ecociel · research note · SAP ECC on IBM i → S/4HANA
Cutting migration QA effort by up to 10× with AI: what holds up and what doesn’t
How far AI and automation can cut functional-equivalence testing, data validation, test-case creation, regression and UAT when an SAP ECC system on IBM i (Db2 for i) moves to S/4HANA. Covers sources, a vendor map, recurring patterns, a first-principles effort model and three playbooks for customers.
Bottom line
≤35%
SAP’s own claim for AI in test management (Cloud ALM)[1]
~15–20×
cheaper per extra test cycle with differential testing estimate
7.8×–10×
total QA reduction, Playbook C mature, M→L program size estimate
- No public source shows 10× on total migration QA. SAP claims up to 35% for AI-supported test management[1]. SI offerings claim 25–40% (Wipro, TCS)[34][35]. The 85–92% figures cover specific buckets (impact-scoped regression on updates, test-script generation for upgrades)[29][23][31].
- The lever behind 10× is not better test writing. It is getting rid of the oracle problem. In a brownfield conversion the requirement is “behaves like before, except for the intended differences”. The legacy system, or production itself, becomes the oracle, and testing turns into diffing. SAP’s own DTV/LTVF already work this way for data[3]. Basis Technologies’ Testimony did record-and-replay from production back in 2017[33].
- AI cuts the cost per test cycle by roughly 15–20×, covering execution, reconciliation and evidence. estimate The total stays capped by fixed human work: approving intended differences, business acceptance, triage decisions, residual UI/authorisation tests and building the engine. My model puts the ceilings at ~2.4× for tool-led, ~9× for an AI equivalence engine and ~13× for production-as-oracle.
- IBM i adds real equivalence risk that differential testing catches cheaply: the EBCDIC→Unicode step (two-step path if non-Unicode)[4], sort-order changes[7], a Windows SUM host for DMO[5], and re-platforming of IBM i-side jobs, spool output and IFS file interfaces.
- What ecociel can sell: a reusable equivalence/replay engine. The fixed cost per customer drops from ~300–350 PD (first build) to ~120–150 PD, and the 10× range opens up for mid-to-large programs. estimate
Context: what is specific about ECC on IBM i
Technical path
- S/4HANA is Unicode-only. A non-Unicode (EBCDIC) ECC cannot convert to Unicode and S/4 in one step. SAP recommends upgrading to ECC 6.0 EHP7 with DMO including Unicode migration first, then running the S/4 conversion[4].
- Db2 for i → HANA runs through SUM with DMO. For S/4HANA 1909 and later, the SUM host has to be a Windows application server, and specific IBM i PTFs are required[5].
- IBM i application servers are not supported for S/4. Uninstall them after DMO, but only after moving SAPGLOBALHOST[6].
- Collect usage data with SCMON/SUSG over a long window (ideally covering year-end) to scope custom code and tests[8].
Equivalence risks that tests must cover
- Collation: EBCDIC sorts letters before digits, Unicode/ASCII sorts digits before letters[7]. This hits report order,
BETWEEN/range selections on character keys and custom code that relies on order. - Code-page boundaries: outbound/inbound flat files on the IFS, partner EDI and RPG/CL programs outside SAP that read SAP data. Diff outbound files byte-for-byte. practice
- Platform glue: IBM i job scheduler/CL scripts, spool/AFP output and Db2-for-i native SQL in custom code all need re-platforming. Each is a classic “silent difference”. practice
- S/4 data-model changes (ACDOCA, MATDOC, BP, PRCD_ELEMENTS) produce expected diffs. They need normalisation rules (KIDs) or every diff turns into noise[3].
