CASE STUDY/GLOBAL CEMENT MANUFACTURER

Up to 80% faster SAP reporting through autonomous agentic AI

AI agents autonomously optimized high-impact SAP ABAP reports for one of India's largest cement manufacturers, eliminating database bottlenecks, reducing runtime by up to 80%, and restoring production stability without altering business logic or output.

Faster Runtime

58-80%

MANUAL ABAP EFFORT

Zero

DELIVERY VELOCITY

1 Dayvs 1 month

FUNCTIONAL PARITY

100%

THE CHALLENGE

Client's critical SAP reports were timing out or running for minutes, creating serious performance bottlenecks and business risk.

Our client operates one of India's most complex SAP landscapes, with hundreds of custom ABAP programs spanning sales and distribution, finance, procurement, and logistics. Several programs, including e-invoice and e-bill detail reporting, customer credit risk and sales reporting, and invoice exception reporting, had become serious operational bottlenecks.

Each program shared the same root cause: ABAP written for smaller data volumes and older database architectures. SELECT SINGLE statements fired inside tight loops over thousands of records. Repeated calls to BAPI_USER_GET_DETAIL fetched the same user master data on every iteration. Linear table scans ran where indexed binary search was possible. Temporary tables accumulated memory without being released.

With S/4HANA migration in progress and the 2027 ECC sunset approaching, these programs needed to be HANA-ready, and manual remediation across all three was estimated to take months of scarce ABAP developer time.

“The optimized programs delivered exactly the same results but in a fraction of the time. What would have been a high-risk manual exercise was completed safely without disrupting production workflows.”

- SAP Technical Lead, Global Cement Manufacturer

THE APPROACH

Automated ABAP Optimization with Zero Functional Risk

Sookti AI deployed autonomous agents to perform end-to-end code optimization with the objective of improving runtime performance, database efficiency, and memory usage, without altering existing business logic or functional output.

The scope of work included identifying and remediating performance bottlenecks such as database access inside loops, inefficient internal table access patterns, redundant function module calls, and delayed memory cleanup.

Through deep static analysis, the agents detected performance anti-patterns across all three programs and systematically transformed each one without manual production intervention or introducing regression risk.

Each program progressed independently through analysis, structural refactoring, automated regression validation, and documentation generation. Functional parity was verified against the original source clone prior to delivery.

Across all programs, the agents:

Replaced every SELECT SINGLE inside a loop with FOR ALL ENTRIES bulk fetch

Introduced SORT + BINARY SEARCH on all preloaded internal tables for O(log n) access

Preloaded lookup data (user master via USR21/ADRP, customer KNA1, vendor BSEG, risk class) outside loops

Eliminated redundant BAPI_USER_GET_DETAIL calls by caching user and email data upfront

Built unique partner and segment lists before function module calls to eliminate duplicate overhead

Introduced parallel cursor logic to process grouped records without full table scans

Applied DELETE ADJACENT DUPLICATES after every sort to ensure correct binary search behaviour

Released all temporary internal tables with explicit FREE statements to reduce peak memory

Preserved every functional condition, business rule, and output structure exactly

All changes were self-documented, regression-tested automatically, and validated to ensure identical results to the legacy programs.

RESULTS (PERFORMANCE GAINS)

Data-driven improvements measured on live DEV 300 production workloads. Results below represent a subset of optimized programs from a broader SAP performance initiative.

PROGRAM / REPORT

BEFORE (LEGACY)

AFTER (SOOKTI AI)

IMPROVEMENT

ZRSD_EBILL_EINV_DETAILS (ZSD1062)

42s

17.6s

58%faster

zhrsd947 (ZSD206)

21.6s

4.3s

80%faster

ZSDR_EXC_INVOICE_CDS (ZSD149)

37.4s

15.5s

59%faster

PROGRAM / REPORT

ZRSD_EBILL_EINV_DETAILS (ZSD1062)

BEFORE (LEGACY)

42s

AFTER (SOOKTI AI)

17.6s

IMPROVEMENT

58%faster

PROGRAM / REPORT

zhrsd947 (ZSD206)

BEFORE (LEGACY)

21.6s

AFTER (SOOKTI AI)

4.3s

IMPROVEMENT

80%faster

PROGRAM / REPORT

ZSDR_EXC_INVOICE_CDS (ZSD149)

BEFORE (LEGACY)

37.4s

AFTER (SOOKTI AI)

15.5s

IMPROVEMENT

59%faster

Validation Methodology

For accurate benchmarking, source clones of the original programs were executed under identical DEV 300 workload conditions. Runtime measurements were captured using SAP Runtime Analysis (SAT), and optimized versions were validated against their respective source clones to confirm 100% output parity.

