The util­i­ties indus­try has long oper­at­ed on a sim­ple man­date: gen­er­ate pow­er, move it through the grid, and bill the cus­tomer. But SAP is bet­ting that AI will push util­i­ties beyond this trans­ac­tion­al iden­ti­ty toward some­thing more ambi­tious. Bruno Pin­col­i­ni, SAP’s Glob­al Lead of Busi­ness AI for Util­i­ties, describes the shift as trans­form­ing util­i­ties ​“from providers of essen­tial ser­vices to plat­forms for inno­va­tion and resilience.”

Joule agents are cen­tral to that trans­for­ma­tion. SAP is cur­rent­ly posi­tion­ing Joule not as a stand­alone chat­bot but as a ​“process-aware copi­lot,” per Pin­col­i­ni, that spans the full util­i­ties val­ue chain, from smart meter­ing and mar­ket oper­a­tions through bill-to-cash work­flows, cus­tomer expe­ri­ence, and ser­vice and asset management.

SAP is now putting spe­cif­ic prod­uct com­mit­ments behind that vision. As part of its 2026 roadmap, the com­pa­ny plans to roll out AI-dri­ven capa­bil­i­ties tar­get­ing cus­tomer self-ser­vice, con­sump­tion analy­sis, and billing excep­tion han­dling. In an ASUG inter­view, Pin­col­i­ni pre­viewed the new and forth­com­ing fea­tures and gov­er­nance archi­tec­ture geared toward mak­ing AI viable in a sec­tor defined by com­pli­ance constraints. 

SAP’s approach is as delib­er­ate in what it pri­or­i­tizes as what it avoids. ​“The aim is not to replace proven trans­ac­tion­al con­trols in SAP for Util­i­ties,” Pin­col­i­ni explains, ​“but to speed insight, deci­sion-mak­ing, and guid­ed exe­cu­tion on top of them, with­out com­pro­mis­ing governance.”

Process-Native Ver­sus Gener­ic AI

Gen­er­al-pur­pose AI copi­lots strug­gle in enter­prise con­texts because they lack access to the data struc­tures and busi­ness log­ic that define how a com­pa­ny actu­al­ly oper­ates. When a cus­tomer calls about an unex­plained spike in their bill, a gener­ic chat­bot can deliv­er vague answers or search pub­lic knowl­edge bases, but it can­not pull the customer’s actu­al meter reads, com­pare them to their rate sched­ule, or check whether a recent ser­vice order might explain the discrepancy.

Pin­col­i­ni not­ed that gen­er­al-pur­pose AI copi­lots ​“risk hal­lu­ci­nat­ing or giv­ing impre­cise answers because they lack real data, con­text, or inte­gra­tion.” SAP’s answer is what it calls ​“ground­ing.” Joule con­nects direct­ly to the SAP Knowl­edge Graph and SAP Busi­ness Data Cloud, anchor­ing its respons­es in a customer’s actu­al enter­prise data rather than sta­tis­ti­cal infer­ence about what a plau­si­ble answer might be.

The depth of that inte­gra­tion appears in Joule’s indus­try-spe­cif­ic vocab­u­lary. The agent is pro­grammed to under­stand com­mer­cial struc­tures like busi­ness part­ners and con­tract accounts along­side tech­ni­cal assets like devices, reg­is­ters, and meter reads. Joule can also nav­i­gate process sta­tus­es such as dun­ning lev­els and ser­vice orders. When a billing inquiry arrives, the sys­tem pulls invoic­es, pay­ment his­to­ry, and ser­vice records to con­struct a response ground­ed in fact rather than inference.

Joule can also coor­di­nate com­plex tasks that span mul­ti­ple SAP appli­ca­tions — S/4HANA Util­i­ties, Ser­vice Cloud, Field Ser­vice Man­age­ment, and Enter­prise Asset Man­age­ment — draw­ing on approved APIs and SAP Inte­gra­tion Suite to exe­cute work­flows end-to-end. Pin­col­i­ni cites this orches­tra­tion capa­bil­i­ty as the def­i­n­i­tion of ​“process-native.” Beyond the vocab­u­lary of util­i­ties oper­a­tions, the AI under­stands the actu­al mechan­ics of how work gets done in SAP systems.

Inside the Con­tact Center

The most con­crete imple­men­ta­tion of Joule today lives in SAP Ser­vice Cloud 2.0, where sev­er­al AI fea­tures tar­get the real­i­ties of util­i­ty con­tact cen­ter oper­a­tions. A Case Clas­si­fi­ca­tion Agent ana­lyzes incom­ing ser­vice tick­ets and cat­e­go­rizes them accord­ing to com­pa­ny-spe­cif­ic busi­ness require­ments. A Case Sum­ma­ry func­tion gen­er­ates con­densed his­to­ries from the full thread of cus­tomer com­mu­ni­ca­tions, spar­ing agents from scrolling through dozens of exchanges to under­stand the situation.

For email-orig­i­nat­ed cas­es, a Case Type Deter­mi­na­tion mod­el assess­es the sub­ject line and descrip­tion to estab­lish the appro­pri­ate case type, then inputs that data to auto­mat­i­cal­ly ini­ti­ate case creation.

The Sim­i­lar Case Rec­om­men­da­tion fea­ture scans the pre­vi­ous twelve months of cat­e­go­rized cas­es, using seman­tic sen­tence-lev­el analy­sis rather than sim­ple key­word match­ing to sur­face the three most rel­e­vant prece­dents. Seman­tic analy­sis cap­tures mean­ing and con­text that key­word match­ing miss­es, which improves rel­e­vance when agents need to see how the orga­ni­za­tion han­dled com­pa­ra­ble situations.

