014/q4/the-uks-automated-trading-landscape; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf] Search query site:bankofengland.co.uk the UK's automated trading landscape pdf identified the seeded Bank of England page as moved or inaccessible, so electronic-trading claims rely on accessible Financial Conduct Authority and Bank for International Settlements sources rather than the broken seed alone.
  • [fact; source: https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109; https://link.springer.com/article/10.1007/s41870-020-00502-z] Search query ERP critical failure factors systematic literature review located the Journal of Business and Technology review, but the site timed out from this environment, so direct ERP support uses the accessible Springer review and search-discovered summaries only where they agree.
  • [fact; source: https://doi.org/10.1145/103162.103188; https://www.forrester.com/blogs/robotic-process-automation-rpa-is-at-a-tipping-point/; https://www.gartner.com/en/newsroom/press-releases/2022-05-24-gartner-says-worldwide-robotic-process-automation-software-revenue-grew-19-5-percent-in-2021; https://www.bankofengland.co.uk/quarterly-bulletin/2014/q4/the-uks-automated-trading-landscape] No additional paper, Digital Object Identifier (DOI), or arXiv source gaps were found beyond the broken seed DOI and moved seed pages already recorded above.

A. Personal computing and broader IT

  • [fact; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download] Brynjolfsson and Hitt argue that the business value of computers is increasingly determined by managers' ability to invent new processes, procedures, and organisational structures rather than by hardware or software acquisition alone.
  • [fact; source: https://sloanreview.mit.edu/article/the-transforming-power-of-complementary-assets/; https://publications.jrc.ec.europa.eu/repository/bitstream/JRC75890/lfna25542enn.pdf] MIT Sloan and the European Commission Joint Research Centre both describe Information and Communications Technology (ICT) value as dependent on complementary assets such as process change, employee training, and organisational capital rather than on technical deployment alone.
  • [fact; source: https://epub.ub.uni-muenchen.de/11507/1/Mahr_Kretschmer_2010_LMU.pdf; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-03-12-ai-force-multiplier-ambition-expansion.md] Complementarity research and the prior repository synthesis both point to decentralisation and decision-right redesign as important conditions for converting computing capacity into measured productivity.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://sloanreview.mit.edu/article/the-transforming-power-of-complementary-assets/; https://publications.jrc.ec.europa.eu/repository/bitstream/JRC75890/lfna25542enn.pdf] The personal-computing wave did not fail because the tools were useless; it underdelivered initially because enterprise complements, workflow redesign, skills, and management practice, accumulated more slowly than local tool adoption.

B. ERP

  • [fact; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109] Recent ERP systematic reviews converge on recurring failure factors that include weak top-management support, inadequate training, mismatch between the system and business strategy, poor project or process management, and user resistance.
  • [fact; source: https://link.springer.com/article/10.1007/s41870-020-00502-z] The Springer meta-synthesis explicitly frames ERP failure as an organisational challenge shaped by stakeholder participation, structural readiness, human capital, culture, and risk-management quality, not as a software-selection problem alone.
  • [inference; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109] ERP success patterns therefore centre on executive sponsorship, process standardisation before heavy customisation, user training, and change management that aligns the system to enterprise operating discipline rather than to legacy local variation.

C. Cloud

  • [fact; source: https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html] AWS states that cloud investments should be tied to measurable business outcomes and that cloud transformation often requires changes to the operating model, organisation structure, roles, responsibilities, and governance controls.
  • [fact; source: https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.pwc.com/gx/en/services/consulting/cloud-transformation/reaching-full-cloud-potential.html] Accessible cloud guidance emphasises leadership alignment, readiness assessment, skills-gap closure, governance-at-scale, and business-technology alignment as prerequisites for realising cloud value.
  • [inference; source: https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.pwc.com/gx/en/services/consulting/cloud-transformation/reaching-full-cloud-potential.html; https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/clouds-trillion-dollar-prize-is-up-for-grabs] The cloud wave underdelivered when firms treated migration as a technical relocation exercise; winners treated it as a platform, governance, and talent redesign program tied to business outcomes.