All optimizations were strictly technical enhancements and no changes were made to business logic, calculations, or functional behavior. The deployment introduced zero regressions and required no manual ABAP intervention.

SAP Runtime Analysis (SAT) DEV 300

Program 1, 42s to 17.6s (58% faster)

Program 2, 21.6s to 4.3s (80% faster)

Program 3, 37.4s to 15.5s (59% faster)

As a result, the transformed programs now deliver improved scalability, faster runtime performance, and enhanced memory efficiency, making them suitable for high-volume datasets and stable background execution in production environments.

Lines of Custom Code Refactored Across Reports

110,000+

Business Logic Changes, No Regression Risk

Zero

Memory-Optimized for High Volume Runs

100%

Documentation, Traceable, and Audit-ready

Auto

Economic Impact

Compared to traditional manual ABAP optimization (estimated 1 FTE per 2–3 objects per month), the AI-driven approach enabled large-scale optimization at a fraction of the time and cost. Estimated cost savings exceeded 80–90% compared to conventional remediation models.

TECHNICAL APPENDIX: DETAILED OPTIMIZATION LOG

What Changed: From SELECT SINGLE inside loops to FOR ALL ENTRIES with Binary Search

Agents identify every instance of repeated database access inside loops and refactor them to a single bulk fetch with sorted binary search, delivering O(log n) lookup performance regardless of dataset size. The same pattern was applied consistently across all programs with zero exceptions.

PROGRAM 1 — ZRSD_EBILL_EINV_DETAILS (ZSD1062)

20 optimizations. 42s → 17.6s. 58% faster.

LINES	OPTIMIZATION	REASON
382–407	VBFA SELECT SINGLE → FOR ALL ENTRIES	Eliminated N DB calls inside loop; replaced with single bulk fetch and in-memory lookup
392–397	SORT L_TAB_VBRK BY VGBEL + binary search	Ensured sorted table before binary search for correctness and performance
400–407	VBFA lookup via READ TABLE ... BINARY SEARCH	Replaced database access with fast in-memory lookup for better scalability
419	FREE it_tsyrws	Released VBFA preload table after last usage to reduce memory footprint
429–437	VBAK ERNAM preload via FOR ALL ENTRIES	Eliminated repeated SELECT SINGLE calls on VBAK inside processing logic
433–435	SORT it_xfv5uf + DELETE ADJACENT DUPLICATES	Prepared unique, sorted lookup table for efficient binary search
675–682	Binary search lookup for VBAK ERNAM	Replaced database access with in-memory lookup to improve runtime
438–446	EKKO ERNAM preload via FOR ALL ENTRIES	Avoided repeated DB access to EKKO by bulk-fetching purchasing document owners
443–445	SORT it_h6fe8f + DELETE ADJACENT DUPLICATES	Ensured unique and sorted entries for binary search usage
702–710	Binary search lookup for EKKO ERNAM	Improved performance by avoiding DB reads inside loop
447–475	USR21 + ADRP user master preload	Reduced dependency on repeated BAPI_USER_GET_DETAIL calls by caching user master data
476–523	Email resolution via ADR6 with priority handling	Deterministic, efficient email selection avoiding repeated database calls
525	FREE lt_all_emails	Released temporary email lookup table after usage to optimize memory.
527	SORT lt_user_contact BY bname	Ensured sorted user contact table for binary search during user lookup.
669–671	SORT IT_CODMAST_IDS + binary search	Optimized repeated lookups by ensuring correct binary search usage.
788–801	SORT IT_DOCTYPE + binary search	Improved pending-invoice processing performance by optimizing table access.
860–865	SORT IT_RMC_INV + binary search	Optimized e-way bill related lookups by replacing linear scans.
953–955	FREE IT_RMC_INV	Released RMC invoice table after last usage to reduce memory consumption.
1502–1504	FREE L_TAB_VKORG	Released authorization validation table after completion of auth checks.
1528–1530	FREE L_TAB_FINAL2	Freed FAE driver table in GET_PENDING_INV after data extraction.

