The AI Reckoning Enterprise Tech Leaders Are Now Bracing For

Three executives from the technology firm Evocs told BW Businessworld, in the executives’ framing of the enterprise AI reckoning, that the test facing enterprise AI leadership in 2026 is not technical capability. Sumit Kumar, the firm’s Enterprise Architect, Tom Marcin, its VP and Senior Advisor, and Arvind Sharma, President of Corporate Strategy, framed the cloud migrations of the previous decade and the AI rollouts of this one as running into the same wall. The wall is the absence of assessment, governance, and people in the room before the technology arrives. The cost of skipping those steps is now arriving on enterprise balance sheets.

Cloud-first bets left expensive workloads, security gaps, and inefficient environments behind that now have to be retrofitted before AI can run safely on top of them. AI-first deployments, the executives warn, are repeating the same mistakes at higher speed. The difference is that regulators have begun to code the discipline into rules while the technology is still being rolled out.

The Cloud Migration Debt AI Will Inherit

Cloud migration promised agility, scale, and cost control. It delivered, in many enterprises, a different set of line items. Enterprises that moved first and audited later ended up with the three things the technology was supposed to prevent: expensive workloads, security gaps, and inefficient cloud environments that required significant optimisation after deployment. The cost of retrofitting cloud infrastructure is the line item the cloud decade rarely budgeted for.

The phrase ‘expensive workloads’ is shorthand for a deeper problem. Compliance requirements were not checked before workloads moved. The fit between the workload and the chosen cloud provider was not evaluated before contracts were signed. Cost optimisation, security architecture, and long-term operational fit were all retrofitted after the fact, when the bill and the breach risk had already arrived. This is the cloud migration debt that AI now inherits.

Secure Data First, Then Enable AI

The sequencing problem shows up clearly in the regulatory calendar. India’s central bank released a draft framework on June 24, 2026 that requires every regulated lender, NBFC, payments bank, cooperative, and asset reconstruction company to install an AI kill switch, put model risk at the board level, and disclose to customers when they are talking to a machine. The deadline for public comments runs through July 24, 2026. The draft assumes the regulated entity already knows what models it is running, where the data lives, and who owns each model, which is exactly the picture cloud-first bets rarely produced.

Our focus is simple: secure your data first, then enable AI. Enterprises need to understand where their data is going, how it is being used and whether it complies with regulatory requirements before adopting AI at scale.

Kumar, Enterprise Architect at Evocs, told BW Businessworld that the same logic applies outside banking. The RBI’s framework is the loudest version of a discipline that should govern every enterprise running AI in production. Most aren’t there yet.

The kill switch the RBI mandates is just the visible edge of a deeper requirement. The draft also demands that every regulated entity maintain an inventory of active, inactive, and decommissioned models and keep decommissioned ones on file for at least ten years. That kind of model hygiene assumes the enterprise already has a clear picture of its data estate. Secure your data first is the sequencing, and the June 2026 draft is now starting to enforce it, as detailed in the June 2026 AI kill switch mandate.

The People Problem Behind Failed Enterprise Transformations

The third leg of the Evocs argument is the one most AI strategies skip past. Tom Marcin, Evocs’s VP and Senior Advisor, said that large-scale digital programmes frequently fail despite significant budgets because organisations neglect the people responsible for implementing change. That sentence lands harder the longer one sits with it.

Marcin described the pattern as designing transformation programmes without involving employees in shaping them. People are not in the room when the future state is decided, and the change is pushed onto them after the fact. Resistance is what comes back, and the pattern repeats at every scale, from individual department rollouts to company-wide cloud migrations. The best transformations, in Marcin’s framing, are those where employees become part of the journey instead of simply being asked to accept the outcome.

The cost lands on the budget first and the operating model after. A transformation that cannot land on its intended users is one that pays twice: once for the rollout and once for the rollback. It is also where the cloud debt and the AI rollout connect most clearly, because AI tools that land on a workforce that was not consulted get treated as another layer pushed onto an already tired operating model.

Adoption in this context is a trust problem. The employees who have absorbed two or three rounds of transformation in the last decade recognise the pattern, and their resistance is the rational response to having their judgement about their own work consistently overridden. Marcin’s prescription is procedural: involve employees in shaping the future state from the start. The cost is the time it takes to convene the right people before the design is locked. The benefit is that the rollout does not have to be sold to a sceptical workforce after the fact, and the technical work supports the discipline.