D. RPA

  • [fact; source: https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/] Verint's summary of Deloitte's 2017 RPA research reports that only 3% of organisations had scaled RPA to 50 or more robots, which indicates that pilot success and enterprise scale were very different milestones.
  • [fact; source: https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] Accessible RPA evidence repeatedly points to the same scale constraints: fragmented ownership, unstable or changing processes, weak collaboration between business units and IT, insufficient governance, and high maintenance load when automations are attached to brittle workflows.
  • [fact; source: https://www.businesswire.com/news/home/20200212005226/en/RPA-Reality-Check-New-Forrester-Research-Identifies-Barriers-to-RPA-Scalability; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] The accessible Forrester summary and the Poussa thesis both frame scale as a governance and operating-model problem, not merely a tooling problem, because bot resiliency, process maturity, and a support model become limiting factors once automation moves beyond a few isolated cases.
  • [inference; source: https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] RPA showed a classic enterprise-adoption failure mode: firms captured local labour-saving gains quickly, but without a Center of Excellence (CoE), process discipline, and lifecycle support, they accumulated brittle automations rather than an enterprise capability.

E. Electronic trading in financial services

  • [fact; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations] The Financial Conduct Authority says algorithmic trading firms are a major source of market liquidity but face inherent risks, so firms' controls and oversight functions, including compliance and risk management, must keep pace with growing market speed and technological complexity.
  • [fact; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.fca.org.uk/firms/senior-managers-certification-regime] The Financial Conduct Authority review organises good practice around governance, development and testing, risk controls, and market-abuse surveillance, and notes that some firms require all material algorithm changes to be approved by a Senior Management Function (SMF), a role defined within the Financial Conduct Authority's Senior Managers and Certification Regime.
  • [fact; source: https://www.bis.org/publ/mktc13.pdf; https://eur-lex.europa.eu/eli/dir/2014/65/oj] The Bank for International Settlements reports that the foreign-exchange market has no market-wide circuit breakers or kill switches, so each provider and user of execution algorithms must maintain adequate safeguards, and it links broader execution-algorithm use to explicit governance oversight and risk-control requirements under Markets in Financial Instruments Directive II (MiFID II), the European Union directive governing investment-services markets.
  • [inference; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf] Electronic trading is the clearest financial-services analogue for enterprise AI because local speed created value only after firms built central testing, monitoring, change-control, and senior-accountability mechanisms that could absorb faster decision cycles without destabilising the system.

F. Cross-wave synthesis and transfer to enterprise AI

  • [fact; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] The 2025 DevOps Research and Assessment (DORA) result is explicit: AI does not fix a team, it amplifies what is already there, and the recommended starting points are policy clarity, internal context, foundational practices, safety nets, internal platform investment, and end-user focus.
  • [fact; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] DORA also reports near-universal AI use at work, more than 80% perceived productivity benefit, a continuing negative relationship to delivery stability, and a direct correlation between internal-platform quality and the ability to unlock AI value.
  • [fact; source: https://hai.stanford.edu/ai-index/2025-ai-index-report] The AI Index reports that organisational AI use rose to 78% in 2024, AI-related incidents are increasing, standardised responsible-AI evaluations remain rare, and AI-related regulation intensified sharply in 2024.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] The repeated pattern across all five waves is best explained as predominantly structural, although technology maturity, vendor-market evolution, and sector-specific regulation affect how severely it appears in each wave.
  • [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-platform-operating-models.md] The transferable enterprise AI pattern is therefore a shared enterprise layer for policy, evaluation, internal knowledge context, safety nets, standards, platform tooling, and talent development, combined with federated domain teams that apply those rails to specific workflows and customer problems.