LINES 382–407

VBFA SELECT SINGLE → FOR ALL ENTRIES

Technical Reason:Eliminated N DB calls inside loop; replaced with single bulk fetch and in-memory lookup

LINES 392–397

SORT L_TAB_VBRK BY VGBEL + binary search

Technical Reason:Ensured sorted table before binary search for correctness and performance

LINES 400–407

VBFA lookup via READ TABLE ... BINARY SEARCH

Technical Reason:Replaced database access with fast in-memory lookup for better scalability

LINES 419

FREE it_tsyrws

Technical Reason:Released VBFA preload table after last usage to reduce memory footprint

LINES 429–437

VBAK ERNAM preload via FOR ALL ENTRIES

Technical Reason:Eliminated repeated SELECT SINGLE calls on VBAK inside processing logic

LINES 433–435

SORT it_xfv5uf + DELETE ADJACENT DUPLICATES

Technical Reason:Prepared unique, sorted lookup table for efficient binary search

LINES 675–682

Binary search lookup for VBAK ERNAM

Technical Reason:Replaced database access with in-memory lookup to improve runtime

LINES 438–446

EKKO ERNAM preload via FOR ALL ENTRIES

Technical Reason:Avoided repeated DB access to EKKO by bulk-fetching purchasing document owners

LINES 443–445

SORT it_h6fe8f + DELETE ADJACENT DUPLICATES

Technical Reason:Ensured unique and sorted entries for binary search usage

LINES 702–710

Binary search lookup for EKKO ERNAM

Technical Reason:Improved performance by avoiding DB reads inside loop

LINES 447–475

USR21 + ADRP user master preload

Technical Reason:Reduced dependency on repeated BAPI_USER_GET_DETAIL calls by caching user master data

LINES 476–523

Email resolution via ADR6 with priority handling

Technical Reason:Deterministic, efficient email selection avoiding repeated database calls

LINES 525

FREE lt_all_emails

Technical Reason:Released temporary email lookup table after usage to optimize memory.

LINES 527

SORT lt_user_contact BY bname

Technical Reason:Ensured sorted user contact table for binary search during user lookup.

LINES 669–671

SORT IT_CODMAST_IDS + binary search

Technical Reason:Optimized repeated lookups by ensuring correct binary search usage.

LINES 788–801

SORT IT_DOCTYPE + binary search

Technical Reason:Improved pending-invoice processing performance by optimizing table access.

LINES 860–865

SORT IT_RMC_INV + binary search

Technical Reason:Optimized e-way bill related lookups by replacing linear scans.

LINES 953–955

FREE IT_RMC_INV

Technical Reason:Released RMC invoice table after last usage to reduce memory consumption.

LINES 1502–1504

FREE L_TAB_VKORG

Technical Reason:Released authorization validation table after completion of auth checks.

LINES 1528–1530

FREE L_TAB_FINAL2

Technical Reason:Freed FAE driver table in GET_PENDING_INV after data extraction.

PROGRAM 2 — zhrsd947 (ZSD206)

21 optimizations. 21.6s → 4.3s. 80% faster.

