In conversation with CIO&Leader, Eric Kaplan, CTO of AHEAD, explains why most enterprise AI initiatives stall before production and how organisations can move from isolated pilots to scalable, governed AI systems.
He outlines the engineering foundations, guardrails, and operating model shifts required to turn AI strategy into measurable business outcomes.
CIO&Leader: What are the top engineering or platform bottlenecks that stop enterprises from taking AI from strategy to production at scale?
Eric Kaplan: The biggest bottlenecks are not exotic AI problems. They are long-standing issues in data, legacy systems, and controls that AI makes harder to ignore. Data sits in too many places, with different definitions, so building a reliable dataset is slow and messy. Legacy platforms were never designed to share data cleanly or support rapid change, which is why integrating AI into real workflows takes longer than expected.
Security and compliance add another layer, because leaders need confidence that sensitive information will not leak or be used in the wrong way. And once AI moves into production, you are paying to run it continuously, so cost and usage control quickly become real constraints.
AI doesn’t fail because the models are weak. It fails because the foundations are.
CIO&Leader: Why do most AI proofs of concept fail to graduate into production systems, and how does AHEAD design differently to avoid this?
Eric Kaplan: Most AI proofs of concept fail because they are treated as side projects. A small team builds something to show that it can work, but it is not tied to a business outcome with a clear owner, success measures, and a plan to fit into day-to-day work. When teams try to scale, they then run into everything that was not accounted for – data access, approvals, security reviews, integration with existing systems, and change management.
At AHEAD, we start from the business problem, not the model. We define what success looks like in measurable terms and design the solution so it plugs into an existing process or workflow from the beginning. We also build with production in mind – including access controls, monitoring, and integration patterns – so we are not trying to bolt those on at the end.
CIO&Leader: What do “AI-ready” data and platform readiness look like in practical terms for large, regulated enterprises?
Eric Kaplan: In practical terms, AI-ready data means the organisation can use its data safely and consistently without rebuilding the basics for every use case. Data should have clear owners, common definitions, and basic quality checks so teams trust what they are using. Sensitive data needs to be well understood and protected, with policies enforced by systems and platforms, not left to individual judgement.
On the platform side, readiness means teams can deploy and update AI workloads safely and repeatably. They should be able to track what changed, roll back if needed, and quickly spot when something is not behaving as expected. For regulated enterprises, it also means strong observability and audit trails – the ability to show what data was used, which model or tool version produced an output, and who approved key changes. Our work with clients focuses heavily on these foundations, because without them AI remains a set of isolated experiments.
You can’t bolt on security, controls, and monitoring after the fact. Production has to be designed from day one.
CIO&Leader: As enterprises move from automation to agentic AI, what guardrails and governance are essential before allowing any level of autonomy?
Eric Kaplan: Once AI starts taking actions instead of just producing content, guardrails have to be very clear. The first step is to define the scope of authority – what the agent is allowed to do, where it can act, and what it must never touch. Access should follow least privilege, and higher risk actions should require explicit approval or a second factor.
You also need strong logging and traceability, so there is a reliable record of what the agent did, when, and based on which inputs. Many enterprises will be best served by rolling this out in stages, starting with well-bounded internal workflows such as IT operations or service management, where the impact is easier to control. At AHEAD, we help clients move through those stages with pattern-based designs that can be reused across multiple agents and domains.
CIO&Leader: How do you design human oversight, auditability, and accountability into production AI systems without slowing innovation?
Eric Kaplan: You bake these requirements into how the system runs, instead of layering manual reviews on top. Human oversight can be risk-based – low risk actions flow through automatically, while higher risk decisions are routed to people with clear response times and criteria. That keeps the review effort focused where it really matters.
Auditability becomes manageable when the system captures records automatically: the inputs, the actions taken, the outputs produced, and the changes made over time. Accountability improves when responsibilities are clearly split between business and technology owners – someone owns the outcome, someone owns the production system. In our experience, teams move faster when they have this structure, because they are innovating inside well-defined boundaries instead of renegotiating controls for every new idea.
AI is delivering real value in security when it augments analysts, not replaces them.
CIO&Leader: How does cloud modernisation change when it is driven by AI outcomes rather than infrastructure efficiency alone?
Eric Kaplan: Cloud modernisation becomes less about moving workloads for the sake of migration and more about building an environment where AI can run reliably and securely. AI work depends on access to data, consistent security controls, and predictable performance. For many regulated firms, this will continue to be a mix of data centres and public cloud.
What changes is the priority. Modernisation programmes shift towards making data easier to use across systems, strengthening controls around identity and access, improving reliability, and managing ongoing usage costs as AI scales. The aim is to support specific AI-driven business outcomes – like faster underwriting, better customer service, or more resilient operations – not only infrastructure efficiency metrics.
CIO&Leader: Where are enterprises seeing real ROI from AI in cybersecurity today, and where are expectations still outpacing reality?
Eric Kaplan: Security is one of the clearer areas where AI is already delivering value, because teams deal with huge volumes of signals and need help finding what matters. Organisations are seeing benefits in faster triage, better grouping of alerts, and quicker response in security operations, especially where AI is used to augment analysts rather than replace them.
Expectations get ahead of reality when enterprises assume AI will fix weak foundations – gaps in identity, incomplete asset visibility, poor data classification, or unpatched systems. There is also a trust and risk dimension. Leaders worry about exposing sensitive data to tools they do not fully control, and about what happens when an AI-assisted decision is wrong. The best results we see come from focused deployments tied to specific workflows, with clear guardrails, strong integration into existing tools, and human oversight built in.
The goal is measurable business impact, not infrastructure efficiency metrics.
CIO&Leader: How is AHEAD leveraging India’s talent and delivery ecosystem to accelerate AI from strategy to production for global clients?
Eric Kaplan: India is a key part of how we execute across the full path from AI strategy to working production systems. Moving AI into production requires more than data science – it needs engineering across data platforms, applications, infrastructure, operations, and security, along with the practical work of making solutions usable in day-to-day workflows.
Our teams in India give us scale and depth across these skills, which helps us move faster and support long-running production programmes, not just pilots. It also allows us to build reusable delivery patterns, reference architectures, and accelerators that can be applied across clients and industries. That combination of global strategy with India-based engineering and delivery has become a core strength in how AHEAD helps enterprises realise value from AI at scale.