What the research says
Only items I found and read are listed. Quotes are verbatim. Vendor-commissioned studies are labelled as such. Numbers refer to sources at the bottom.
| Source | Finding |
|---|---|
| SAP | Test management with AI in SAP Cloud ALM: “Up to 35% effort savings and 6–15% reduction in project duration using AI.” [1] |
| SAP | DTV is “SAP’s standard solution for automated data comparison during system conversions, selective transitions, and system upgrades”. LTVF adds rule-based comparisons, e.g. KONV in the source compared with PRCD_ELEMENTS in the target, and can “accept expected differences, compare substrings, concatenations, tolerances”. [3] |
| SAP | IBM i specifics: S/4HANA “runs with Unicode only”. An “integrated non-unicode to unicode conversion together with the conversion to SAP S/4HANA is not available”, so a non-Unicode system needs a two-step procedure. [4] |
| SAP | For target S/4HANA 1909 and later, IBM i is no longer supported as an application server platform. DMO from Db2 for i needs a Windows application server. [5] |
| SAP | After DMO, the remaining IBM i instances have to be uninstalled. SAPGLOBALHOST has to be moved first, otherwise “your S/4HANA system will be damaged”. [6] |
| IBM | EBCDIC and Unicode sort sequences differ: in EBCDIC, letters sort before digits; in Unicode/ASCII, digits sort before letters. Sort order, character range selections and ORDER BY results can therefore change. [7] |
| Horváth | 200 companies surveyed (Q1 2025): projects take “on average 30 percent longer than originally planned”; 65% report strong to very strong quality deficits; “underestimated testing and data migration phases” is among the main reasons. [9] |
| ASUG | 56% of ASUG members are live on S/4HANA or moving to it. Of those, 44% chose brownfield, 29% selective/bluefield and 26% greenfield. 31% report costs above expectations. [10] |
| DSAG | In the DACH region, SAP ERP/Business Suite is still in use at 51% (2025) and S/4HANA on-premise at 42%. 68% say AI has high or medium relevance (n=243). [11] |
| SAPinsider | 55% cite project duration as a major concern (37% in 2024). 62% cite high project costs. Automated testing tools are used or planned by 28%. [12] |
| Sogeti/Capgemini | 56% allocate 5–15% of department budget to SAP end-to-end testing. 25.4% still rely on a manual approach; 42.7% have automated over 50% of their SAP testing. [13] |
| World Quality Report | 89% are piloting or deploying GenAI-augmented QE workflows (37% in production, 52% in pilots). Only 15% have reached enterprise-wide deployment. [14] |
| Gartner | First Magic Quadrant for AI-Augmented Software Testing Tools (6 Oct 2025, 10 vendors). The 2022 Market Guide forecast: “By 2027, 80% of enterprises will have integrated … AI-augmented testing tools into their software engineering toolchain”, up from 10% in 2022. [15] |
| Forrester (Tricentis-commissioned) | TEI 2026: 403% ROI, $9M NPV, and testing scope reduced by 84%. [17] |
| IDC (Worksoft-sponsored) | 83% less staff time to test equivalent business processes; 61% less staff time to create and maintain scripts; 23% faster completion of major migrations/projects; 548% five-year ROI. [18] |
| McKinsey | Orchestrated gen-AI agents cut a bank’s 700–800 hour mainframe migration estimate by 40%. One insurer improved code modernization efficiency and testing by more than 50%. [19] |
| Deloitte | Ascend Test AI produced a complete suite of test scenarios, cases and steps for a broad ERP scope “in under 48 hours”, a process that “would typically take weeks” (French pharma S/4HANA). [21] |
| Deloitte/UiPath | Customer Zero S/4HANA: “60% of test cases automated” and a “10% acceleration in project delivery”. [22] |
| PwC | Accelleron (S/4HANA Cloud upgrade regression, 40 countries): GenAI agents on SAP BTP delivered a “92% reduction in testing effort”, about CHF 296,000 per upgrade cycle. [23] |
| PwC | Data Quality & Migration Suite: “Saves up to 70% of business user validation time during mock cycles or cutover data loads.” [24] |
| KPMG | QA has to stay independent of business integration. “UAT should be led by the business”. [26] |
| Accenture | GenWizard platform claims “5-8X faster speed to market” and “>50% improvement in software quality”. These are platform-wide, not SAP-QA-specific. [28] |
| EY | Intelligence Realized Platform: integrates EY assets for SAP transformation. No QA-specific figure found. [27] |
How to read this: Survey data (Horváth, ASUG, SAPinsider, Sogeti, WQR) shows testing is a known overrun driver and AI adoption is broad but shallow. Effort-reduction claims come almost entirely from vendors or vendor-commissioned studies, are measured on narrow scopes, and rarely refer to a full ECC→S/4 conversion. Nothing found from Gartner/Forrester/IDC quantifies QA reduction specifically for S/4 migrations.
