Enterprise AI programmes often fail through incompleteness. A model is selected without preparing data, a workflow is automated without redesigning roles, or a pilot succeeds while legacy systems prevent it from scaling. Infosys proposes a six-part implementation framework that makes those gaps easier to see.
Through its Topaz Fabric, the services company is working on AI implementations with 90% of its top 200 clients and reports more than 4,600 projects in progress. Its framework covers AI strategy and engineering, data for AI, process AI, legacy modernisation, physical AI, and AI trust.
The six areas are not a guaranteed recipe. They are a useful diagnostic: an organisation can ask which transformation has been funded, which has an owner, and which has quietly been assumed to solve itself.
Strategy, Data and Process Form the Core
Strategy and engineering connect models, agents, third-party tools, infrastructure, and business objectives. Without that common architecture, departments can accumulate separate assistants that duplicate cost and apply inconsistent controls.
Data for AI addresses structured and unstructured information, including what Infosys calls data fingerprinting and synthetic training-data services. The objective is to turn isolated assets into reliable model inputs. The important word is reliable. More data does not help when definitions conflict, permissions are unclear, or updates arrive too late for the decision.
Process AI focuses on placing agents inside work rather than beside it. This may require redesigning a workflow so humans and software exchange tasks efficiently. Leaders should map the complete outcome, identify where judgment is required, and measure whether automation removes effort or merely relocates it into checking and exception handling.
These three areas reinforce one another. Strategy selects the valuable process, data supplies trustworthy context, and workflow design turns capability into an operating result.
Legacy and Physical Systems Expand the Boundary
Legacy modernisation recognises that enterprise AI rarely begins on a clean platform. Agents can help analyse dependencies, interpret old code, and plan staged replacement. They may accelerate discovery, but modernisation still requires tests, domain knowledge, and careful sequencing so essential services continue to operate.
Physical AI extends the framework into sensors, devices, digital twins, robotics, autonomous systems, and edge computing. Here, an incorrect decision can affect equipment, inventory, safety, or production rather than only a document.
That demands collaboration across information technology, operational technology, engineering, and line-of-business teams. A model that performs well in a laboratory may face intermittent connectivity, noisy sensors, strict latency, and machines that cannot be paused for an easy software update.
The framework is valuable because it prevents leaders from treating physical operations as a simple extension of cloud AI. The risks, maintenance cycles, and evidence required are different.
Trust Must Accompany Every Other Workstream
Infosys places risk assessment, policy, security testing, ethics, guardrails, and lifecycle management under AI trust. Governance should not wait until the other five areas are complete. It determines which data can be used, which process can be automated, how a legacy-analysis agent is isolated, and when a physical system may act.
Clear accountability reduces regulatory and reputational risk. Each deployment needs a named owner, documented purpose, access limits, evaluation before release, monitoring after release, and a route for users to challenge or correct an outcome.
Leaders should also budget for workforce change. Process redesign requires training, while legacy and physical projects need specialised expertise. Those costs belong in the business case rather than appearing later as unexpected resistance.
Infosys’ framework presents AI implementation as organisational transformation, not software installation. Its strength is the insistence that six workstreams move together.
An enterprise does not need equal investment in every area for every project. It does need an explicit answer for each one. The missing transformation is often where an impressive pilot becomes an expensive exception—and where a realistic implementation plan can still recover it.
