What the largest US bank did well, what it has not yet shown, and what a board can take from both.
Terence KokExecutive Director, AI Governance & Assurance, Orion Five Engineering
About your speaker
Terence Kok
Executive Director, AI Governance & Assurance Practice, Orion Five Engineering
Former Chief AI and Innovation Officer, Meinhardt Group
Twenty-five years leading digital and AI transformation across Asia and the Middle East: national smart-city programmes, critical infrastructure and enterprise operations. Today I help leadership teams put AI into production with controls they can show a board, an auditor and a regulator.
BuiltA governed AI platform serving more than 6,000 engineers, with zero security incidents
CertifiedISO/IEC 42001 Lead Auditor Professional (BSI) · AIGP (SMU) · Registered Management Consultant
ServingCTCB CPSC Advisory Board · UN ESCAP AI for Developing Countries Forum
AwardsCDO Vision Award 2026 · Singapore's Most Influential Data & AI Leaders 2026 · Google Trailblazer Transformation Award 2024
Independent analysis from public sources. The views in this session are mine.
Case study · JPMorgan Chase
This session
Seven parts, from the foundations to the board agenda.
The foundations2016 to 2023: COiN, cloud, data and senior owners.
What the bank builtLLM Suite, the tools on top and the technology budget.
What they did wellSeven strengths and the results the bank has published.
Where the record falls shortFive gaps a board should notice.
The value claimHow the $2 billion figure moved and how the bank counts it.
Jobs, controls and DBSWorkforce statements, the penalties and two ways of counting.
Part seven is for the board: seven lessons, a controls checklist and questions for discussion.
Case study · JPMorgan Chase
The case in one slide
The strongest AI record in banking, with one number still unexplained.
#1
on the Evident AI Index in every edition to October 2026
200,000+
colleagues given LLM Suite, the bank’s own generative AI platform
$19.8bn
technology spend expected in 2026, up 10%
~$2bn
of benefit for ~$2bn of AI expense, per the CEO; no published method
Sources: [26] Evident, 6 Oct 2026; [2] Annual Report 2024; [7] Company Update, 23 Feb 2026; [19] Bloomberg, 7 Oct 2025.
Case study · JPMorgan Chase
Why boards study this case
Three reasons it belongs on a board agenda.
01
It is the benchmark
When a board asks what good AI looks like in a bank, this is usually the bank it means.
02
Both halves are visible
A well-run build, and the gaps that remain when value has to be shown outside the bank.
03
Much of it can be copied
Platform first, senior owners and named results do not need a $20 billion budget.
Sources: [26] [7] [2]. The reasons are my reading.
Case study · JPMorgan Chase
01
The foundations came first.
2016 to 2023: contract review, cloud, data and senior owners, before generative AI.
Case study · JPMorgan Chase
2016 · Contract Intelligence
AI at the bank predates ChatGPT by six years.
COiN went live in June 2016. Bloomberg reported in 2017 that it did in seconds the review of about 12,000 commercial loan agreements a year that had taken 360,000 hours of work.
The 360,000 hours is the bank’s own figure, relayed by Bloomberg. No headcount effect was reported.
Sources: [29] ABA Journal, Mar 2017; [30] Bloomberg via Data Center Knowledge, 6 Mar 2017.
Case study · JPMorgan Chase
Modern infrastructure · Investor Day, May 2025
Most applications had moved to modern infrastructure by 2025.
65%
of applications ran a large part of their workloads on public or private cloudup from 50% a year earlier
80%
including applications on virtual servers
~$18bn
technology expense outlook for 2025about $1bn more than 2024
Peak
"probably past the point of peak modernization spend"CFO, Investor Day 2025
Generative AI arrived on top of a platform that was already governed. That is the sequence most organisations try to skip.
Source: [4] Investor Day 2025 transcript. The last line is my reading.
Case study · JPMorgan Chase
2023 · owners and controls
The bank gave AI senior owners and early limits.
Jun 2023
A data and analytics chief
Teresa Heitsenrether became the first Chief Data & Analytics Officer, on the Operating Committee.
2,000+
Specialists in house
"More than 2,000 AI/machine learning (ML) experts and data scientists", per the 2023 letter.
Feb 2023
ChatGPT restricted
Under normal third-party software rules, "not because of a particular issue", before the bank had its own platform.