Vendor landscape
Promise = the vendor’s own words or a close paraphrase from the linked page. Not verified. “Fit” = where it slots into Playbooks A/B/C. All links were checked on 13 Oct 2026: they returned HTTP 200 or loaded in a browser-equivalent fetch. Some sites block scripted clients.
| Vendor / offering | Stated promise | Fit | Link |
|---|---|---|---|
| Tricentis LiveCompare Change impact analysis | “LiveCompare customers reduce testing efforts by 85% or more” (footnote: average 87%, sample of 120 SAP enterprise customers). Mostly measured on updates/transports. | A, B (scoping) | link |
| Tricentis Tosca / SAP Enterprise Continuous Testing by Tricentis Test automation (SAP Solution Extension) | “Codeless, AI-powered test automation.” SAP ECT now generates “complete, end-to-end test cases using natural language prompts” (May 2026). | A, B (residual) | link |
| Worksoft Certify Business-process test automation | “Accelerate project timelines, reduce risk, and increase efficiency with the world’s most trusted test automation platform for SAP and beyond.” | A | link |
| SAP Cloud ALM: Test Management Assistant (Joule) ALM + AI agents | “Automate test scoping, preparation, data determination and provisioning, and execution support for continuous, risk-based validation at scale.” SAP blog: up to 35% effort savings. | A, B (backbone) | link |
| SAP Data Transition Validation (DTV) / LTVF Data validation | DTV: snapshot source and target, compare automatically, highlight differences; pre-delivered FI/MM/SD content. LTVF: rule-based, cross-table, tolerances, expected differences. | A, B, C | link |
| SAP Signavio Process Intelligence / Process Insights Process mining | “Improve process performance at scale using AI agents to capture workflows, analyze execution, and accelerate transformation.” | A, B (scoping) | link |
| SAP Build Process Automation Low-code automation | “Automate and transform business processes with an easy-to-use no-code experience.” Not a QA product. Relevant only as a host for custom agents. | (B) optional | link |
| Panaya Change intelligence + test management | “SAP S/4HANA transformations with zero go-live issues”. ECC upgrade prep with “50% fewer resources and cutting project timelines in half”. Testing page: “Agentic SAP testing at enterprise scale.” | A | link |
| Basis Technologies Change/DevOps for SAP; Testimony (RTA) | Current site: “De-risk S/4HANA migration with automated dual maintenance.” Testimony learns tests from real production activity and replays them (2017 launch article). Not prominent on today’s site; check whether it is still sold. | C (precedent) | link |
| Celonis Process mining | HARMAN story: “34% of custom T-codes and 22% of custom programs revealed to be low-usage and low-value.” The S/4 test instance gets connected to track the transition. | A, B (scoping) | link |
| Qualitest (QualityAI) QE services | “AI can perform test scenario, case and script generation and even deliver these assets to the CI/CD pipeline.” | A, B (delivery partner) | link |
| Testsigma Agentic test automation | “Test 10x faster - powered by AI Agents”, “90% Lesser Test Maintenance”. SAP GUI automation per docs. | B (residual UI) | link |
| Original Software Code-free testing / UAT | “Code-free SAP test automation for ECC, S/4HANA and SuccessFactors. Manage BAU changes and migrate with confidence.” | A (UAT) | link |
| Keysight Eggplant Model-based, image-based UI testing | Tests “at the UI level with image recognition, so it runs on any … application that renders on a screen, SAP …”. Keysight Generator turns requirements into Gherkin scenarios. | B (residual UI) | link |
