Down­load the full inter­view here. 

As the gen­er­al man­ag­er and VP of SAP prod­ucts at insight­soft­ware, Axel Stre­ichardt inter­faces direct­ly with many of the key chal­lenges busi­ness­es face while prepar­ing to migrate to SAP S/4HANA. As the end of ECC main­te­nance approach­es next year, com­pa­nies con­tin­ue to find them­selves embark­ing on 18-month projects. To com­ple­ment this tran­si­tion, insight­soft­ware offers ser­vices focused on gaps in cross-sys­tem inte­gra­tion, self-ser­vice ana­lyt­ics, advanced ana­lyt­ics and trend analy­sis, BI tool inte­gra­tion, AI-enhanced intel­li­gence, and process automation.

A recent ASUG sur­vey found that while 39% of SAP land­scapes are build­ing foun­da­tion­al knowl­edge for AI imple­men­ta­tion and 41% are amid active exper­i­men­ta­tion or pilot­ing, only 10% of those sur­veyed have been able to achieve enter­prise-scale rollout.

Fol­low­ing SAP’s announce­ment of its vision for the Autonomous Enter­prise at this year’s SAP Sap­phire & ASUG Annu­al Con­fer­ence, Stre­ichardt sat down to dis­cuss the data reme­di­a­tion that has to pre­cede every migra­tion, AI’s role in iden­ti­fy­ing pre­vi­ous­ly uncaught errors, and the future of inte­gra­tion with SAP products.

This inter­view has been edit­ed and con­densed for length and clarity.

Q: You’ve spent more than 20 years work­ing with SAP, at SAP itself, and at AWS, EMC, and Pure Stor­age before insight­soft­ware. Across all those seats, what do SAP cus­tomers con­sis­tent­ly get wrong about a tran­si­tion like this?

Cus­tomers struc­ture migra­tion around capa­bil­i­ty readi­ness: Do we have the right infra­struc­ture? Do we have to be in the cloud? Is the appli­ca­tion con­fig­ured? What they don’t ask is, ​“What data do we actu­al­ly have, and is it fit for pur­pose on the oth­er side?” That’s the dis­crep­an­cy — they have no idea how bro­ken their data is until they’re halfway through. If 30% is bro­ken, that’s a six-week reme­di­a­tion effort, and some­one owns this data going for­ward or it decays again. What­ev­er data qual­i­ty sins you let slip will come back to haunt you when you try to acti­vate AI 18 months later.

Q: At SAP Sap­phire, SAP put its vision for the autonomous enter­prise at the cen­ter of every­thing: agents orches­trat­ing end-to-end process­es, appli­ca­tions that exe­cute work rather than record it. What’s your hon­est read on that vision?

At SAP Sap­phire, I was talk­ing about the $15 mil­lion gap that every CFO has. That’s real, not abstract — it’s los­ing mon­ey because you can’t see the gaps in your data. In one con­ver­sa­tion, a CFO said, ​“I have vis­i­bil­i­ty for 60% of our spend. The rest is dark data. I don’t know if we’re over­pay­ing sup­pli­ers, miss­ing vol­ume dis­counts, or car­ry­ing redun­dant inven­to­ry, because I can’t con­nect the dots between pro­cure­ment, finance, and sup­ply chain.”

Mar­gin leak­age hap­pens because your data sits in silos, and sup­ply chain issues are the worst. A cus­tomer dis­cov­ered they were sys­tem­at­i­cal­ly pay­ing pre­mi­um freight costs for expe­dit­ed ship­ments that were going to the wrong ware­house because their order sys­tem wasn’t talk­ing to their inven­to­ry sys­tem. That cost them $2.3 mil­lion annu­al­ly. The data gap was invis­i­ble, but then we piv­ot­ed to SAP autonomous enter­prise agents orches­trat­ing work­flows. Con­nect your mas­ter data so finance, sup­ply chain, and oper­a­tions are talk­ing, then lay­er AI on top to auto­mate find­ing prob­lems. We’ve been doing that foun­da­tion work for over 20 years. Adding AI is the log­i­cal next step, which is why we launched AI automa­tion capa­bil­i­ties at the end of June to sur­face cost and sup­ply-chain chal­lenges auto­mat­i­cal­ly that usu­al­ly go undetected.

Q: Only about 1 in 10 cus­tomers have scaled AI past a pilot, and most are after some­thing more mod­est than full auton­o­my — AI that sup­ports peo­ple rather than replaces them. How do you square where SAP is point­ing with where cus­tomers actu­al­ly are?

