This part­ner insight was authored by Bri­an Rogan, Growth Leader, U.S. at LG CNS Per­fecTwin.

ECC main­stream main­te­nance ends in 2027, and the cus­tomers run­ning it can’t address that dead­line in iso­la­tion. Migrat­ing to S/4HANA is the obvi­ous move, but two oth­er shifts come with it: quar­ter­ly upgrades become a per­ma­nent fea­ture of life under Cloud ERP, and Clean Core requires that cus­tomiza­tions move off the core sys­tem and onto SAP BTP. Each of those changes inval­i­dates some­thing test­ing orga­ni­za­tions spent the last ten years build­ing around.

Most of the test­ing play­books in use today were designed for an envi­ron­ment SAP is leav­ing behind. UI-based replay tools, which car­ried the pre­vi­ous wave of SAP test automa­tion, work by record­ing what a user does on screen and replay­ing it lat­er. They depend on the screen look­ing the same the next time. Fiori doesn’t coop­er­ate; quar­ter­ly updates rou­tine­ly shift things around, and scripts that worked last quar­ter need to be repaired this one.

When the team’s time is going into fix­ing what used to work, new cov­er­age just doesn’t get built. The data side of the prob­lem is dif­fer­ent, but adds up to the same thing. If your test data is some­thing the QA team designed rather than some­thing pulled out of the pro­duc­tion sys­tem, you’re test­ing what peo­ple thought cus­tomers do, not what cus­tomers actu­al­ly do.

The first month-end close after go-live is usu­al­ly where this catch­es up with you. Spe­cial dis­counts that nobody flagged, for­eign-cur­ren­cy reval­u­a­tions on trans­ac­tions the test scripts nev­er saw, tax rules from a coun­try that wasn’t in the test scope. These show up the same week, and most of them weren’t in any pre-go-live regres­sion run.

These two issues com­pound each oth­er. Time spent repair­ing scripts reduces resources for new cov­er­age. Missed excep­tions extend Hyper­care beyond its planned dura­tion, redi­rect­ing bud­gets meant for future automa­tion. By the next quar­ter­ly release, teams often stop using test­ing tools, dash­boards become out­dat­ed, and regres­sion is tracked man­u­al­ly in spreadsheets.

What AI Doesn’t Fix

The ven­dors haven’t ignored these prob­lems, and most of the major test­ing plat­forms have added some kind of AI fea­ture aimed at one or anoth­er of them. Self-heal­ing is the most vis­i­ble: when a UI ele­ment moves, the tool tries to fig­ure out where it went and rewires the bro­ken script accordingly.

There’s also a cat­e­go­ry of fea­tures around gen­er­a­tion — describe a sce­nario in plain lan­guage, get back some­thing resem­bling a test — that ven­dors are pitch­ing under the agen­tic-AI ban­ner. And on the analy­sis side, some tools now use AI to pre­dict which parts of a sys­tem a change will touch, which lets teams scope regres­sion more nar­row­ly than run­ning every­thing every time.

The fea­tures are use­ful in their own right, but they all run into the same wall when the under­ly­ing tool is UI-based. SAP’s UI is also its most volatile lay­er. Anchor­ing automa­tion to a lay­er that changes con­stant­ly is a struc­tur­al prob­lem no AI fea­ture can solve, because the AI is repair­ing dam­age the archi­tec­ture didn’t have to pro­duce in the first place.

A Dif­fer­ent Architecture

Per­fecTwin, the SAP test­ing solu­tion from LG CNS, takes the oppo­site approach. Rather than record­ing user inter­ac­tions and replay­ing them through a brows­er, it sends test inputs straight to the SAP appli­ca­tion lay­er, where cal­cu­la­tions, data flows, and busi­ness rules are decid­ed. From that archi­tec­tur­al choice, four rein­forc­ing capa­bil­i­ties follow.