That connects directly to the second philosophical position Marcin offered. In Marcin’s framing, AI is one input into the decision. The cost comes when AI is treated as the whole answer.

AI Should Strengthen Human Leadership

Marcin’s view is that AI should strengthen human leadership. Automation can improve efficiency and support decision-making, and the qualities he names as uniquely human are not soft skills in the consultant sense. They are the core of how an organisation actually decides. Empathy, judgement, and contextual understanding remain the inputs that AI outputs do not contain.

AI, in this framing, provides direction based on data. Leaders make the final call by balancing that direction with experience and emotional intelligence. The split between them is operational, with AI handling pattern recognition, scenario surfacing, and option generation, and leaders absorbing the ambiguity, accountability, and human consequence.

The risk on the other side is one the RBI’s draft names explicitly: automation bias, the tendency for employees to over-rely on AI outputs without applying their own judgement. The RBI treats this as a model risk on the same footing as hallucination and adversarial input. In an enterprise setting, automation bias shows up quietly: a junior analyst who has been told that an AI agent handles a particular reconciliation stops checking the agent’s work, or a risk officer who receives an AI-generated credit memo reads it as a starting point and then approves it on the same shape because the model said so. The error compounds because the audit trail still looks complete. The output was generated, reviewed, and approved, and the judgement that should have caught the failure was outsourced to the same system that produced it.

The risk lands harder in 2026 because the AI agents deployed into enterprise finance, customer service, and operations are doing more than the previous generation of automation did. They generate content, draft decisions, and act on data in ways that look like judgement from the outside. Marcin’s argument is that this is exactly the moment leadership cannot step back, because the data can be presented by AI while the decision stays one leadership has to absorb. The leadership gap grows wider when AI is treated as a substitute for the discipline.

The Customer-Value Filter for AI Investment

Arvind Sharma, President of Corporate Strategy at Evocs, said sustainable growth starts with understanding customer expectations instead of making assumptions about value. The phrasing matters because the trap it is describing is the one most AI strategies fall into. The technology is adopted because competitors are adopting it. The cost is justified because the alternative is falling behind. The speed of rollout becomes the metric, and the value the rollout produced goes unmeasured.

Value is what customers and key stakeholders believe in, not what organisations assume. Every technology investment should be evaluated on the value it creates and the return it delivers. AI should be adopted where it strengthens business outcomes, not simply because it is the latest trend.

Sharma, President of Corporate Strategy at Evocs, told BW Businessworld that the discipline is not theoretical. Most enterprises cannot answer the basic question of what an AI rollout has produced in customer or operational terms because the measurement was never built in. The same executives who authorised a cloud migration without proper assessment are now authorising AI pilots without a value baseline. The pattern repeats, and the bill compounds.

That is the discipline enterprise finance teams are now attempting to operationalise, with one industry analysis putting the potential cost reduction from an AI-first finance function at up to 40 percent when the redesign reaches the operating model. The full case is laid out in the AI-first finance transformation case, where the 40 percent figure is tied to end-to-end deployment. Pilot-stage token spend on its own does not arrive there, which is the gap most enterprise AI rollouts are running into today.

What Disciplined AI Leadership Looks Like in Practice

Taken together, the three Evocs executives describe a leadership discipline that enterprise AI rollouts rarely meet. The pieces are not new, and the combination is rarely assembled in the order it needs to be. The discipline is four moves stacked in sequence. Each one builds on the one before, and skipping any of them is what turns a deployment into a retrofit.

• Assessment before deployment. Identify compliance requirements, cost fit, security posture, and operational fit before workloads are moved or AI is enabled on top of them.
• Governance around data. Secure the data estate first, then enable AI on top of it, with clear ownership of every model and every data source.
• People in the room. Involve employees in shaping the future state of the work, not in approving a finished design after the fact.
• Value as the filter. Measure every AI investment against customer and operational outcomes before approving it.

