Why IT and AI Keep Clashing Inside Large Companies
IT and AI teams are often grouped under a single “technology” label, yet they approach their work with opposite assumptions. IT prioritizes control, standardization and auditability; AI teams prioritize experimentation, flexibility and speed. When those priorities collide, the two functions can undermine each other. A new Harvard Business Review article, based on advisory work with three companies, details the pattern and the fixes that emerged.
The three cases—diversified conglomerate Baringa Group, Southern Bank with 36,000 employees, and telecom PacificTel with 9,200 staff—show different roots of conflict. At Baringa, the fight was over what counts as data. At Southern Bank, it was about competence, governance and accountability. At PacificTel, it was whether aging IT infrastructure could support AI at all.
Each company resolved the standoff structurally rather than rhetorically. Baringa expanded its data inventory to treat call transcripts, contracts, emails and workflow logs as strategic assets. Southern Bank split the Head of AI and Head of IT roles and created a taxonomy separating analytical from generative AI. PacificTel split AI projects into low-foundation, data-dependent and foundation-critical categories and linked the AI roadmap to its IT consolidation roadmap. The common thread: both functions kept distinct responsibilities, but were measured on shared business outcomes and given structures that made collaboration possible.
How Baringa, Southern Bank and PacificTel Reshaped the Relationship
Baringa Group: Making Governance Relevant to AI's Raw Material
The conflict at Baringa Group was definitional. IT teams thought of data as tables, columns and charts; AI teams relied on customer interactions, call centre transcripts, service agreements and behavioural insights. The company's response was to expand its enterprise data inventory to include unstructured sources and classify them as strategic assets, with executive-level governance extending existing standards to those inputs. That move kept IT's control frameworks intact while legitimising the data AI teams were already using. The rationale: the definition of data is a governance question, not a philosophical one. Excluding the inputs AI actually uses from formal governance guarantees the two teams will work on different assumptions.
Southern Bank: Trading a Competence Fight for Clear Ownership
The standoff at Southern Bank was framed as a dispute about competence. IT pointed to sound governance and auditability; AI pointed to competitors already using AI for fraud detection, personalised offers and faster model development. Southern Bank separated the Head of AI and Head of IT roles, created a mandatory taxonomy separating analytical AI (machine learning, predictive modelling, decisioning) from generative AI (text, summaries, conversation), and required project intake forms and risk templates to specify which branch applied. It also rewrote role architecture so data engineers, data scientists, machine learning engineers and IT had clear handover points. This is persuasive because a demand to be “brought inside governance earlier” is easier to satisfy when both sides agree on what kind of AI is being discussed.
PacificTel: Matching AI Ambition to IT Reality
PacificTel's conflict was about foundations. IT was years into consolidating legacy systems and flat-file integrations; AI teams wanted to deploy customer sentiment analysis and network fault prediction immediately. The company categorised AI use cases as low-foundation (document drafting, knowledge tools), data-dependent (sentiment analysis, treated as insight signals rather than automated decisions) and foundation-critical (fault detection, churn intervention, automated responses). It then tied the AI roadmap to the IT consolidation roadmap. The structural point: foundation-critical AI cannot succeed on fragmented, late-arriving data, and treating all AI as one program invites failures that each side blames on the other.
One Mechanism, Three Variations
Across the three cases, the lesson is consistent: companies resolved conflict not by merging IT and AI or choosing one side, but by making each side responsible for the conditions under which the other could succeed. Baringa gave both teams a shared business outcome and joint KPIs; Southern Bank made accountability visible through role separation and taxonomy; PacificTel made sequencing explicit. Because the HBR article is based on anonymised advisory work and reports no quantitative outcomes, these should be read as management patterns rather than proven results.
Steps for Executives Trying to End the IT-AI Standoff
Executives who oversee both functions can apply the following moves from the three cases:
- Expand the data inventory before debating definitions. Baringa formally included call transcripts, service notes, contracts, emails and workflow logs as strategic assets. If governance only covers structured databases, AI teams will keep working outside official systems.
- Separate roles by accountability, not by reporting convenience. Southern Bank split the Head of AI and Head of IT positions after combining them blurred responsibility. Give IT infrastructure, security, compliance and resilience; give AI model development, experimentation and deployment.
- Adopt a taxonomy that differentiates analytical from generative AI. Southern Bank requires explainability and auditability for analytical AI used in fraud, credit and pricing, while allowing lighter controls for generative work such as drafting and summarisation.
- Segment AI projects by foundation dependency. PacificTel separated low-foundation, data-dependent and foundation-critical use cases. Do not start predictive fault detection or automated customer responses until data pipelines, identifiers and real-time monitoring are stable.
- Link the AI roadmap to the IT consolidation roadmap and share the KPIs. PacificTel connected the two plans, and Baringa measured IT and AI teams on the same commercial results, so both had a structural incentive to move together.
Risk & Opportunity Assessment
| Commercial Risk | Medium | Unresolved IT-AI conflict delays AI deployment and creates duplicated tools and parallel efforts, as seen at Southern Bank, potentially wasting investment in models that never reach production. |
| Competitive Risk | Medium | AI leaders at Southern Bank and PacificTel said competitors were already using AI for fraud detection, personalised offers and faster model development; internal friction slows the same capabilities. |
| Regulatory Risk | Medium | In regulated environments such as Baringa and Southern Bank, AI must be auditable and explainable; if IT and AI cannot agree on governance early, compliance risks are discovered only after deployment. |
| Reputation Risk | Low | Internal team conflict is mostly invisible externally, but PacificTel's experience shows AI models that degrade on fragmented data can weaken customer-facing services and internal confidence in AI. |
| Technology Disruption | Medium | Generative and agentic AI depend on unstructured, conversational data that does not fit legacy IT systems, forcing companies like Baringa and PacificTel to redefine data assets and rebuild foundations. |
| Commercial Opportunity | Medium | Baringa integrated AI directly into customer service, claims, advisory and logistics workflows, and PacificTel combined sentiment analysis with IT's hard data; resolving conflict can turn AI into day-to-day decision support. |
Comments 0