§3 Reasoning

  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations] The strongest observed pattern is cross-wave convergence on the same organisational complements: executive sponsorship, process discipline, training, governance, shared platforms or support functions, and explicit control loops.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.bis.org/publ/mktc13.pdf; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] Because the same complement bundle reappears in five substantively different technology waves, consumer personal computing, enterprise systems, infrastructure platforms, back-office automation, and market automation, the best-supported interpretation is that the pattern is predominantly structural, although each wave is also shaped by technology maturity and context-specific regulation.
  • [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] Enterprise AI belongs in the same class because the current evidence already shows the same mismatch between near-universal local use and incomplete enterprise control, evaluation, and operating-model maturity.
  • [assumption; source: https://www.forrester.com/blogs/robotic-process-automation-rpa-is-at-a-tipping-point/; https://www.gartner.com/en/newsroom/press-releases/2022-05-24-gartner-says-worldwide-robotic-process-automation-software-revenue-grew-19-5-percent-in-2021; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] The inaccessible Forrester and Gartner seed pages do not materially reverse the RPA pattern used here. Justification: the final conclusion rests on multiple independent sources, including the Deloitte statistic cited by Verint and the Poussa thesis, rather than on either blocked page alone.

§4 Consistency Check

  • [fact; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://publications.jrc.ec.europa.eu/repository/bitstream/JRC75890/lfna25542enn.pdf] There is no contradiction between the early productivity paradox and later firm-level gains, because the complements literature explains the time lag between tool adoption and measured enterprise outcomes.
  • [fact; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] There is also no contradiction between local productivity gains and enterprise underdelivery, because DORA and RPA scaling research both show that faster local output can coexist with unstable downstream systems and weak organisational absorption capacity.
  • [inference; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/] Confidence should be highest on the qualitative failure-mode pattern and lower on any single headline percentage from ERP or RPA analyst material, because several seed analyst pages were blocked or moved.
  • [inference; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf] The electronic-trading analogy is stronger for governance and control design than for user-adoption behaviour, but that difference does not weaken the main conclusion because the transfer claim is about enterprise control architecture, not about desk-level change management alone.

§5 Depth and Breadth Expansion

  • [inference; source: https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] Technologies that accelerate change volume reward standard interfaces, testing, monitoring, and strong feedback loops, so the technical lens reinforces the case for shared enterprise rails before broad AI rollout.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-03-02-transaction-costs.md] The economic lens shows that value migrates from the tool to the organisational complements, because once the tool becomes easier to buy, the scarce asset becomes the shared operating capability that prevents each team from reinventing governance, integration, and support.
  • [inference; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] The behavioural lens shows that user resistance, incentive misalignment, and fear of disruption recur across ERP, RPA, and AI, which means capability building must include training, role clarity, and visible leadership backing.
  • [inference; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://hai.stanford.edu/ai-index/2025-ai-index-report] The regulatory lens strengthens the enterprise-AI transfer claim because the control burden is increasing while standardised responsible-AI evaluation remains immature, which raises the value of central policy, testing, and auditability capabilities.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.bis.org/publ/mktc13.pdf] The historical lens shows that the most reusable lesson is not a specific org chart but a sequence: first define the shared enterprise layer, then standardise high-value processes, then scale local adoption through federated teams operating on common rails.

§6 Synthesis

Executive summary:

  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] Enterprise AI capability building succeeds when firms treat AI as an operating-model and shared-enterprise-rails program rather than as a distributed tool rollout, because every relevant prior wave delivered enterprise value only after governance, process redesign, training, and platform standards caught up with local adoption.
  • [fact; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://www.bis.org/publ/mktc13.pdf] Personal computing, ERP, cloud, RPA, and electronic trading all show the same sequence: local productivity gains appear first, then enterprise value depends on complementary organisational capabilities that are slower and harder to build than the technology itself.
  • [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-platform-operating-models.md] The best-supported transfer pattern for enterprise AI is therefore specific: centralise policy, evaluation, internal context, safety nets, and platform ownership, while federating workflow redesign and use-case application near domains.
  • [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] The main uncertainty is not whether AI can generate local productivity gains, but how quickly each enterprise can build those complements before adoption outpaces control and creates duplicated governance, support, and integration friction.