New and Upcom­ing Features

SAP’s cur­rent roadmap for Joule in util­i­ties extends the assistant’s reach into cus­tomer self-ser­vice and con­sump­tion analy­sis, while final­ly tar­get­ing one of billing’s per­sis­tent pain points: excep­tion handling.

Since Q4 2025, a Util­i­ties Cus­tomer Self-Ser­vice Agent has been equipped to han­dle cus­tomer inter­ac­tions with vis­i­bil­i­ty into the customer’s full pro­file: rate struc­tures and active prod­ucts, his­tor­i­cal usage, and poten­tial upgrade paths. SAP’s stat­ed goal is per­for­mance ​“equal to han­dling by human agents,” which, if achieved, can cut oper­at­ing costs while poten­tial­ly improv­ing cus­tomer loy­al­ty and iden­ti­fy­ing rev­enue oppor­tu­ni­ties that pure­ly trans­ac­tion­al self-ser­vice misses.

Also live since Q4 2025, an AI Sum­ma­ry for Billed Con­sump­tion capa­bil­i­ty com­piles a rolling year of billing his­to­ry for each metered ser­vice, sur­fac­ing usage trends and flag­ging what drove any fluc­tu­a­tions from one cycle to the next. This gives util­i­ty admin­is­tra­tors the con­text they need for informed deci­sion-mak­ing and proac­tive ser­vice management.

In the sec­ond quar­ter of 2026, SAP plans to intro­duce a Joule Agent for resolv­ing out­sort­ed billing doc­u­ments. These excep­tions are a man­u­al bot­tle­neck in an oth­er­wise auto­mat­ed billing sys­tem and require human judg­ment. The agent will review the full con­text behind each excep­tion — the Busi­ness Process Excep­tion Man­age­ment (BPEM) case file, the customer’s mas­ter data and con­sump­tion his­to­ry, and the billing doc­u­ment itself — then pro­pose a res­o­lu­tion with sup­port­ing ratio­nale. The aim is to sup­port billing spe­cial­ists by pro­vid­ing analy­sis and rec­om­men­da­tions, while leav­ing final deci­sions to humans.

The Gov­er­nance Question

Util­i­ties oper­ate under reg­u­la­to­ry con­straints that make gov­er­nance non-nego­tiable. Billing reg­u­la­tions, shut­off pro­tec­tions, and com­plaint-han­dling require­ments cre­ate a com­pli­ance land­scape where autonomous AI rec­om­men­da­tions could eas­i­ly cause problems.

SAP address­es this through what Pin­col­i­ni describes as mul­ti­ple lay­ers of guardrails:

• Tech­ni­cal secu­ri­ty con­trols sit along­side role-based access that ensures agents only see and do what they’re autho­rized to see and do.
• Agent orches­tra­tion runs through a con­trolled hub.
• Audit log­ging enables accountability.
• Reg­u­la­to­ry-change man­age­ment capa­bil­i­ties help orga­ni­za­tions adapt as rules evolve.

The archi­tec­ture is designed for reg­u­lat­ed, risk-sen­si­tive domains where an AI’s rec­om­men­da­tions must remain with­in the bounds of what is legal­ly and pro­ce­du­ral­ly permissible.

Cal­cu­lat­ing The Cost of AI

SAP’s AI com­mer­cial mod­el is divid­ed into two tiers. Many embed­ded AI capa­bil­i­ties come includ­ed with cloud appli­ca­tion sub­scrip­tions and inher­it the data res­i­den­cy, role-based access, and audit con­trols of the host appli­ca­tion. Basic AI fea­tures and a starter allo­ca­tion of Joule mes­sages typ­i­cal­ly require no addi­tion­al licensing.

Pre­mi­um capa­bil­i­ties fol­low a con­sump­tion-based mod­el metered through SAP Busi­ness Tech­nol­o­gy Plat­form (BTP). This tier includes plat­form ser­vices like the Gen­er­a­tive AI Hub and AI Core, spe­cial­ized tools like Doc­u­ment Infor­ma­tion Extrac­tion and Joule Stu­dio for build­ing cus­tom agents, and usage-based charges for high-vol­ume Joule inter­ac­tions and agent orches­tra­tion. All require AI Units, billed against SAP BTP enti­tle­ments under CPEA or pay-as-you-go terms.

Joule oper­ates across both mod­els. The in-appli­ca­tion expe­ri­ence is includ­ed with eli­gi­ble SAP cloud sub­scrip­tions. How­ev­er, when Joule uses pre­mi­um SAP BTP AI ser­vices or third-par­ty large lan­guage mod­els via the Gen­er­a­tive AI Hub, that usage is billed against SAP BTP consumption.

On the pri­va­cy front, SAP com­mits to keep­ing cus­tomer data out of foun­da­tion mod­el train­ing unless the SAP user explic­it­ly autho­rizes it. For an indus­try where reg­u­la­to­ry trust is foun­da­tion­al, that com­mit­ment to data bound­aries is rein­forc­ing the guardrails approach Pin­col­i­ni empha­sizes over­all for the sector.

Stay ahead with First Five, ASUG­’s exclu­sive newslet­ter. Sub­scribe today for a curat­ed email fea­tur­ing top insights tai­lored for SAP users.