LINES	OPTIMIZATION	REASON
389–393	SORT it_rpmkr BY kunnr + binary search	Ensured sorted table before binary search for correctness and performance
403–409	SORT it_rpmkr BY kunnr rpmkr + binary search	Enabled efficient deterministic lookup during repeated reads inside loop
729–735	risk_class SELECT SINGLE → FOR ALL ENTRIES	Avoided repeated DB access by preloading risk class data in a single call
733–735	SORT it_efxjg + DELETE ADJACENT DUPLICATES	Ensured unique partner entries and prepared table for binary search
738–757	Unique partner + segment list build	Eliminated redundant partner-segment combinations; reduced repeated function calls
759–767	Pre-read partner and segment lists before function	Reduced function module overhead by passing unique values only
774	FREE lt_pairs, lt_partner, lt_segments	Released temporary helper tables after last usage to reduce memory footprint
783–791	risk_class binary search lookup	Replaced DB calls inside loop with reads from preloaded internal table
812	FREE it_efxjg	Freed preload table after last usage to release memory early
996–1002	altkn SELECT SINGLE → FOR ALL ENTRIES	Eliminated repeated DB calls for customer alternate key retrieval
1000–1001	SORT it_bknib8 + DELETE ADJACENT DUPLICATES	Ensured correct and efficient binary search lookup
1011–1014	Binary search lookup for altkn	Replaced database access with fast in-memory lookup
1022	FREE it_bknib8	Released lookup table after final usage to reduce memory
1031–1037	KNA1 name1 + zzfinance_scheme → FOR ALL ENTRIES	Prevented repeated KNA1 reads by bulk-fetching customer master data
1035–1036	SORT it_w8extk + DELETE ADJACENT DUPLICATES	Prepared lookup table for efficient binary search
1046–1050	Binary search for customer name and finance scheme	Improved runtime by replacing DB reads with internal table lookup
1058	FREE it_w8extk	Freed customer master preload table after usage
1139–1145	oitol SELECT SINGLE → FOR ALL ENTRIES (T691F)	Eliminated repeated DB access to T691F during loop execution
1142–1143	SORT it_yvvk9o + DELETE ADJACENT DUPLICATES	Ensured correct binary search behaviour and removed duplicates
1152–1155	Binary search lookup for oitol	Replaced DB reads with in-memory access for better performance
1164	FREE it_yvvk9o	Released temporary lookup table after last usage

LINES 389–393

SORT it_rpmkr BY kunnr + binary search

Technical Reason:Ensured sorted table before binary search for correctness and performance

LINES 403–409

SORT it_rpmkr BY kunnr rpmkr + binary search

Technical Reason:Enabled efficient deterministic lookup during repeated reads inside loop

LINES 729–735

risk_class SELECT SINGLE → FOR ALL ENTRIES

Technical Reason:Avoided repeated DB access by preloading risk class data in a single call

LINES 733–735

SORT it_efxjg + DELETE ADJACENT DUPLICATES

Technical Reason:Ensured unique partner entries and prepared table for binary search

LINES 738–757

Unique partner + segment list build

Technical Reason:Eliminated redundant partner-segment combinations; reduced repeated function calls

LINES 759–767

Pre-read partner and segment lists before function

Technical Reason:Reduced function module overhead by passing unique values only

LINES 774

FREE lt_pairs, lt_partner, lt_segments

Technical Reason:Released temporary helper tables after last usage to reduce memory footprint

LINES 783–791

risk_class binary search lookup

Technical Reason:Replaced DB calls inside loop with reads from preloaded internal table

LINES 812

FREE it_efxjg

Technical Reason:Freed preload table after last usage to release memory early

LINES 996–1002

altkn SELECT SINGLE → FOR ALL ENTRIES

Technical Reason:Eliminated repeated DB calls for customer alternate key retrieval

LINES 1000–1001

SORT it_bknib8 + DELETE ADJACENT DUPLICATES

Technical Reason:Ensured correct and efficient binary search lookup

LINES 1011–1014

Binary search lookup for altkn

Technical Reason:Replaced database access with fast in-memory lookup

LINES 1022

FREE it_bknib8

Technical Reason:Released lookup table after final usage to reduce memory

LINES 1031–1037

KNA1 name1 + zzfinance_scheme → FOR ALL ENTRIES

Technical Reason:Prevented repeated KNA1 reads by bulk-fetching customer master data

LINES 1035–1036

SORT it_w8extk + DELETE ADJACENT DUPLICATES

Technical Reason:Prepared lookup table for efficient binary search

LINES 1046–1050

Binary search for customer name and finance scheme

Technical Reason:Improved runtime by replacing DB reads with internal table lookup

LINES 1058

FREE it_w8extk

Technical Reason:Freed customer master preload table after usage

LINES 1139–1145

oitol SELECT SINGLE → FOR ALL ENTRIES (T691F)

Technical Reason:Eliminated repeated DB access to T691F during loop execution

LINES 1142–1143

SORT it_yvvk9o + DELETE ADJACENT DUPLICATES

Technical Reason:Ensured correct binary search behaviour and removed duplicates

LINES 1152–1155

Binary search lookup for oitol

Technical Reason:Replaced DB reads with in-memory access for better performance

LINES 1164

FREE it_yvvk9o

Technical Reason:Released temporary lookup table after last usage

PROGRAM 3 — ZSDR_EXC_INVOICE_CDS (ZSD149)

19 optimizations. 37.4s → 15.5s. 59% faster.