Key findings:

  1. [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://sloanreview.mit.edu/article/the-transforming-power-of-complementary-assets/; https://publications.jrc.ec.europa.eu/repository/bitstream/JRC75890/lfna25542enn.pdf; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] The personal-computing and broader IT evidence shows that enterprise value came from complementary organisational change, process redesign, training, and decision-right redesign, not from workstation deployment alone. Confidence: high
  2. [inference; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109] ERP retrospectives repeatedly identify executive sponsorship, process fit, user training, change management, and stakeholder participation as the dominant determinants of value realisation, which suggests ERP underdelivery was mainly organisational rather than technical. Confidence: high
  3. [inference; source: https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.pwc.com/gx/en/services/consulting/cloud-transformation/reaching-full-cloud-potential.html] Cloud transformations underdelivered when firms treated migration as infrastructure relocation, because realised value depended on business-outcome alignment, governance-at-scale, skill development, and platform-style operating models. Confidence: high
  4. [inference; source: https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.businesswire.com/news/home/20200212005226/en/RPA-Reality-Check-New-Forrester-Research-Identifies-Barriers-to-RPA-Scalability] RPA produced visible pilot wins but weak enterprise scale because brittle processes, fragmented ownership, support gaps, and missing Center of Excellence mechanisms turned automations into maintenance burdens instead of reusable capability. Confidence: high
  5. [inference; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf] Electronic trading in financial services scaled only with central testing, monitoring, risk controls, change approval, and senior-accountability structures, which suggests that faster automated decision cycles increase the need for shared control mechanisms rather than reducing it. Confidence: high
  6. [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations] The repeated failure modes across all five waves are best explained as predominantly structural, although technology maturity, vendor-market evolution, and regulation affect their severity in each wave. Confidence: medium
  7. [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md] For enterprise AI, the best-supported reusable capability is a shared enterprise layer for policy, evaluation, internal context, platform tooling, and talent systems, while use-case delivery should remain federated near business domains. Confidence: medium
  8. [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] Because AI adoption is already widespread while incidents and regulation are rising, organisational absorption capacity is a likely current bottleneck, which makes capability building more urgent than additional tool proliferation. Confidence: medium

Evidence map:

Claim Source Confidence Notes
[inference] IT value required organisational complements, not deployment alone. https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://sloanreview.mit.edu/article/the-transforming-power-of-complementary-assets/; https://publications.jrc.ec.europa.eu/repository/bitstream/JRC75890/lfna25542enn.pdf high Personal-computing analogue for AI productivity gaps.
[inference] ERP outcomes were dominated by sponsorship, process fit, training, and change management. https://link.springer.com/article/10.1007/s41870-020-00502-z; https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109 high Qualitative pattern is stronger than any single failure-rate statistic.
[inference] Cloud value depended on operating-model, governance, and talent change. https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.pwc.com/gx/en/services/consulting/cloud-transformation/reaching-full-cloud-potential.html high Migration alone was insufficient.
[inference] RPA scaled poorly without CoE, governance, process maturity, and support. https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.businesswire.com/news/home/20200212005226/en/RPA-Reality-Check-New-Forrester-Research-Identifies-Barriers-to-RPA-Scalability high Pilot wins did not equal enterprise capability.
[inference] Electronic trading required central testing, monitoring, and accountability controls. https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf high Strong financial-services analogue for AI control design.
[inference] Cross-wave failures are best explained as predominantly structural, although context affects severity. https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations medium Same complement bundle appears across five unlike waves, but each wave also has its own maturity and regulatory conditions.
[inference] Enterprise AI should centralise shared rails and federate domain use. https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-platform-operating-models.md medium Direct transfer from historical pattern plus current AI evidence, but public evidence is thinner on exact operating-model shape than on the underlying capability need.
[inference] Organisational absorption capacity is a likely current AI bottleneck. https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report medium Adoption is already broad while governance maturity lags, but the evidence does not prove a single universal bottleneck.