| smartShift Custom code transformation | “SAP Custom Code Transformation with 99.99% Precision.” Also automated dual maintenance. | adjacent (code) | link |
| Syniti (SAP ADMM) Data migration & quality | “Proven to increase project efficiency by 46%”; “95% reduction in data quality errors”. | adjacent (data) | link |
| SNP Kyano Validate Data validation | “Automating data verification activities, enabling large-scale testing with minimal effort.” Run most tests “during the go-live downtime”. | A, B, C | link |
| Natuvion (DCS / Automatic Report Check) Data validation | ARC runs identical report variants in source and target and compares them. “Significantly reduces the manual validation effort, shortens reconciliation cycles.” | A, B, C | link |
| Opkey AI ERP test automation | “Reduce manual testing effort by 80%”; “Speed up your ECC to S/4HANA timeline by months & reduce costs up to 50%.” | A | link |
| Avo Automation (Avo Assure) No-code test automation | “Reduce SAP regression effort by 80% with no-code test automation across SAP ECC, Fiori, SuccessFactors, and S/4HANA migrations.” | A | link |
| Sofy AI-native SAP test agents | “Validate SAP business processes end-to-end — across GUI, Fiori, and integrations — without fragile scripts or manual maintenance.” | B (residual UI) | link |
| Qorelo AI-native SAP delivery layer | “The AI layer of your S/4HANA migration”. Up to 45% less total time to completion on its measured (greenfield) path. Raised €3M seed (2026). | adjacent (discovery) | link |
| Synthesized Test data / synthetic data | “SAP TDMS isn’t released for S/4HANA. Build lean, masked test clients another way.” | B, C (data) | link |
| Accenture GenWizard SI AI platform | “5-8X faster speed to market”, “50-75% reduction in IT costs” (platform-wide). | SI option | link |
| Infosys Topaz for SAP S/4HANA Cloud SI AI suite | “Fast track adoption timelines by using 40+ AI infused accelerators.” | SI option | link |
| Cognizant SI (S/4HANA services, Tricentis partner) | S/4HANA transformation services; “S/4HANA migration increases user productivity by 25%” (featured story). No QA-specific figure on page. | SI option | link |
| TCS ConvertCore SI conversion framework | “Accelerate your conversion journey by up to 40% by leveraging automation, business reference suite, and industry-specific accelerators.” | SI option | link |
| Deloitte Ascend Test AI (+UiPath) SI AI test generation | Generated scenarios, cases and steps for a broad ERP scope “in under 48 hours”. | SI option | link |
| PwC Data Quality & Migration Suite SI data accelerator | “2,000+ embedded data quality rules”; “up to 70% of business user validation time” saved. | SI option | link |
| Wipro AI-powered SAP S/4HANA Assurance SI QA offering | “More than a 25% reduction in test assurance efforts”; “30% reduction in test cycle time”. | SI option | link |
Patterns and a common methodology
1 · Scope by evidence
Usage logs (SCMON/SUSG, ST03N/STAD) and process mining replace workshop-driven scope. Typically a third of custom objects and transactions turn out to be dead or low-value[32][8].2 · Test what changed
Impact analysis maps changed objects to business processes and tests. Kimberly-Clark found only 13% of normally tested transactions were impacted[31][29].3 · Legacy as oracle
Compare instead of asserting: report/table snapshots (DTV, LTVF, Kyano, ARC) and record-and-replay (Testimony)[3][33].4 · Generate, don’t write
LLMs generate test cases from process models, documents and recordings. Hours instead of weeks[21][23][2].5 · Self-healing / agentic execution
Computer-use and model-based agents remove most script maintenance[25][36].6 · Intended differences as a first-class artefact