LINES	OPTIMIZATION	REASON
390–392	SORT i_bsis1 BY bukrs belnr gjahr	Ensured sorted internal table for correct and efficient binary search during lookup
397–404	READ TABLE i_bsis1 ... BINARY SEARCH	Replaced linear search with binary search for faster in-memory access inside loop
412–415	FREE l_tab_bseg	Released temporary BSEG working table after last usage to reduce memory footprint
418–420	FREE i_bsis1	Explicitly freed BSIS lookup table after completion of processing
577–579	SORT it_faglflexa + SORT i_bsis	Prepared both tables for efficient binary search and deterministic access
590–593	Binary search on it_faglflexa	Replaced database access with fast internal table lookup for FAGLFLEXA data
601–603	FREE it_faglflexa	Released FAGLFLEXA preload table after final usage to optimise memory
663–704	Parallel cursor logic on i_bsis	Improved performance by iterating only over matching key ranges instead of full table scans
714–735	Bulk SELECT on BSEG via FOR ALL ENTRIES	Eliminated repeated DB calls by fetching BSEG data in a single bulk operation
721–724	SORT lt_bseg_sa56gg + DELETE ADJACENT DUPLICATES	Ensured unique key combinations and prepared table for efficient access
733–735	SORT lt_bseg_sa56gh + duplicate removal	Reduced redundant entries and ensured deterministic lookup behaviour
738–740	FREE lt_collected_keys	Released collected key table after DB fetch to reduce memory usage
742–745	SORT i_hkont + DELETE ADJACENT DUPLICATES	Prepared G/L account lookup table for optimised binary search
749–767	Parallel cursor on i_bsis with indexed looping	Processed only contiguous matching records, avoiding unnecessary full-table comparisons
878–880	Binary search on i_hkont	Improved lookup performance for G/L account resolution
893–895	FREE i_bsis, t_i_bsis, lt_bseg_sa56gg/gh	Released multiple temporary processing tables after last usage to minimise memory
972–987	Replaced SELECT SINGLE on BSEG with internal table lookup	Avoided repeated DB access by reading vendor data from preloaded BSEG table.
990–1005	Binary search on p_ltl_bseg	Improved vendor lookup performance by using sorted internal table access.

LINES 390–392

SORT i_bsis1 BY bukrs belnr gjahr

Technical Reason:Ensured sorted internal table for correct and efficient binary search during lookup

LINES 397–404

READ TABLE i_bsis1 ... BINARY SEARCH

Technical Reason:Replaced linear search with binary search for faster in-memory access inside loop

LINES 412–415

FREE l_tab_bseg

Technical Reason:Released temporary BSEG working table after last usage to reduce memory footprint

LINES 418–420

FREE i_bsis1

Technical Reason:Explicitly freed BSIS lookup table after completion of processing

LINES 577–579

SORT it_faglflexa + SORT i_bsis

Technical Reason:Prepared both tables for efficient binary search and deterministic access

LINES 590–593

Binary search on it_faglflexa

Technical Reason:Replaced database access with fast internal table lookup for FAGLFLEXA data

LINES 601–603

FREE it_faglflexa

Technical Reason:Released FAGLFLEXA preload table after final usage to optimise memory

LINES 663–704

Parallel cursor logic on i_bsis

Technical Reason:Improved performance by iterating only over matching key ranges instead of full table scans

LINES 714–735

Bulk SELECT on BSEG via FOR ALL ENTRIES

Technical Reason:Eliminated repeated DB calls by fetching BSEG data in a single bulk operation

LINES 721–724

SORT lt_bseg_sa56gg + DELETE ADJACENT DUPLICATES

Technical Reason:Ensured unique key combinations and prepared table for efficient access

LINES 733–735

SORT lt_bseg_sa56gh + duplicate removal

Technical Reason:Reduced redundant entries and ensured deterministic lookup behaviour

