Modernizing Legacy Systems with AI: A Practical Playbook for Indian Enterprises

The Indian enterprise IT estate in 2026 is older than the LinkedIn slides suggest. Behind every AI strategy deck is a real environment that includes a Cobol-era core banking system, an SAP installation deployed in 2008, a Tally instance that finance refuses to give up, a custom Java application written by a team that has long since left, and a dozen Excel sheets that constitute the actual source of truth for processes nobody has documented in a decade.

This is the legacy. It is not a problem to solve in a quarter. It is the operating environment.

For most of the last fifteen years, "modernisation" has meant one of three things: a full rewrite (slow, expensive, usually scope-creeping), a migration to a cloud-hosted SaaS equivalent (fast in marketing terms, slow and disruptive in practice), or a strangler-fig refactor that has been ongoing for so long that nobody remembers the original deadline.

AI does not eliminate these options. But it changes the economics of specific modernisation patterns enough that the playbook is worth rewriting.

Where AI Actually Helps with Legacy Modernisation

Not everywhere. The pitches that promise AI will rewrite your mainframe automatically should be treated with the suspicion they deserve. The places where AI genuinely shifts the modernisation economics:

1. Documentation generation from undocumented code. LLMs can now produce useful documentation for legacy codebases that have no current documentation. Not perfect documentation, but a working starting point. For a Java codebase last documented in 2014, this changes the cost of understanding the system before changing it.

2. Code translation between legacy and modern stacks. Cobol to Java, VB6 to .NET, PL/SQL to Postgres. The translations are not push-button, but the AI-assisted version of these projects is meaningfully faster than the human-only version. Validation remains essential.

3. Document intelligence on legacy unstructured data. A lot of legacy "data" is actually scanned PDFs, faxes, and paper-form records sitting in shared drives. Document AI converts this into structured data without modernising the underlying systems that consume it.

4. API generation in front of legacy systems. Wrapping a legacy system with an AI-assisted API layer lets modern applications consume legacy capabilities without modifying the legacy system itself. Wrap, don't rewrite.

5. Workflow automation that bridges legacy and modern. AI agents that operate across legacy GUIs, modern APIs, and human-in-the-loop steps. RPA, but meaningfully smarter.

6. Testing and regression generation for legacy systems. AI-generated test cases over legacy code with poor test coverage. Not a substitute for human review, but a meaningful accelerator.

What AI does not do: rewrite the system for you, eliminate the need to understand the business logic, or substitute for the organisational change management that any real modernisation requires.

The Indian Legacy Profile, Honestly

Indian enterprises have a distinct legacy profile that international modernisation playbooks tend to miss.

Mid-size SAP and Oracle EBS estates are everywhere. Most large Indian enterprises have an ERP installation that is at least a decade old, deeply customised, and tied to processes that have grown around the customisations. Changing the system means changing the processes means changing how people work — which is the hard part.

Custom Java and .NET applications are the silent majority. Built in the IT services boom of the 2000s, often by teams now dispersed, with documentation in unknown states. Every Indian enterprise has at least one application where the institutional knowledge lives entirely in one person's head.

Tally remains the financial source of truth in surprising places. Including in companies that have invested heavily in enterprise software. Finance teams are loyal to what they trust and know, and Tally has earned that loyalty.

Excel and email are the integration layer. The actual flow of data between systems in many Indian enterprises is mediated by manual export, manipulation, and re-import. This shadow integration layer is invisible to the architecture diagram and essential to the business.

Paper and PDF persist at scale. KYC documents, contracts, statutory filings, employment records. The volume of paper-anchored business processes in Indian enterprises is one of the most underestimated facts in modernisation planning.

A modernisation strategy that does not acknowledge these realities will fail in predictable ways.

A Five-Step Modernisation Playbook

Step 1: Map the actual estate, including the Excel layer.

Before any modernisation decision, document what is actually running — not just the official applications. The Excel sheets. The email-based workflows. The Access databases. The shared drives. This step is unglamorous and frequently skipped, with predictable consequences.

AI helps here by accelerating documentation generation from code. It does not help with the Excel and email layer, which requires interviews.

Step 2: Classify by value, risk, and friction.