CFOs and sup­ply chain lead­ers want AI and have fund­ing for AI projects, but first they need to trust it. Trust only comes one way: The AI has to be right. So, the con­ver­sa­tion usu­al­ly starts dif­fer­ent­ly — not ​“Here’s an AI agent that will auto­mate your deci­sions,” but ​“Here’s AI that will give you the right answers to your hard­est ques­tions.” Once the cus­tomer sees that the analy­sis is accu­rate, you can move to proac­tive and have the sys­tem auto­mat­i­cal­ly flag finan­cial anom­alies and sup­ply chain challenges.

Q: Mas­ter data is one of the biggest unsolved prob­lems our mem­bers report. This is your home turf. When you get into a customer’s envi­ron­ment, what’s actu­al­ly bro­ken in how their data sits?

Mas­ter data lives in silos, so there’s no sin­gle source of truth for iden­ti­ty. A glob­al man­u­fac­tur­er has GL codes in SAP, cost cen­ter codes in their sup­ply chain sys­tem, account codes in a lega­cy order sys­tem — same cost, three dif­fer­ent iden­ti­fiers. When finance asks, ​“What did we spend on this cost cen­ter?”, the answer depends on which sys­tem you’ve asked. Nobody owns rec­on­cil­i­a­tion, and there’s no data val­i­da­tion process across sys­tems, so bad data prop­a­gates from sys­tem to system.

Q: A lot of your cus­tomers are still mid-migra­tion, and mov­ing to S/4HANA is the sin­gle biggest chal­lenge our mem­bers name. Mean­while, RISE now com­mits them to switch­ing on Joule assis­tants in year one. How does the AI con­ver­sa­tion land for a com­pa­ny that hasn’t fin­ished the migration?

It doesn’t, real­ly. One CIO made it very clear to me: ​“We are still load­ing mas­ter data and fix­ing cutover issues. Joule is year two at the ear­li­est.” Migra­tions and Joule are com­pet­ing for the same resources, and migra­tion wins. So, we’re help­ing cus­tomers today with AI to accel­er­ate the migra­tion itself dur­ing that bru­tal cleanup and mas­ter data reme­di­a­tion phase. That gets them to clean data faster when Joule arrives. They can start using AI today to win their migration.

Q: ASUG research found that how a com­pa­ny approach­es AI pre­dicts suc­cess bet­ter than its size or bud­get does. What sep­a­rates the cus­tomers pulling ahead from those stuck in pilots?

The ones that win start with a process, not a mod­el: ​“Here’s the process that’s bro­ken. If we fix this, here’s the busi­ness val­ue. Can AI help?” Often the answer is ​“no,” because you need gov­er­nance first, but when the answer is ​“yes,” they have clear suc­cess cri­te­ria from day one. Pilots that fail usu­al­ly start with, ​“Hey, let’s build an ML mod­el to pre­dict X.”

Sec­ond, suc­cess­ful cus­tomers mea­sure busi­ness out­comes, not mod­el accu­ra­cy. Pilots that fail mea­sure ​“We are reduc­ing month and close by x days.” Mod­el accu­ra­cy is table stakes. Out­comes are what real­ly matter.

Third, suc­cess­ful cus­tomers have a point per­son with actu­al author­i­ty. Pilots that scale have a CFO or CEO, or even a VP of finance, who spon­sors the work and takes account­abil­i­ty for results.

Con­nect with Axel Stre­ichardt on LinkedIn. 

Q: You frame AI as increas­ing trust in the data. At the same time, gov­er­nance and secu­ri­ty are the top con­cerns cus­tomers cite when it comes to adopt­ing AI. How do you rec­on­cile those, and why should a CFO trust an AI-gen­er­at­ed insight more than the spread­sheets they’ve used for 15 years?

Fif­teen years of spread­sheets have not solved the $15 mil­lion gap I talked about. A spread­sheet is pas­sive; it shows you what you already know to look for. A CFO then has to hunt through rows of data to find anom­alies, vari­ances, and hid­den prob­lems. AI is dif­fer­ent. Gov­er­nance doesn’t con­strain AI; it enables trust. If I can see how the AI mod­el was trained, which rule it’s over­rid­ing, or who approved the excep­tion, I can trust it.

Here’s a con­crete exam­ple: We imple­ment­ed auto­mat­ed GL rec­on­cil­i­a­tion for a finance ser­vice firm. The AI sys­tem flags rec­on­cil­ing items out­side nor­mal vari­ances and bands and sur­faces root cause anom­alies auto­mat­i­cal­ly. Every flagged item gets a rea­son code and val­i­da­tion rules, and approvals are locked with busi­ness jus­ti­fi­ca­tion. Finance went from 40 hours a week on GL-rec­on­cil­i­a­tion to under an hour.

Q: Your posi­tion­ing has been to com­ple­ment SAP rather than replace it. SAP has acquired Rel­tio and Dremio, and the com­pa­ny is build­ing SAP Busi­ness Data Cloud to com­bine SAP and non-SAP data, an area close to what you’ve described as your own. How do you see the rela­tion­ship evolv­ing from here?