Sources: [17] FinTech Futures, Jun 2023; [1] Annual Report 2023 letter; [31] CNN, 22 Feb 2023.
Case study · JPMorgan Chase
02
What the bank built.
One governed platform for most staff, tools on top, and agents next.
Case study · JPMorgan Chase
LLM Suite · how it fits together
One controlled platform sits between staff and the models.
Sources: [2] Annual Report 2024; [22] CNBC, 30 Sep 2025; [3] Annual Report 2025; [23] CNBC, 9 Jun 2026. The dashed box is my reading.
Case study · JPMorgan Chase
LLM Suite reach · five reports, four measures
From one division to most of the workforce in about eighteen months.
Jul 2024Access, Asset & Wealth Mgmt first
50,000
Sep 2024Users, as reported
140,000
2024Colleagues it was rolled out to
200,000+
Sep 2025Access; about half use it near daily
~250,000
Feb 2026Weekly users, per the CEO
150,000
Sources: [36] Fortune reporting the FT, Jul 2024; [18] CIO Dive, Sep 2024; [2] Annual Report 2024; [22] CNBC, Sep 2025; [7] Company Update, Feb 2026. Access, roll-out and weekly use are different measures, not one trend.
Case study · JPMorgan Chase
On top of the platform
Tools built for specific jobs.
Coach AI
Helps wealth advisers find information "up to 95% faster", per the bank to Reuters.
GPT-Insights
Investment research tool launched December 2024; research task time cut "by up to 85%".
Code assistants
Used by over 90% of engineers in the corporate and investment bank.
Employee Assistant
A personal AI agent that answers staff questions and takes action for them.
Use cases doubled
AI use cases in production doubled in the year to February 2026.
Agents next
Agents that run on their own "for hours", announced June 2026.
Sources: [38] Reuters, May 2025; [2] Annual Report 2024; [3] Annual Report 2025; [7] Company Update, Feb 2026; [23] CNBC, Jun 2026. All company-reported.
Case study · JPMorgan Chase
Technology expense, US$ billion
AI spend of about $2 billion sits inside a budget near $20 billion.
9.6
2017As reported
15.5
2023
~17
2024Outlook
~18
2025~$2bn of it AI
~19.8
2026Outlook, +10%
Sources: [30] Bloomberg, 2017; [6] Investor Day 2024 presentation; [4] Investor Day 2025; [7] Company Update, Feb 2026; [19] Bloomberg, Oct 2025 (AI expense).
Case study · JPMorgan Chase
03
What they did well.
Seven strengths the public record supports, and the results the bank has published.
Case study · JPMorgan Chase
What the record supports
Strengths a board can copy.
Strength
What the record shows
Platform before use cases
Cloud and data work came first, so LLM Suite had somewhere governed to run. [4] [2]
One controlled platform
A single model-agnostic tool reached 200,000+ colleagues on an opt-in roll-out. [2] [21]
Senior ownership
Data and analytics chief and CIO both on the Operating Committee. [1] [7]
Talent at scale
2,000+ AI and ML specialists; prompt training in onboarding. [1] [40]
Named results
KYC, engineering, screening and fraud results quantified by function. [2] [3] [4]
Plain speaking
Leaders say publicly what they cannot measure. [4] [8]
Early controls
ChatGPT restricted in 2023; AI under model risk; supplier security letter. [31] [9] [14]
Case study · JPMorgan Chase
Results the bank has published, by function
Specific, named and unaudited.
~40%
lower KYC unit costsKnow-your-customer, Annual Report 2024
10–20%
engineer productivity from coding assistantsAnnual Report 2024
2× / ½
double the screening volume, half the manual checksAnnual Report 2025
Flat
fraud cost in basis points while attacks grew 12% a yearInvestor Day 2025
85%
less time on research tasks, at mostGPT-Insights, Annual Report 2024
~90%
less manual cash-flow work for some clientsBloomberg, Mar 2024, bank-sourced
Sources: [2] [3] [4] [35]. Company-reported, unaudited, no published method. More specific than most banks disclose.
Case study · JPMorgan Chase
Fraud · Investor Day, May 2025
The fraud result names its measure.
“With a 12% CAGR in attack rate, the cost of fraud in basis points has been flat ... AI has been a very significant part of the solution.”
Marianne Lake, CEO of Consumer & Community Banking, Investor Day 2025.
