Accelerating SAP S/4HANA Migration and Custom Code Documentation with Generative AI

axfood_/_harman · Other · 2025 · source Agentic: no · promotional bias: med · source-backed · extracted with prompt 0.5.2 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: SAP modernization

Harman needed to document 30,000 legacy ABAP custom objects for S/4HANA migration after decades of poor documentation.

Constraints:

30,000 custom objects
12-consultant manual plan
Tight migration timeline

Architecture

Harman's S/4HANA migration documentation system wraps AWS generative AI around a large legacy SAP estate. Custom ABAP objects from ECC landscapes are scanned by Bedrock and later Amazon Q Developer with Claude Sonnet; prompts produce business summaries, business-scenario descriptions, key functionality, and step-by-step technical documentation; consultants and functional teams use that structured output to map custom code to business processes, testing, simplification, and S/4 standard replacements. The architecture replaces a slow consultant-only documentation factory with repeatable AI documentation for 30,000 custom objects.

Diagram confidence: explicit (topology described by the source).

flowchart LR
  sap-ecc-landscapes[SAP ECC landscapes] --> custom-abap-object-corpus[Custom ABAP object corpus]
  custom-abap-object-corpus --> aws-bedrock-scanner[AWS Bedrock scanner]
  aws-bedrock-scanner --> prompt-iteration[Prompt iteration]
  prompt-iteration --> structured-documentation-output[Structured documentation output]
  custom-abap-object-corpus --> amazon-q-developer[Amazon Q Developer]
  amazon-q-developer --> claude-sonnet[Claude Sonnet]
  amazon-q-developer --> structured-documentation-output
  structured-documentation-output --> business-process-mapping[Business process mapping]
  business-process-mapping --> s-4hana-migration-planning[S/4HANA migration planning]

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`SAP ECC landscapes`	Legacy SAP environments containing custom ABAP objects and business-process customizations.	`Custom ABAP object corpus`
`Custom ABAP object corpus`	Large set of undocumented or poorly documented custom code requiring migration analysis.	`SAP ECC landscapes`, `AWS Bedrock scanner`
`AWS Bedrock scanner`	Initial generative AI path for scanning programs and producing structured documentation.	`Custom ABAP object corpus`, `Prompt iteration`
`Prompt iteration`	Collaboration with AWS to improve prompts and outputs beyond raw technical details.	`AWS Bedrock scanner`, `Structured documentation output`
`Amazon Q Developer`	Later code-documentation path using a newer Claude Sonnet version with better output.	`Custom ABAP object corpus`, `Structured documentation output`
`Claude Sonnet`	Model family used through AWS services for code-understanding documentation.	`Amazon Q Developer`
`Structured documentation output`	Business summary, business scenario, key functionality, and technical step-by-step documentation.	`Prompt iteration`, `Amazon Q Developer`, `Business process mapping`
`Business process mapping`	Maps custom code to L1-L4 processes, unit testing, SIT, and UAT needs.	`Structured documentation output`, `S/4HANA migration planning`
`S/4HANA migration planning`	Uses documentation to identify dependencies, testing scope, simplifications, standard replacements, and cleanup.	`Business process mapping`

Data flow

Harman starts with custom ABAP objects in ECC, scans them with AWS Bedrock and later Amazon Q Developer/Claude Sonnet, iterates prompts until the output includes both business-language summaries and technical documentation, then uses those documents to map code to business processes, test scopes, dependencies, S/4 simplifications, and replacement opportunities.

3 supporting quotes

“So we have 30,000 custom objects in one of the ECC landscapes and on second landscape we have nearly 20,000 CBOs.”transcript, implied

“So the initial output from the tool was not up to the expectation, even though it was doing automatically and it was only producing the technical details, like it scans the program and provided only that tables that program was utilizing and few technical details.”transcript, stated

“So the top four or five lines, if you look at, it provides a key summary, and then in the bottom it provides business scenario description of the program.”transcript, stated

Why this architecture

Harman chose AI-generated documentation because the manual process was too slow, costly, inconsistent, and hard for business users to consume. The migration required understanding dependencies, testing scope, simplification opportunities, and business-process mapping across a decades-old SAP estate, so a repeatable structured-output workflow was more useful than consultant prose in mixed formats.