Expected differences get encoded as comparison rules (LTVF “accept expected differences”, ecociel’s KID register)[3].7 · Evidence by default
Execution evidence comes out of runs automatically, aligned to SOX/ICS[25].8 · Humans keep the decisions
Independent QA and business-led UAT remain[26]. SAP keeps “key decisions human-led”[1].Common methodology
Discover → Capture baseline behaviour → Normalise (KIDs) → Generate → Execute differentially → Triage → Accept → Evidence. Run it once per technical step (Unicode, DMO) and once per mock cycle.First principles: where QA effort actually goes
| Bucket | What the effort really is | What AI changes | Irreducible floor |
|---|---|---|---|
| Test design | Deciding what to test and what correct looks like (the oracle problem) | Generation from processes/documents is cheap. In conversions the oracle is free (legacy behaviour). | Review of coverage; acceptance of intended differences |
| Test data | Finding representative, consistent, protected data across document chains | Production snapshots + masking; AI picks representative documents per process variant | Data-protection decisions (nDSG/GDPR), masking validation |
| Execution | Clicking through cases, capturing evidence, repeated per cycle | API-level replay is cheapest; agentic UI where needed; marginal cost → ~0 | Oversight; things only humans can judge (UX) |
| Reconciliation | Comparing balances, open items, stock, valuations | Full-population diffs instead of samples; AI explains diffs | Approving tolerances and residual diffs |
| Defect triage | Is it a defect, intended, environment, or data? Who owns it? | Clustering, KID matching, root-cause hypotheses | Decision and ownership |
| Documentation | Evidence packs, reports, audit trail | Generated from run logs | Sign-off |
| UAT | Business confidence and accountability | Focus on differences only | Accountable owners must still accept |
Limits AI does not remove
- Sign-off accountability: process owners and the CFO sign, not a model. Independent QA has to stay independent[26].
- Intended differences: simplification items, new number formats (40-char material), BP and credit management changes all create expected diffs that someone has to approve.
- Audit/regulatory: ICS/SOX ITGC evidence, GxP CSV in life sciences, Swiss nDSG/GDPR when production data is used for testing. LLMs must not see unmasked personal data without a lawful basis and suitable hosting.
- Non-determinism: use LLMs to produce deterministic artefacts (tests, rules, explanations). The diff engine itself has to be deterministic and pass seeded-defect evals.
- No diff ≠ correct: equivalence proves “same as before”, which includes old bugs. New functionality still needs classic tests.
Why 10× is hard (the arithmetic)
Total effort = fixed human work + per-cycle work × cycles. AI pushes per-cycle work towards zero (≈420 PD/cycle → ≈20–30 PD/cycle in my model). The total is then dominated by work that doesn’t go away: KID approval, UAT acceptance, triage decisions, residual UI/authorisation tests and engine setup.
The ceiling is baseline ÷ scaled human residual: A ≈ 2.4×, B ≈ 9.2×, C ≈ 12.9× estimate. Getting past 10× requires (1) a productised engine, (2) a program large enough to amortise it, (3) business/auditor acceptance of diff evidence in place of scripted UAT.
The reasoning model
estimate Reference program (my assumption): mid-to-large manufacturer, ECC 6.0 on IBM i, non-Unicode, modules FI/CO, MM, SD, PP, ~1,200 test cases, ~30 interfaces. Two-step path = 6 validation cycles (Unicode mock + Unicode go-live validation, 3 S/4 mocks, dress rehearsal). Excludes technical conversion work (Basis/SUM/DMO, custom-code remediation), licences, compute and training.