Each of the four moves is missing from a different set of enterprise AI rollouts, and the cloud-first decade proved what skipping assessment costs: the retrofit bill arrives regardless. The same retrofit is now waiting for the AI-first decade. Assessment, governance, people, and value is a sequencing question, with the order mattering because the cost of skipping any one of them is paid by the next. The RBI’s June 2026 draft is the visible signal that regulators are now coding the discipline into rules, with the kill switch and the customer disclosure requirements framed as governance moves that assume the underlying pieces exist. Enterprises that skipped those pieces will find the new rules landing on top of the cloud debt they are already paying down.

The board-level signal matters because the discipline the executives describe cannot be implemented from below. Marcin’s people-in-the-room prescription requires the leadership in the room to invite the people, and Kumar’s secure-the-data-first sequencing requires the budget authority to delay AI deployment in favour of data estate work. All three moves assume a leadership layer willing to be the friction the rollout would otherwise avoid.

Sharma’s value filter requires the strategy function to veto initiatives that cannot articulate their business case. Marcin, Kumar, and Sharma told BW Businessworld that the choice sits with leadership. The cost of skipping the discipline shows up on the retrofit bill, and that bill arrives regardless of how well the technology itself performs. The technology is ready before the leadership is, in most enterprise rollouts, and the gap between the two is what the retrofit bill measures.

The cloud decade absorbed the cost in infrastructure spend and security retrofits. The AI decade is on track to absorb it in governance enforcement and adoption friction. The four disciplines the executives describe work as a sequencing problem. They are not a checklist, and the gap between the two framings is where most enterprise AI rollouts fall apart. The discipline is what most enterprises have to build first.

Frequently Asked Questions

Why do enterprise AI projects fail before they start?

Enterprise AI projects fail before they start when the assessment step is skipped, and the deployment decision is made before the cost, compliance, and security implications have been worked through. Tom Marcin of Evocs said that large-scale digital programmes frequently fail despite significant budgets because organisations neglect the people responsible for implementing change. The technology is rarely the bottleneck in these cases. The implementation is, and the implementation is a leadership decision.

What is the cloud migration debt that AI inherits?

Cloud migration debt is the retrofit cost that arrives after workloads are moved into a cloud environment without proper assessment. Sumit Kumar of Evocs described the result in three terms: expensive workloads, security gaps, and inefficient cloud environments that require significant optimisation after deployment. AI deployed on top of these estates inherits the same gaps at higher speed, which is why the AI decade starts where the cloud decade’s retrofit bill was already overdue.

How should enterprises sequence AI and data governance?

Enterprises should secure their data first and enable AI second, in that order and not the reverse. Kumar said the focus is simple: understand where the data is going, how it is being used, and whether it complies with regulatory requirements before adopting AI at scale. India’s central bank has effectively codified this sequencing into its June 24, 2026 draft framework for banks, which requires an AI kill switch, board-level model risk management, and customer disclosure for AI interactions. The deadline for public comments on the framework runs through July 24, 2026. Enterprises outside banking that ignore the sequencing are running the same risk under the same name.

Why does leadership matter more than technology in AI transformation?

In AI transformation, leadership shapes the gaps that cause rollouts to fail. Marcin’s position is that AI should strengthen human leadership, with empathy, judgement, and contextual understanding remaining uniquely human. The RBI’s draft framework warns of automation bias, the tendency to over-rely on AI outputs, as a model risk on the same footing as hallucination or adversarial input. A leadership gap dressed up as an efficiency gain is still a leadership gap.

How should AI investments be measured for value?

AI investments should be measured against customer and operational outcomes, with adoption speed and competitive parity ruled out as the right metrics. Arvind Sharma, President of Corporate Strategy at Evocs, said that value comes from what customers and key stakeholders believe in, while organisations’ assumptions about value often miss the mark. Every technology investment should be evaluated on the value it creates and the return it delivers, and AI is no exception to that rule.

What role do employees play in AI transformation?

Employees are the implementation layer of any enterprise AI rollout, and the rollout only works when that layer is part of the design from the start. Marcin said that the best transformations are those where employees become part of the journey instead of simply being asked to accept the outcome. Involving employees in shaping the future state is the procedural step that converts a rollout into an adoption. Skipping the step does not save time; it moves the cost from the design phase to the adoption phase, where it compounds. The pattern is the same one cloud-first rollouts ran into a decade ago, and the AI decade is now repeating it.