This part­ner insight was authored by Kyle Bialek, Man­ag­ing Direc­tor at KPMG. 

Many orga­ni­za­tions are invest­ing heav­i­ly in AI to dri­ve bet­ter fore­cast­ing, smarter asset deci­sions, improved cus­tomer expe­ri­ence, and clean­er reg­u­la­to­ry report­ing. On the sur­face, the speed and results look great: clean dash­boards, con­fi­dent pre­dic­tions, faster decisions.

What most peo­ple don’t see is every­thing required under­neath to make those results trust­wor­thy. At the cen­ter of that foun­da­tion is the trust­ed busi­ness con­text lay­er: the busi­ness def­i­n­i­tions, seman­tic rela­tion­ships, own­er­ship, and lin­eage that pre­serve mean­ing across decades of orga­ni­za­tion­al change and shift­ing nomen­cla­ture. With­out it, AI pro­duces answers that sound right but aren’t, and trust is quick­ly lost.

Con­text is not a tech­ni­cal solu­tion — it is the lan­guage of your busi­ness. Orga­ni­za­tions that treat it that way turn AI exper­i­ments into com­pet­i­tive advantages.

The Tip of the Iceberg

AI is not new to util­i­ties; many are see­ing real results. Pre­dictabil­i­ty of fail­ures can cut unplanned out­ages. Smart-meter bill fore­cast­ing and high-bill alerts help improve cus­tomer expe­ri­ence and reduce call-cen­ter vol­ume. Auto­mat­ed report­ing helps speed up fil­ings and reduces errors.

Boards and reg­u­la­tors have been push­ing for these out­comes. The prob­lem is they stay reli­able only when the foun­da­tion under­neath is sol­id, and demand from the board and reg­u­la­tors con­tin­ues to grow. When data lacks clear mean­ing, con­sis­tent def­i­n­i­tions, and clean his­to­ry, out­comes drift and com­mit­ments become hard­er to meet.

The Ice­berg Effect

Think of an ice­berg. The small part above the water is what lead­ers see — AI insights, dash­boards, auto­mat­ed deci­sions. The much larg­er mass below the water­line is the work that makes those insights trust­wor­thy: data qual­i­ty, gov­er­nance, trust­ed busi­ness def­i­n­i­tions, seman­tic rela­tion­ships, lin­eage, and espe­cial­ly busi­ness context.

Many pro­grams spend the major­i­ty of their atten­tion on the tip. They invest in mod­els and use cas­es while under-invest­ing in the foun­da­tion that deter­mines whether those invest­ments will scale. In util­i­ties, where deci­sions affect reli­a­bil­i­ty, billing accu­ra­cy, safe­ty, and reg­u­la­to­ry stand­ing, that imbal­ance can have real consequences.

Why Con­text Matters

Con­text answers the basic busi­ness ques­tions raw data can­not: What does this num­ber actu­al­ly mean? Who owns it? How was it cal­cu­lat­ed? When was it last updat­ed? What oth­er names does this term go by?

In util­i­ties, these ques­tions are rarely sim­ple. The same term — cus­tomer, meter, out­age, work order, rev­enue — can mean dif­fer­ent things across oper­a­tions, cus­tomer ser­vice, finance, and reg­u­la­to­ry teams. With­out a shared def­i­n­i­tion, AI sys­tems fill the gaps with incom­plete or con­flict­ing logic.

The poten­tial busi­ness impacts can show up quickly:

• Deci­sion qual­i­ty drops: Fore­casts and rec­om­men­da­tions rest on shaky definitions.
• Trust erodes: When peo­ple can­not explain or defend an AI rec­om­men­da­tion, they stop using it.
• Reg­u­la­to­ry expo­sure grows: Clear lin­eage and mean­ing become hard­er to demonstrate.
• Scal­ing gets expen­sive: Every new use case requires cus­tom data work instead of build­ing on a con­sis­tent foundation.

Strong con­text revers­es this by estab­lish­ing a com­mon lan­guage, let­ting AI oper­ate with the same con­text expe­ri­enced employ­ees already use. Like humans, AI solu­tions can­not accu­rate­ly respond with­out the insti­tu­tion­al knowl­edge you have amassed through­out your career.

This foun­da­tion becomes even more impor­tant as util­i­ties mod­ern­ize their tech­nol­o­gy land­scapes with plat­forms such as SAP S/4HANA, SAP Ana­lyt­ics Cloud, and SAP Busi­ness Data Cloud. These tech­nolo­gies can enable sig­nif­i­cant advances in ana­lyt­ics, plan­ning, and AI, but they still depend on clear busi­ness mean­ing and trust­ed con­text. With­out it, even sophis­ti­cat­ed tools risk deliv­er­ing pol­ished but unre­li­able outputs.

What This Looks Like in Practice

Dur­ing a recent enter­prise plan­ning and AI readi­ness effort at a large North Amer­i­can elec­tric util­i­ty, the ini­tial focus was on high-val­ue out­comes such as asset avail­abil­i­ty pre­dic­tion, finan­cial fore­cast­ing, and intel­li­gent deci­sion sup­port. As the work pro­gressed, it became clear that tech­nol­o­gy itself was not the pri­ma­ry con­straint. Sig­nif­i­cant effort was required to har­mo­nize plan­ning data, align busi­ness def­i­n­i­tions and report­ing seman­tics, and estab­lish con­sis­tent gov­er­nance across finance and oper­a­tional systems.

