This Part­ner Insight was authored by Tra­cy Ben­nett, Senior Vice Pres­i­dent, Amer­i­c­as at PiLog Group.

For most of its his­to­ry, mas­ter data at util­i­ty orga­ni­za­tions was nobody’s job — a man­u­al, unglam­orous data entry func­tion, treat­ed as a nec­es­sary cost of doing busi­ness rather than a strate­gic asset. Nobody ques­tioned the process, because nobody believed there was a bet­ter one.

But in recent years, as data’s role in inno­va­tion has grown, some orga­ni­za­tions — guid­ed by part­ners like PiLog Group, which has spent three decades work­ing with asset-inten­sive indus­tries — have begun to chal­lenge this assump­tion, reimag­in­ing mas­ter data as the foun­da­tion for busi­ness advantage.

The teams who depend­ed on data began tak­ing own­er­ship of it. Busi­ness units built cen­ters of excel­lence. Shared ser­vices func­tions claimed their domains. Mate­ri­als, assets, finan­cials, and ven­dors each devel­oped their own stan­dards, gov­er­nance, and tools. On paper, this looked like progress.

In prac­tice, it built walls. The silos that formed around data own­er­ship meant asset records, main­te­nance his­to­ries, and mate­r­i­al mas­ters were cre­at­ed in par­al­lel and in iso­la­tion, and then stitched togeth­er after the fact through slow, expen­sive, error-prone process­es. The data entry prob­lem nev­er dis­ap­peared; it just became more com­pli­cat­ed, and more expen­sive to fix.

The Board Wants AI. The Data Says Otherwise.

When AI arrived, util­i­ties respond­ed the way boards want­ed them to: fast. Gartner’s 2025 CIO sur­vey found that 94% of pow­er and util­i­ty CIOs planned to increase AI invest­ment that year, with aver­age spend­ing up 38.3%, and Gart­ner expect­ed 40% of util­i­ties to deploy AI-dri­ven con­trol room oper­a­tors by 2027. ICF’s util­i­ty sur­vey found every respon­dent already using AI some­where, with 65% call­ing it a ​“game chang­er.” Itron’s sur­vey of 500 North Amer­i­can util­i­ty exec­u­tives put adop­tion at 81%, led by grid opti­miza­tion, safe­ty, and demand forecasting.

While the invest­ment in AI is mas­sive, the results are more uneven. S&P Global’s 2026 research found that only 43% of AI projects launched are expect­ed to deliv­er ROI with­in 12 months, even as spend­ing keeps climb­ing. Deloitte puts the share of ener­gy, resources, and indus­tri­al com­pa­nies using AI to fun­da­men­tal­ly trans­form their busi­ness at just 23%. 

Set those num­bers side by side, and a pat­tern appears. Util­i­ties are mov­ing fast on AI and slow on the one thing that AI suc­cess depends on: data.

Indus­try ana­lysts now say it plain­ly: busi­ness­es do not have an AI prob­lem; they have a data prob­lem. In util­i­ties specif­i­cal­ly, asset hier­ar­chies, func­tion­al loca­tions, mate­ri­als, and main­te­nance his­to­ries were built over decades by dif­fer­ent teams, in dif­fer­ent sys­tems, using dif­fer­ent nam­ing con­ven­tions, and for dif­fer­ent pur­pos­es. AI inher­its that incon­sis­ten­cy and sur­faces it faster than any tech­nol­o­gy before.

This is not a rea­son to dis­trust AI, but instead a rea­son to be hon­est about what AI needs from an orga­ni­za­tion before it can deliv­er what that orga­ni­za­tion wants.

Two Util­i­ties, One Famil­iar Problem

At a major water util­i­ty oper­a­tor, more than 23,000 mate­r­i­al records in SAP car­ried dupli­cate iden­ti­ties, no UNSP­SC clas­si­fi­ca­tion, and no cross-site vis­i­bil­i­ty. Pro­cure­ment deci­sions were being made against data that no longer matched phys­i­cal real­i­ty, and non-mov­ing spares were absorb­ing cap­i­tal with no oper­a­tional justification. 