Assumptions:

  • [assumption; source: https://www.forrester.com/blogs/robotic-process-automation-rpa-is-at-a-tipping-point/; https://www.gartner.com/en/newsroom/press-releases/2022-05-24-gartner-says-worldwide-robotic-process-automation-software-revenue-grew-19-5-percent-in-2021; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] The blocked Forrester and Gartner seed pages do not contain materially different headline claims from the accessible summaries used here. Justification: the final conclusion is triangulated with independent sources and does not depend on those pages alone.

Analysis:

  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations] The evidence was weighted primarily by recurrence across unlike contexts, because repeated appearance of the same complement bundle across five waves is more decision-useful than any single market-size statistic or vendor claim.
  • [inference; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] Financial-services trading evidence was weighted heavily on control design because it shows what happens when automated decisions become fast, opaque, and systemically consequential, which is closer to enterprise AI governance than personal-computing rollout is.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/] The main trade-off is central control versus local speed, but the historical record suggests that shared rails improve enterprise speed over time because they reduce duplicated governance, duplicated support, and duplicated integration effort.

Risks, gaps, uncertainties:

  • [fact; source: https://www.forrester.com/blogs/robotic-process-automation-rpa-is-at-a-tipping-point/; https://www.gartner.com/en/newsroom/press-releases/2022-05-24-gartner-says-worldwide-robotic-process-automation-software-revenue-grew-19-5-percent-in-2021; https://www.bankofengland.co.uk/quarterly-bulletin/2014/q4/the-uks-automated-trading-landscape; https://doi.org/10.1145/103162.103188] Several seeded pages were blocked, moved, or broken in this environment, so the evidence base is strongest on qualitative mechanisms and somewhat weaker on some original analyst phrasing or legacy primary links.
  • [inference; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/] ERP and RPA percentage claims vary materially by sample and definition, so the review should treat exact failure-rate numbers as less stable than the recurring organisational failure pattern.
  • [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] The AI transfer claim is strongest for capability-building logic and control design, but weaker for exact org-chart prescriptions, because current public evidence still describes practices more often than fully disclosed operating models.

Open questions:

  1. Which leading financial-services firms have published enough detail to compare centralised versus federated enterprise AI shared-enterprise layers directly rather than by analogy?
  2. How should enterprises sequence capability building when they already have substantial cloud and data-platform maturity but weak AI evaluation maturity?
  3. Which measures best detect when AI adoption is creating duplicated governance, support, and integration friction faster than the organisation is building shared rails?

§7 Recursive Review

  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] The final conclusion is justified by accessible evidence from every wave under study and does not depend on any blocked seed page as sole support.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] The remaining uncertainty is about prevalence and sequencing, not about the direction of the core historical pattern.

Findings

Executive Summary

[inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] Enterprise AI capability building succeeds when firms treat AI as a shared organisational capability program, not as a collection of local productivity tools, because every relevant prior wave produced enterprise value only after governance, process redesign, training, and platform standards caught up with adoption.

[fact; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://www.bis.org/publ/mktc13.pdf] Personal computing, ERP, cloud, RPA, and electronic trading each show the same historical sequence: local productivity gains appear before enterprise value, and the gap is closed by complementary organisational capabilities rather than by more technology alone.

[inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-platform-operating-models.md] The transferable pattern for enterprise AI is therefore to centralise policy, evaluation, internal context, safety nets, and platform ownership while federating workflow redesign and domain-specific application near business units.

[inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] The main uncertainty is not whether AI can create local gains, but whether each enterprise can build those complements before adoption outpaces control and creates governance, quality, and support debt.