Each system gets sorted along three axes. Value: how much business activity depends on it. Risk: what happens if it breaks or is breached. Friction: how painful it is to work around its limitations today. The intersection of high friction and either high value or high risk is where modernisation effort should focus first.

Step 3: Choose the modernisation pattern per system.

Not every system needs the same treatment:

Replace — a modern system fully replaces the legacy. Highest cost, highest risk, highest payoff when justified.

Wrap — leave the legacy in place, build an AI-assisted API and agent layer around it. Lower cost, lower risk, preserves the institutional knowledge embedded in the legacy.

Refactor in place — gradual rewriting within the legacy stack. AI helps with code translation and test generation. Multi-year, with steady incremental value.

Retire — some systems no longer carry their own weight. Identifying these honestly is one of the most underrated modernisation activities.

Step 4: Bridge unstructured data with document intelligence.

For the paper-and-PDF layer that sits alongside legacy systems, deploy document intelligence as an independent stream. This does not modernise the underlying systems, but it converts historically unstructured data into a format that modern applications can use.

This is where AI delivers the most reliable, fastest-payback modernisation value in Indian enterprises — OCR, semantic search, and structured extraction across thousands of documents, deployed on-premise to keep sensitive content local.

Step 5: Build the integration layer.

The thing that ties wrapped legacy, modernised replacements, and the document intelligence layer together is an integration layer that can operate across legacy GUIs, modern APIs, and human-in-the-loop steps. This becomes the durable platform investment. Underlying systems change. The integration layer carries the institutional logic forward.

Common Modernisation Failure Modes

Big-bang replacement. A multi-year project to replace a legacy system entirely in one go. Misses value milestones, accumulates risk, and frequently misses the original requirements by the time it ships.

Modernisation without business process review. Replacing the system without examining whether the underlying process should be changed. The new system inherits all the legacy's awkwardness plus the change management cost.

Ignoring the Excel and email layer. Modernising the official systems while ignoring the shadow IT that actually mediates the data flow. The result is a modernised system surrounded by the same unmodernised workflow.

AI as magic wand. Believing the vendor pitch that AI can auto-rewrite the legacy. AI helps, sometimes meaningfully. It does not substitute for the engineering and business work required.

No retirement plan. Modernising systems that should have been retired. The CFO eventually notices that the modernisation budget is larger than the business value of the modernised system.

The Compliance Angle

For Indian enterprises, modernisation increasingly intersects with compliance. The DPDP Act, sectoral regulations, and customer contractual expectations all push for clear data location, retention, and deletion controls. Many legacy systems were not designed for any of this.

Two practical consequences:

Modernisation creates the opportunity to improve compliance posture. A wrapped or replaced system can be brought into the enterprise's modern IAM, audit, retention, and deletion frameworks in ways the legacy never could.

On-premise document intelligence is increasingly the right answer for the paper-and-PDF layer. Putting sensitive document data through cloud OCR services may itself be a compliance issue. On-premise document intelligence converts the data without sending it outside the organisation.

A Realistic Eighteen-Month Roadmap

Q1 — Mapping and classification. Complete inventory including the Excel and email layer. Classification by value, risk, and friction.

Q2 — Quick wins via document intelligence. On-premise document AI deployed for the highest-volume paper-anchored workflow. ROI demonstrated early.

Q3 — Wrap and integration. API and agent layer around two priority legacy systems. Modern applications start consuming legacy capabilities cleanly.

Q4 — First replacement. One legacy system replaced where the business case is strong.

Q5 — Refactor in place. Ongoing AI-assisted code modernisation within the largest remaining legacy stack.

Q6 — Retirement and consolidation. Systems identified for retirement actually retired. Estate documentation updated.

The shape matters more than the specific quarter assignments. The principle is: value early through wraps and document intelligence, replacement only where justified, refactor as ongoing work, retire honestly.

Conclusion

Legacy modernisation in Indian enterprises is not a project. It is an ongoing discipline that will continue for the next decade. AI does not change that fundamentally, but it changes the economics of specific high-cost modernisation activities enough that the playbook needs to be rewritten.

The enterprises that get modernisation right will use AI surgically — documentation generation, code translation, document intelligence, API generation, agent-driven integration. The ones that get it wrong will either ignore AI's genuine help or treat it as a magic wand.

To discuss how AI-assisted legacy modernisation can work for your organisation, reach out to admin@setidure.com.