“this is not working at all. And also, we were unable to meet the timelines for the subsequent tasks planned based on this documentation”transcript, stated

“And the outputs are very structured, easy to interpret, and very consistent. So this is all achieved with AWS Bedrock.”transcript, stated

Models used: Claude through Bedrock · Claude Sonnet through Amazon Q Developer

Scale

One ECC landscape had 30,000 custom objects and a second had nearly 20,000 CBOs.
“So we have 30,000 custom objects in one of the ECC landscapes and on second landscape we have nearly 20,000 CBOs.”transcript, stated
Manual documentation with 12 consultants was still estimated at 15 months.
“we ramped up to 12 consultants. Even with that there is minor improvement in the speed, still the estimation to achieve the goal was 15 months.”transcript, stated
Bedrock improved speed 6-7x, reducing the timeline from 15 months to 2 months and cost by more than 70%.
“So the speed has improved by six to seven times and reduce the total timeline from 15 months to 2 months. And the total cost reduction was more than 70%.”transcript, stated

Alternatives considered (roads not taken)

Manual documentation by consultants

Rejected because: Even after scaling to 12 consultants, the process was slow, expensive, inconsistent, and missed migration timelines.

“each person was documenting in different format, and it's not providing the value we were looking for. So these were the practical challenges.”transcript, stated

Accept initial Bedrock output as raw technical detail

Rejected because: The first output scanned code but did not provide the business-language documentation Harman needed.

“the initial output from the tool was not up to the expectation, even though it was doing automatically and it was only producing the technical details”transcript, stated

Document only selected custom objects

Rejected because: Harman concluded all 30,000 objects had to be documented for migration, testing, dependency, and simplification planning.

“So with all this clear understanding, we know that now we have to document all 30,000 customer objects, so there is no alternative for it. So initially we started documenting the custom code manually.”transcript, stated

Patterns

`batch-offline-processing` (migration)

The documentation process scanned a large static corpus of SAP custom objects.

“scan all 30,000 objects”transcript, stated

`structured-output-contract` (documentation)

The goal was repeatable documentation for mapping custom code into migration plans.

“And then in the bottom, as you see, so this is a detailed step by step documentation in technical perspective.”transcript, stated

Tradeoffs

How to document legacy SAP customizations for S/4 migration.

Chose Bedrock and Amazon Q Developer generated structured documentation over manual consultant documentation for every object.

Rationale: The AI workflow produced uniform business and technical documentation much faster and at lower cost than the scaled manual plan.

“because same prompt was used across all 30,000 customer objects. So the output was very structured and very consistent.”transcript, stated

How to make code documentation useful across roles.

Chose business-language summaries plus technical step-by-step details over technical-only scans of tables and code references.

Rationale: Harman needed both business stakeholders and ABAP developers to understand purpose, functionality, and implementation details.

“It is very difficult for a business person or a functional person to understand what is the purpose of the code.”transcript, stated

Gotchas (what bit them)

The existing custom-code estate had poor documentation.

Root cause: Decades of accumulated ABAP objects made manual migration analysis expensive and inconsistent.

Resolution: The team used Bedrock/Q Developer to generate documentation at scale.

“customizations in the system, and majority of those customizations have very poor documentation, and in some cases very”transcript, stated

Evaluation

The case reports timeline reduction against a consultant-based manual plan.

Metrics: 30,000 objects · 15 months · 2 months

“by six to seven times and reduce the total timeline from 15 months to 2 months. And the total cost reduction was more than 70%.”transcript, stated

Outcomes

Documentation timeline dropped from a 15-month consultant plan to about 2 months.

The workflow covered all 30,000 custom objects.

“we know that now we have to document all 30,000 customer objects, so there is no alternative for it.”transcript, stated