Key Findings

  1. [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://sloanreview.mit.edu/article/the-transforming-power-of-complementary-assets/; https://publications.jrc.ec.europa.eu/repository/bitstream/JRC75890/lfna25542enn.pdf; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] High confidence: The personal-computing and broader Information Technology (IT) evidence shows that enterprise value came from complementary organisational change, process redesign, training, and decision-right redesign, not from workstation deployment alone.
  2. [fact; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109] High confidence: ERP retrospectives repeatedly identify executive sponsorship, process fit, user training, change management, and stakeholder participation as the dominant determinants of value realisation, which means ERP underdelivery was mainly organisational rather than technical.
  3. [fact; source: https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.pwc.com/gx/en/services/consulting/cloud-transformation/reaching-full-cloud-potential.html] High confidence: Cloud transformations underdelivered when firms treated migration as infrastructure relocation, because realised value depended on business-outcome alignment, governance-at-scale, skill development, and platform-style operating models.
  4. [fact; source: https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.businesswire.com/news/home/20200212005226/en/RPA-Reality-Check-New-Forrester-Research-Identifies-Barriers-to-RPA-Scalability] High confidence: RPA produced visible pilot wins but weak enterprise scale because brittle processes, fragmented ownership, support gaps, and missing Center of Excellence (CoE) mechanisms turned automations into maintenance burdens instead of reusable capability.
  5. [fact; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf] High confidence: Electronic trading in financial services scaled only with central testing, monitoring, risk controls, change approval, and senior-accountability structures, which shows that faster automated decision cycles increase the need for shared control mechanisms rather than reducing it.
  6. [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations] Medium confidence: The repeated failure modes across all five waves are best explained as predominantly structural, although technology maturity, vendor-market evolution, and regulation affect their severity in each wave.
  7. [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md] Medium confidence: For enterprise AI, the best-supported reusable capability is a shared enterprise layer for policy, evaluation, internal context, platform tooling, and talent systems, while use-case delivery should remain federated near business domains.
  8. [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] Medium confidence: Because AI adoption is already widespread while incidents and regulation are rising, organisational absorption capacity is a likely current bottleneck, which makes capability building more urgent than additional tool proliferation.

Evidence Map

Claim Source Confidence Notes
[inference] IT value required organisational complements, not deployment alone. https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://sloanreview.mit.edu/article/the-transforming-power-of-complementary-assets/; https://publications.jrc.ec.europa.eu/repository/bitstream/JRC75890/lfna25542enn.pdf high Historical analogue for AI productivity gaps.
[inference] ERP outcomes were dominated by sponsorship, process fit, training, and change management. https://link.springer.com/article/10.1007/s41870-020-00502-z; https://jbt.sljol.info/articles/10.4038/jbt.v7i1.109 high Organisational pattern is stronger than any single ERP failure-rate claim.
[inference] Cloud value depended on operating-model, governance, and talent change. https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.pwc.com/gx/en/services/consulting/cloud-transformation/reaching-full-cloud-potential.html high Migration without redesign underdelivered.
[inference] RPA scaled poorly without CoE, governance, process maturity, and support. https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.businesswire.com/news/home/20200212005226/en/RPA-Reality-Check-New-Forrester-Research-Identifies-Barriers-to-RPA-Scalability high Pilot wins did not equal enterprise capability.
[inference] Electronic trading required central testing, monitoring, and accountability controls. https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf high Strong financial-services analogue for AI control design.
[inference] Cross-wave failures are best explained as predominantly structural, although context affects severity. https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations medium Same complement bundle appears across five unlike waves, but each wave also has its own maturity and regulatory conditions.
[inference] Enterprise AI should centralise shared rails and federate domain use. https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-capability-model.md; https://github.com/davidamitchell/Research/blob/main/Research/completed/2026-04-22-enterprise-ai-platform-operating-models.md medium Direct transfer from history plus current AI evidence, but public evidence is thinner on exact operating-model shape than on the underlying capability need.
[inference] Organisational absorption capacity is a likely current AI bottleneck. https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report medium Adoption is already broad while governance maturity lags, but the evidence does not prove a single universal bottleneck.