Traditional baseline
| Effort bucket | PD | Share | How the number is built |
|---|---|---|---|
| Scoping & test design | 800 | 17% | ~1,200 test cases written or updated by functional consultants at ~0.5 PD each (600) plus scoping workshops and process walkthroughs (200). |
| Test data & environments | 600 | 12% | 6 validation cycles × 100 PD: data selection, client copies, masking, master-data fixes, interface stubs. |
| Test execution (SIT / regression) | 1,200 | 25% | 6 cycles × 200 PD: ~400 end-to-end and regression cases per cycle at ~0.5 PD incl. screenshots/evidence. |
| Data validation & reconciliation | 720 | 15% | 6 cycles × 120 PD: key users and consultants reconciling FI/CO balances, open items, stock and valuation, open sales/purchase documents. |
| User acceptance testing | 600 | 12% | 2 UAT rounds × ~30 key users × ~10 days. |
| Defect triage & retest | 600 | 12% | ~1,000 defects/observations × ~0.6 PD (analysis, routing, fix verification, retest). |
| Documentation, evidence & sign-off | 300 | 6% | Test reports, evidence packs, audit/ICS documentation, sign-off meetings. |
| Total traditional QA effort | 4,820 | 100% | ≈ 22 person-years |
Projected effort per playbook
| Playbook | First deployment (PD) | Factor | Mature / productised (PD) | Factor | Ceiling at large scale |
|---|---|---|---|---|---|
| Traditional baseline | 4,820 | 1.0× | 4,820 | 1.0× | – |
| A · Scope & Automate | 2,275 | 2.1× | 2,114 | 2.3× | 2.4× |
| B · Equivalence Engine | 1,220 | 4.0× | 743 | 6.5× | 9.2× |
| C · Production-as-Oracle | 1,262 | 3.8× | 615 | 7.8× | 12.9× |
Sensitivity to program size
Fixed costs (engine/harness, evals, setup) don’t scale with program size; everything else does. For small programs, AI approaches only pay off if the engine is productised.
| Program size | Baseline PD | A first | A mature | B first | B mature | C first | C mature |
|---|---|---|---|---|---|---|---|
| S (small, 0.25×) | 1,205 | 1.9× | 2.0× | 1.9× | 3.4× | 1.7× | 3.6× |
| M (reference, 1×) | 4,820 | 2.1× | 2.3× | 4.0× | 6.5× | 3.8× | 7.8× |
| L (large, 2.5×) | 12,050 | 2.2× | 2.3× | 5.1× | 7.9× | 5.1× | 10.2× |
| XL (5×) | 24,100 | 2.2× | 2.4× | 5.6× | 8.5× | 5.7× | 11.4× |
Three playbooks
Conservative · tool-led
A · Scope & Automate
2.1×
first deployment, ~13 months QA calendar estimate
Usage-based scoping, impact analysis, AI-assisted test design in Cloud ALM, automation of the top 60%, DTV/Kyano for data. Low risk, standard vendors, will not reach 10×.
AI-agentic · differential
B · Equivalence Engine
4.0× → 6.5×
first → mature, ~10–11 months estimate
Agents mine real production documents into scenarios, run them on ECC and S/4 copies, and diff the results against the KID register. Humans approve differences and sign off.
Radical · data-diff first
C · Production-as-Oracle
3.8× → 7.8×
first → mature; ~10× at L size; ~10–11 months estimate
Convert the period-start backup, replay a real closed period, and diff the period-end state against actual production. Agentic tests cover only what replay can’t see.
Reuse of ecociel material: the existing Test Strategy (usage-based scoping, E2E process tests, automated reconciliation, parallel-run diffing, key-user UI recording), the KID register and the seeded-defect Evals concept map directly onto these playbooks. A uses pillars 1–2, B industrialises pillars 2–4 plus KID plus evals, and C is pillar 3 (parallel-run diffing) pushed to the limit.
Caveats
- All reductions and timelines are my own estimates from the bucket model above, not sourced figures. Confidence: A medium-high (±25%), B medium (±40%), C low-medium (±50%, depends on how much of the volume can be replayed).
- Vendor claims are self-reported, often on BAU updates rather than conversions, and not comparable with each other.