1. Native SAP design. Per­fecTwin treats table struc­tures, SAP sys­tem mes­sages, busi­ness objects, and T‑Codes as first-class ele­ments of the test­ing mod­el rather than as fields and but­tons to be locat­ed on a screen. Sce­nar­ios assem­ble along the con­tours of how SAP actu­al­ly works, not along the con­tours of how SAP is rendered.
2. Back­end exe­cu­tion is what pro­duces the speed advan­tage. LG CNS reports exe­cu­tion speeds up to 50 times faster than UI-based alter­na­tives, which mat­ters when thou­sands of regres­sion tests need to fin­ish inside an upgrade window.
3. Real pro­duc­tion data, drawn straight from the oper­a­tional data­base through a Data Extrac­tor. Users spec­i­fy the busi­ness area and time win­dow they want cov­ered and receive the match­ing set of live trans­ac­tions back as test inputs. The edge cas­es that sim­pli­fied test data nev­er pro­duce — the mul­ti-cur­ren­cy set­tle­ments, the unusu­al dis­count struc­tures, the reg­u­la­to­ry vari­a­tions of a glob­al foot­print — show up because they’re already in the data. The dis­tance between test and pro­duc­tion envi­ron­ments nar­rows because the test data is the pro­duc­tion data.
4. No-code sce­nario assem­bly. PerfecTwin’s flow-dia­gram inter­face lets func­tion­al con­sul­tants and busi­ness users build sce­nar­ios by snap­ping reusable units togeth­er rather than writ­ing scripts, draw­ing on a library of pre-built tem­plates for stan­dard SAP process­es. Onboard­ing takes days rather than months, which means automa­tion stops being the domin­ion of a sin­gle spe­cial­ist whose depar­ture leaves the entire pro­gram stranded.

These four capa­bil­i­ties rein­force each oth­er in a way that the indi­vid­ual fea­tures don’t ful­ly con­vey. Back­end exe­cu­tion is what keeps the no-code library from decay­ing through Fiori updates, and real pro­duc­tion data is what makes the speed advan­tage actionable.

Three Out­comes Across the Lifecycle

Three cus­tomer cas­es show the sys­tem at work across the three life­cy­cle stages where SAP test­ing has his­tor­i­cal­ly bro­ken down.

The migra­tion case is the most demand­ing of the three, and it’s where LG CNS has its strongest ref­er­ence: a glob­al man­u­fac­tur­er that ran end-to-end val­i­da­tion against more than 10,000 sce­nar­ios across over 50 trans­ac­tion codes dur­ing its ECC-to‑S/4HANA con­ver­sion, includ­ing full cov­er­age of cross-enti­ty trans­ac­tions and glob­al sys­tem inte­gra­tion touch­points. That cus­tomer went live with zero day-one defects and cut its ver­i­fi­ca­tion time by 40%, an out­come that wasn’t going to come out of sam­ple-data test­ing because the rel­e­vant edge cas­es weren’t in the samples.

The oth­er two stages of the life­cy­cle look dif­fer­ent in their par­tic­u­lars. A finan­cial insti­tu­tion run­ning quar­ter­ly releas­es used the tool to con­sol­i­date its Group­ware-to-SAP work­flow into one end-to-end test, then ran that test against actu­al ECC data ahead of cutover. The team caught RFC inte­gra­tion errors before they reached pro­duc­tion, and the institution’s upgrade ver­i­fi­ca­tion time came down 70%. A sep­a­rate engage­ment, this one with a glob­al chem­i­cal com­pa­ny, replaced an entrenched man­u­al regres­sion rou­tine with an auto­mat­ed regres­sion pro­gram weight­ed by process fre­quen­cy, busi­ness risk, and ROI. Oper­a­tional inci­dents were down 80%.

LG CNS is bring­ing the AI-enabled cloud ver­sion of Per­fecTwin to SAP Sap­phire 2026, and the fram­ing mat­ters as much as the fea­tures. AI is what every test­ing ven­dor is rac­ing to add right now, but its lever­age depends on what it’s applied to. Lay­ered onto UI replay, it’s a sophis­ti­cat­ed patch on a frag­ile foun­da­tion. Built into a back­end-direct, SAP-native, pro­duc­tion-data-dri­ven archi­tec­ture, it extends a struc­ture that already addressed the under­ly­ing problem.

The 2027 dead­line is what’s forc­ing the ques­tion. The cus­tomers who answer it by look­ing at where their test­ing actu­al­ly runs, not just which tool runs it, give them­selves an archi­tec­ture that holds up after migra­tion ends.

Bri­an Rogan is Growth Leader, U.S. at LG CNS PerfecTwin.