SAP owns the trans­ac­tion sys­tem and process lay­er, but what SAP doesn’t own is the work of inte­grat­ing SAP data with sup­ply chain plat­forms and non-SAP ERP lega­cy sys­tems like Ora­cle or Sales​force​.com. We’ve been doing data inte­gra­tions for 20-plus years across thou­sands of cus­tomer engage­ments. That expe­ri­ence can’t be acquired — it has to be built. That’s why SAP part­ners with us as a SolEx part­ner. They don’t want to replace us; they want to work with us.

Q: Microsoft’s Copi­lot is in active use at most orga­ni­za­tions today, while adop­tion of SAP’s own embed­ded AI is still ear­ly. Your prod­ucts work across both. How should a cus­tomer think about where to place their bets while the AI lay­er of their land­scape is still tak­ing shape?

You don’t want to tell some­one that the invest­ments they did, the mil­lions of dol­lars they spent, should be replaced by some­thing else. That’s why the answer isn’t choos­ing one over the oth­er; it’s let­ting both coex­ist. Copi­lot is a gen­er­al intel­li­gence lay­er: sum­ma­rize this vari­ance report, draft an email explain­ing this. SAP is embed­ded AI trained on SAP data and process­es. We help cus­tomers inte­grate both into their stan­dards, process­es, and governance. 

A real exam­ple: a cus­tomer is using SAP AI to iden­ti­fy the top vari­ance to sug­gest the GL codes. Finance ana­lysts review the rec­om­men­da­tion, then use Copi­lot to draft an expla­na­tion for the CFO. So, SAP AI isn’t being replaced by Copi­lot; they’re com­ple­men­tary. One finds the prob­lem, the oth­er helps com­mu­ni­cate the solu­tion, and we are the plat­form that makes both work togeth­er seamlessly.

Q: Pic­ture the prac­ti­tion­er who came home from Sap­phire ener­gized and now faces their actu­al land­scape on Mon­day. What are the unglam­orous first steps that nev­er make it into a keynote slide before any of this is real?

Audit your data. This prob­a­bly takes four to six weeks, and it’s bor­ing. You’re run­ning SQL queries, sam­pling records, check­ing for orphan entries. Start with your most crit­i­cal data sets: GL accounts, cost cen­ters, cus­tomer mas­ter, sup­pli­er mas­ter. Pick one sam­ple. Once you under­stand the scope of the prob­lem, you have a baseline.

Sec­ond, doc­u­ment one crit­i­cal process end-to-end. Pick a process that’s man­u­al, with lots of human steps involved, error-prone — for instance, close oper­a­tion — then doc­u­ment it, not as a process map but as an actu­al walk­through. Who does what, when, with what sys­tems? Where are the hand­offs? This is one week of work.

Third, fix your mas­ter data, start­ing with the high­est-risk items. Don’t try to fix every­thing. Pick the data that cre­ates the most busi­ness impact when it’s wrong for finance, GL accounts, and cost cen­ters. Audit those spe­cif­ic data sets, iden­ti­fy the junk, and cre­ate a reme­di­a­tion plan. This is six to eight weeks of work. Base­line: expect 20 – 30% of mas­ter data to need reme­di­a­tion; com­pa­nies that do this upfront save 60% of migra­tion rework downstream.

Q: A lot of ven­dors talk about these AI capa­bil­i­ties, and when you ask whether they’re real­ly there, the answer is often no. Where do you see the gap between ambi­tion and real­i­ty right now, and what should cus­tomers be most dis­cern­ing about?

What frus­trates me the most is when ven­dors put a lit­tle icon on the slides for data cleanup, data con­text, data enhance­ment, then spend the rest of their pre­sen­ta­tion on flashy dash­boards. Every­thing is focused on the results, not on how to get to these results. You can’t auto­mate your way around the foun­da­tion­al work of data cleanup con­text and enhance­ments. So, cus­tomers should be dis­cern­ing about ven­dors who gloss over the 80% data cleanup con­text and enhance­ment, because it’s invis­i­ble and hard to demo. Ask them, ​“How many cus­tomers have you actu­al­ly done this with? What expe­ri­ence do you have in data cleanup and enhance­ment at scale?” That’s where the real val­ue lives, even if it’s the hard­est to show.

Q: If every cus­tomer read­ing this took away one thing before chas­ing that vision of SAP’s autonomous enter­prise, what would it be?

Get the fun­da­men­tals right first. The autonomous enter­prise vision is real, no ques­tion. It’s also three to five years away for most orga­ni­za­tions. You can’t get there from here by buy­ing more soft­ware; you get there by doing hard work first. Don’t chase autonomous enter­prise — chase the next lev­el of oper­a­tional maturity.

Vis­it the insight­soft­ware web­site.