Assumptions

  • [assumption; source: https://www.forrester.com/blogs/robotic-process-automation-rpa-is-at-a-tipping-point/; https://www.gartner.com/en/newsroom/press-releases/2022-05-24-gartner-says-worldwide-robotic-process-automation-software-revenue-grew-19-5-percent-in-2021; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1] The blocked Forrester and Gartner seed pages do not contain materially different headline claims from the accessible summaries used here. Justification: the final conclusion is triangulated with independent sources and does not depend on those pages alone.

Analysis

  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://link.springer.com/article/10.1007/s41870-020-00502-z; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://lutpub.lut.fi/bitstream/handle/10024/161151/Master%27sThesis_Henri_Poussa_Final.pdf?sequence=1; https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations] I weighted recurrence across unlike contexts more heavily than any isolated statistic, because repeated appearance of the same complement bundle across five waves is more decision-useful than any single market-size or failure-rate estimate.
  • [inference; source: https://www.fca.org.uk/publications/multi-firm-reviews/algorithmic-trading-controls-high-level-observations; https://www.bis.org/publ/mktc13.pdf; https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report] I weighted electronic-trading evidence heavily on control design because it shows what happens when automated decisions become fast, opaque, and systemically consequential, which is the closest regulated analogue to enterprise AI governance.
  • [inference; source: https://repository.upenn.edu/bitstreams/efe7c90d-e2ef-4d5d-83e5-f222a4d6cd96/download; https://docs.aws.amazon.com/prescriptive-guidance/latest/strategy-cloud-transformation/resolving-cloud-transformation-challenges.html; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/] The main trade-off is central control versus local speed, but the historical record suggests that shared rails improve enterprise speed over time because they reduce duplicated governance, duplicated support, and duplicated integration work.
  • [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] I resolved competing interpretations in favour of capability-building urgency because current AI evidence already shows both broad local use and incomplete enterprise governance, which matches the early stage of prior waves more closely than a mature equilibrium.

Risks, Gaps, and Uncertainties

  • [fact; source: https://www.forrester.com/blogs/robotic-process-automation-rpa-is-at-a-tipping-point/; https://www.gartner.com/en/newsroom/press-releases/2022-05-24-gartner-says-worldwide-robotic-process-automation-software-revenue-grew-19-5-percent-in-2021; https://www.bankofengland.co.uk/quarterly-bulletin/2014/q4/the-uks-automated-trading-landscape; https://doi.org/10.1145/103162.103188] Several seeded pages were blocked, moved, or broken in this environment, so the evidence base is strongest on qualitative mechanisms and somewhat weaker on original analyst phrasing or legacy link continuity.
  • [inference; source: https://link.springer.com/article/10.1007/s41870-020-00502-z; https://www.verint.com/blog/how-to-scale-rpa-beyond-a-pilot/] ERP and RPA prevalence figures vary materially by sample and definition, so exact percentages should be treated cautiously even though the organisational failure pattern is well supported.
  • [inference; source: https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report; https://hai.stanford.edu/ai-index/2025-ai-index-report] The transfer claim to AI is strongest for capability-building logic and control design, but weaker for exact org-chart prescriptions, because firms still disclose practices more often than full operating-model detail.

Open Questions

  1. Which financial-services firms have published enough detail to compare centralised versus federated enterprise AI shared-enterprise layers directly rather than by analogy?
  2. How should enterprises sequence capability building when they already have substantial cloud and data-platform maturity but weak AI evaluation maturity?
  3. Which operational measures best detect when AI adoption is creating duplicated governance, support, and integration friction faster than the organisation is building shared rails?

Output

Related

Navigation

Home

By Tag

bureaucracy

change-management

coase

constraint-analysis

control-model

decision-rights

delegation

delivery-risk

demand-segmentation

enterprise

exception-handling

execution

flow

flow-design

flow-metrics

governance

governance-patterns

incentives

instability

institutional-economics

leading-indicators

operating-model

organisation

organisational-design

queue-design

queueing

regulated-enterprise

routing

throughput

throughput-risk

transaction-costs

triage

williamson

Clone this wiki locally