LINES 738–740

FREE lt_collected_keys

Technical Reason:Released collected key table after DB fetch to reduce memory usage

LINES 742–745

SORT i_hkont + DELETE ADJACENT DUPLICATES

Technical Reason:Prepared G/L account lookup table for optimised binary search

LINES 749–767

Parallel cursor on i_bsis with indexed looping

Technical Reason:Processed only contiguous matching records, avoiding unnecessary full-table comparisons

LINES 878–880

Binary search on i_hkont

Technical Reason:Improved lookup performance for G/L account resolution

LINES 893–895

FREE i_bsis, t_i_bsis, lt_bseg_sa56gg/gh

Technical Reason:Released multiple temporary processing tables after last usage to minimise memory

LINES 972–987

Replaced SELECT SINGLE on BSEG with internal table lookup

Technical Reason:Avoided repeated DB access by reading vendor data from preloaded BSEG table.

LINES 990–1005

Binary search on p_ltl_bseg

Technical Reason:Improved vendor lookup performance by using sorted internal table access.

SOOKTI AI'S VALUE

ASPECT

BEFORE (MANUAL)

NOW WITH SOOKTI AI

Code Transformation Time/ Time to Go-Live

1 week per report (7 man-days per 5,000+ lines)

Minutes per report, no manual intervention

Performance

Bottlenecks from outdated patterns (SELECT * , loop-in-loop)

50–90%faster execution with HANA-specific techniques

Code Optimization

Manual identification and patching of inefficient code

Automatic identification and correction; rewritten for HANA (CDS views, parallel cursors)

Quality Assurance

Manual testing; high dependency on key developers

100%functional parity validated with automated regression testing

Risk Management

High dependency on key developers; risk of human error

No “hero-developer” dependency; consistent, repeatable, auditable process with self-documenting output

SAP Clean-Core & Future/Upgrade-Proof

Customizations break with SAP upgrades, manual checking to ensure HANA-compliant code

Built-in clean-core compliance; SAP Standard CDS, Functions and Classes

Long-Term Maintainability

Difficult to maintain; lack of documentation and poor readability

Annotated ABAP with rationale per change; easy to maintain

ASPECT

Code Transformation Time/ Time to Go-Live

BEFORE (MANUAL)

1 week per report (7 man-days per 5,000+ lines)

NOW WITH SOOKTI AI

Minutes per report, no manual intervention

ASPECT

Performance

BEFORE (MANUAL)

Bottlenecks from outdated patterns (SELECT * , loop-in-loop)

NOW WITH SOOKTI AI

50–90%faster execution with HANA-specific techniques

ASPECT

Code Optimization

BEFORE (MANUAL)

Manual identification and patching of inefficient code

NOW WITH SOOKTI AI

Automatic identification and correction; rewritten for HANA (CDS views, parallel cursors)

ASPECT

Quality Assurance

BEFORE (MANUAL)

Manual testing; high dependency on key developers

NOW WITH SOOKTI AI

100%functional parity validated with automated regression testing

ASPECT

Risk Management

BEFORE (MANUAL)

High dependency on key developers; risk of human error

NOW WITH SOOKTI AI

No “hero-developer” dependency; consistent, repeatable, auditable process with self-documenting output

ASPECT

SAP Clean-Core & Future/Upgrade-Proof

BEFORE (MANUAL)

Customizations break with SAP upgrades, manual checking to ensure HANA-compliant code

NOW WITH SOOKTI AI

Built-in clean-core compliance; SAP Standard CDS, Functions and Classes

ASPECT

Long-Term Maintainability

BEFORE (MANUAL)

Difficult to maintain; lack of documentation and poor readability

NOW WITH SOOKTI AI

Annotated ABAP with rationale per change; easy to maintain

TESTIMONIALS

- SAP Technical Lead, Global Cement Manufacturer

ENGAGEMENT DETAILS

Engagement Details

Industry

Manufacturing (Cement)

Landscape

SAP ECC → S/4HANA (Brownfield)

Scope

Custom Code Optimization

Environment

Live Enterprise Production Workloads

Ready to Unlock Similar Performance Gains?

Run a controlled pilot on your most critical SAP workloads and validate measurable results before scaling enterprise-wide.

Book a call with us or write to: aditya@sookti.ai | tanushree@sookti.ai