PiLog Group — a glob­al spe­cial­ist in mas­ter data man­age­ment, data qual­i­ty gov­er­nance, and EAM solu­tions — inter­vened and brought the cat­a­log under con­trol with a struc­tured mas­ter data gov­er­nance pro­gram cov­er­ing dedu­pli­ca­tion, four-lev­el UNSP­SC clas­si­fi­ca­tion, and ISO 8000-aligned enrichment. 

The result: a 27% reduc­tion in inven­to­ry val­ue, 15% sav­ings in pro­cure­ment and inven­to­ry costs, and enter­prise-wide vis­i­bil­i­ty into mate­ri­als for the first time.

At anoth­er large elec­tric­i­ty, water, and gas util­i­ty oper­a­tor, the chal­lenge sat one lev­el down, at the phys­i­cal assets them­selves. Sub­sta­tion equip­ment records lacked GPS coor­di­nates, had unread­able name­plates, and could not be rec­on­ciled with what was actu­al­ly installed in the field. 

In this ongo­ing project, PiLog Group has inspect­ed, pho­tographed, GPS-tagged, and clas­si­fied hun­dreds of assets across the utility’s sub­sta­tions, val­i­dat­ing every record against sin­gle-line dia­grams before it reached SAP EAM. The out­come is the pre­con­di­tion for AI deploy­ment — asset data that the utility’s sys­tems, and even­tu­al­ly its AI, could trust.

Both cas­es share a struc­ture worth nam­ing. The invest­ment that unlocked val­ue was in the data foun­da­tion that AI depends on, not in AI capability.

What AI Changed Isn’t the Prob­lem. It’s the Fix.

Util­i­ties have always known their data had gaps. Beyond aware­ness, what has gen­uine­ly changed is the tool­ing avail­able to close them. 

Work that once required months of man­u­al field sur­veys, engi­neers inter­pret­ing draw­ings, and data teams build­ing hier­ar­chies by hand can now be sub­stan­tial­ly accel­er­at­ed by AI-assist­ed extrac­tion and clas­si­fi­ca­tion, pro­vid­ed the under­ly­ing con­tent, con­text, and gov­er­nance are in place first. 

This does not elim­i­nate the need for sub­ject mat­ter experts. It moves them from data entry to val­i­da­tion, which is a bet­ter use of their time and a faster path to AI that can actu­al­ly be trust­ed with oper­a­tional decisions.

The Math Nobody Puts in the Board Deck

It is worth being pre­cise about why this mat­ters — both finan­cial­ly and oper­a­tional­ly. Rev­enue-gen­er­at­ing ini­tia­tives typ­i­cal­ly return 10 – 15% of oper­at­ing income for every dol­lar invest­ed. Oper­a­tional sav­ings from elim­i­nat­ing dupli­cate records, right-siz­ing inven­to­ry, and reduc­ing unplanned down­time flow to the bot­tom line at close to 100%. 

Mas­ter data qual­i­ty and gov­er­nance pro­grams sit almost entire­ly in the sec­ond cat­e­go­ry. Even though they are rarely the excit­ing line in an AI roadmap, they are the one that pays for itself the fastest.

Mas­ter data was nobody’s job for a long time because nobody had to answer for what it cost. AI has changed that qui­et­ly but per­ma­nent­ly. Every AI ini­tia­tive that stalls after the pilot stage now has a trace­able cause, and it is rarely the mod­el itself. 

The util­i­ties that address their data foun­da­tion before their next AI ini­tia­tive, rather than after it dis­ap­points them, will be the ones that scale past the pilot stage while their peers explain to the board why the num­bers still do not add up.

Learn more about the PiLog Data Qual­i­ty Gov­er­nance (DQG) Suite, which pro­vides end-to-end data har­mo­niza­tion, enrich­ment, and gov­er­nance for orga­ni­za­tions pur­su­ing reli­able ERP per­for­mance and agen­tic AI readiness.

Tra­cy Ben­nett is Senior Vice Pres­i­dent, Amer­i­c­as at PiLog Group.