- Baseline effort varies widely between customers. Calibrate it with the customer’s last upgrade/support-pack test effort before quoting a factor.
- Using production data for testing needs a data-protection assessment (Swiss nDSG / GDPR) and, for LLM use, suitable hosting and contractual controls.
Sources
Checked on 13 Oct 2026. Several sites (SAP Community, PwC, TCS, Horváth, IBM Docs, Keysight, Opkey) block scripted clients, so their pages were read through a fetch service instead. The McKinsey pages blocked both and were read through the search index only.
- SAP (SAP Community, Jun 2026): Your Path to AI-Powered transformation with SAP Cloud ALM
- SAP: Test Management Assistant (Joule, SAP Cloud ALM)
- SAP (SAP Community, Feb 2026): Business Data Validation: comparative overview of LTVF and DTV
- SAP Help: Conversion to SAP S/4HANA using SUM: ABAP Systems on IBM i (SUM 2.0 SP25 guide, PDF)
- SAP (SAP Community): SAP on IBM i: Database Migration Option with target S/4HANA 1909 and beyond
- SAP (SAP Community): SAP on IBM i: Mandatory Post-DMO Activities for SAP S/4HANA Migrations
- IBM Docs: Differences between Unicode and EBCDIC sorting sequences (Db2)
- SAP Help: Custom Code Management during an SAP S/4HANA Conversion (PDF)
- Horváth (Apr 2025): Study: SAP S/4HANA transformations rarely go as planned
- ASUG Research (2025): SAP S/4HANA Adoption Trends, Insights, and Advice
- DSAG (Mar 2025): DSAG-Investitionsreport 2025
- SAPinsider (2025): SAP S/4HANA Migration Benchmark Report 2025 (PDF)
- Sogeti / Capgemini (2023): State of Worldwide Business Assurance for SAP Solutions 2023 (PDF)
- Capgemini / OpenText / Sogeti (Nov 2025): World Quality Report 2025-26 (summary PDF)
- Gartner via Tricentis (Oct 2025): Magic Quadrant for AI-Augmented Software Testing Tools
- Gartner via Appvance (2022): Market Guide for AI-Augmented Software Testing Tools (reprint)
- Forrester Consulting TEI, commissioned by Tricentis (2026): Total Economic Impact of SAP enterprise application QA by Tricentis
- IDC, sponsored by Worksoft: The Value of Automating Business Process Testing with Worksoft (PDF)
- McKinsey: AI for IT modernization: Faster, cheaper, better
- McKinsey: Unleashing developer productivity with generative AI
- Deloitte & UiPath: Agentic Test Automation (incl. Ascend Test AI case, PDF)
- UiPath (Jun 2025): UiPath and Deloitte redefine ERP modernization (Customer Zero)
- SAP / PwC Switzerland (case study): Automated creation of test scripts leveraging GenAI agents
- PwC (Dec 2025): Data Quality & Migration Suite for SAP S/4HANA
- PwC: Rethinking enterprise testing with AI-powered adaptability
- KPMG Advisory Belgium (2026): Quality assurance and business integration on SAP S/4HANA transformations (PDF)
- EY: Intelligence Realized Platform
- Accenture: GenWizard
- Tricentis (vendor blog): SAP key users need LiveCompare, not hypercare
- Tricentis (case study): OMV: automated impact analysis for S/4HANA
- Tricentis via CaseStudies.com: Kimberly-Clark Australia reduces SAP testing by 40% with LiveCompare
- Celonis (customer story): HARMAN S/4HANA transformation
- InsideSAP (2017): Basis Technologies harnesses automation to reinvent SAP testing (Testimony)
- Wipro: AI-powered SAP S/4HANA Assurance
- TCS: TCS ConvertCore: Defining SAP S/4HANA Conversion Roadmap
- Tricentis (May 2026): Tricentis releases agentic AI testing for SAP business transformation