In one com­mon exam­ple, teams dis­cov­ered that terms such as ​“asset,” ​“loss,” and relat­ed avail­abil­i­ty met­rics car­ried dif­fer­ent def­i­n­i­tions and cal­cu­la­tion rules across oper­a­tions, account­ing, and report­ing. Until those mean­ings were aligned, the under­ly­ing data remained sparse and incon­sis­tent, mak­ing it dif­fi­cult to build reli­able mod­els or pro­duce AI-assist­ed fore­casts that could be trusted.

The expe­ri­ence rein­forced a broad­er les­son for util­i­ties: enter­prise AI readi­ness begins with trust­ed busi­ness context.

The Work That Has to Hap­pen Below the Surface

Reli­able AI requires delib­er­ate invest­ment in the sub­merged lay­ers — the accu­mu­lat­ed knowl­edge that made your enter­prise what it is today.

It starts with busi­ness lead­ers defin­ing and main­tain­ing the mean­ing of crit­i­cal data ele­ments. Leav­ing inter­pre­ta­tion only to IT or data teams does not work. A liv­ing busi­ness glos­sary that reflects how the orga­ni­za­tion actu­al­ly oper­ates is essential.

It con­tin­ues with con­text — how data moves from source sys­tems through the lay­ers and into the AI envi­ron­ment. Lin­eage needs to be vis­i­ble so that when results are ques­tioned, the path back is transparent.

Trust­ed busi­ness con­text also must be built into ini­tia­tives from the start. Treat­ing it as doc­u­men­ta­tion after the fact almost always fails. The orga­ni­za­tions that get this right treat busi­ness con­text as an enter­prise asset with defined stan­dards, ongo­ing main­te­nance, and sus­tained investment.

Final­ly, gov­er­nance should con­nect data and AI teams with the busi­ness domains that own the out­comes. With­out that link, even the newest tools pro­duce frag­ment­ed results.

None of this work is glam­orous, and it rarely shows up in the suc­cess sto­ries. But it can deter­mine whether those suc­cess sto­ries can be sus­tained and scaled.

Lessons from Enter­prise Implementation

Two pat­terns stand out from enter­prise plan­ning and AI pro­grams. First, orga­ni­za­tions that treat trust­ed busi­ness con­text as foun­da­tion­al infra­struc­ture — rather than a side activ­i­ty — move more quick­ly from iso­lat­ed exper­i­ments to repeat­able capa­bil­i­ties. Sec­ond, enter­prise plan­ning is often an effec­tive prov­ing ground: fore­cast­ing, sce­nario analy­sis, and dri­ver-based plan­ning force finan­cial and oper­a­tional data to car­ry con­sis­tent mean­ing across the orga­ni­za­tion, mak­ing gaps in def­i­n­i­tions and gov­er­nance vis­i­ble early.

The Cost of Weak Foundations

When orga­ni­za­tions under-invest below the water­line, the prob­lems often sur­face grad­u­al­ly and then all at once. Fore­casts become hard­er to rec­on­cile, AI rec­om­men­da­tions become dif­fi­cult to explain, teams recre­ate def­i­n­i­tions for each new use case, and reg­u­la­to­ry or audit ques­tions become dif­fi­cult to answer. The result is not sim­ply a data prob­lem; it is slow­er adop­tion, high­er cost, and dimin­ished trust in AI-enabled decisions.

Rec­om­men­da­tions

Lead­ers who want AI to deliv­er sus­tain­able val­ue should treat trust­ed busi­ness con­text as an enter­prise capa­bil­i­ty rather than a one-time data exercise.

• Make busi­ness con­text a lead­er­ship pri­or­i­ty: Estab­lish named busi­ness own­ers for the crit­i­cal terms and data prod­ucts that AI and ana­lyt­ics will touch.
• Build con­text into ini­tia­tives from the start: Do not treat def­i­n­i­tions, lin­eage, and gov­er­nance as doc­u­men­ta­tion to be com­plet­ed after implementation.
• Design for reuse: Cre­ate trust­ed busi­ness con­text that can sup­port mul­ti­ple fore­cast­ing, plan­ning, report­ing, and AI use cas­es rather than rebuild­ing mean­ing for each project.
• Mea­sure trust as well as tech­ni­cal per­for­mance: Mod­el accu­ra­cy mat­ters, but so do explain­abil­i­ty, con­sis­ten­cy, lin­eage, and the abil­i­ty of busi­ness users to under­stand and defend AI-sup­port­ed decisions.

Clos­ing: Ris­ing Above the Waterline

The next gen­er­a­tion of enter­prise AI will not be dif­fer­en­ti­at­ed sim­ply by access to increas­ing­ly pow­er­ful mod­els. Those capa­bil­i­ties will become broad­ly avail­able. The advan­tage will come from how well orga­ni­za­tions con­nect those mod­els to the lan­guage, mean­ing, and oper­at­ing con­text of their businesses.

For util­i­ties, that foun­da­tion sits below the water­line: trust­ed def­i­n­i­tions, seman­tic rela­tion­ships, lin­eage, gov­er­nance, and enter­prise data that car­ries con­sis­tent mean­ing from source to deci­sion. Orga­ni­za­tions that invest in this foun­da­tion will be bet­ter posi­tioned to deploy AI, trust it, explain it, and scale it.

The most vis­i­ble AI out­comes may hap­pen above the water­line, but the work that makes them pos­si­ble hap­pens below it.

Kyle Bialek is a Man­ag­ing Direc­tor at KPMG. 

Some or all of the ser­vices described here­in may not be per­mis­si­ble for KPMG audit clients and their affiliates.