Why it is strong
A measure that existed before AI, tracked over time, against a stated pressure.
What it is not
Audited, or split between AI and other controls.
Source: [4] Investor Day 2025 transcript. The reading is mine.
Case study · JPMorgan Chase
Plain speaking about measurement
The bank’s leaders say what they cannot attribute.
Marianne Lake · May 2025
"While AI has definitely contributed here, a lot of this is good old-fashioned process automation and organizational efficiency."
On consumer-bank operations productivity. [4]
Daniel Pinto · May 2024
"Difficult to measure, but roughly the value that we assign to our artificial intelligence use cases ..."
Introducing the $1 billion to $1.5 billion figure. [5]
Jeremy Barnum · Oct 2025
Proving AI savings "turns out to be a very hard thing to do, hard to prove".
CFO, third-quarter call. [8]
Jeremy Barnum · Oct 2025
It "might ... result in people scrambling around to use AI in ways that are actually not efficient".
Why the bank does not make teams prove savings. [8]
Case study · JPMorgan Chase
Controls and suppliers
Controls came early, and the bank pushed its suppliers.
The 10-K brings AI and machine learning models under model risk management. ChatGPT was restricted in 2023, before the bank’s own platform existed.
Software-as-a-service "is quietly enabling cyber attackers", and AI agents amplify the risk.Patrick Opet, CISO, open letter to suppliers, 26 Apr 2025
Sources: [9] Form 10-K for 2025; [31] CNN, Feb 2023; [14] JPMorgan Chase, Apr 2025.
Case study · JPMorgan Chase
Evident AI Index · 6 October 2026
First in every edition, on what Evident measures.
#1
overall, fifth edition in a row
#1
Leadership and Transparency pillars, 2026
#2
Talent and Innovation pillars, 2026
12 of 50
banks disclose any realised or projected AI returns
The index scores capability and disclosure. It does not measure the value AI has delivered.
Sources: [26] Evident AI Index, 6 Oct 2026; [27] FF News, 6 Oct 2026.
Case study · JPMorgan Chase
04
Where the record falls short.
Five gaps that a board, an investor or a supervisor would notice.
Case study · JPMorgan Chase
What the public record does not show
Gaps in an otherwise strong record.
A headline number without a methodThe value figure moved four times and lives in interviews and conferences, not the reports. [28]
Data-control failures in a regulated functionTrade-surveillance gaps penalised in 2024; recordkeeping in 2022. Not AI failures. [10] [13]
Workforce messages that are easy to misread10% in operations, "not fewer headcount", 30% to 40% in discrete areas. [4] [3] [24]
Pressure on staff to adoptCoding-assistant usage reportedly feeds engineers’ goals. [39]
Concentration and rising running costsTwo model providers, no disclosed exit plan; usage costs accelerating. [22] [24]
Case study · JPMorgan Chase
05
The value claim.
How the $2 billion figure moved, and how the bank says it counts.
Case study · JPMorgan Chase
The headline value statement, 2023 to 2026
One figure, five occasions, no published method.
When
Who
Figure
Words and venue
Basis given
2023
Bank target
$1bn → $1.5bn
"Business value" target, raised in the year; reported by Bloomberg [16]
None
20 May 2024
Daniel Pinto
$1–1.5bn
"Difficult to measure, but roughly the value that we assign"; Investor Day [5]
Fields listed
Sep 2024
Daniel Pinto
~$2bn
Raised toward $2bn; "fraud is a big beneficiary"; conference [18]
None
7 Oct 2025
Jamie Dimon
$2bn for $2bn
"For $2 billion of expense, we have about $2 billion of benefit"; Bloomberg TV [19]
Partly measurable
23 Feb 2026
Jamie Dimon
NPVs
Projects tracked as NPVs; generative AI time savings excluded; Company Update [7]
Closest to a method
Fortune counts 13 earnings calls since mid-2023 without the figure [28]; a check of three transcripts agrees [8] [45] [7].
Case study · JPMorgan Chase
Reading the figure
What the $2 billion is, and what it is not.
How it is often reported
"JPMorgan is saving $2 billion a year with AI"
Read as audited annual savings, and sometimes as the size of the bank’s AI investment.
What was said
About $2bn of benefit for about $2bn of expense
Benefit mixes revenue, cost avoidance and risk and fraud. On its own framing, benefit roughly matches spend. The spend sits inside an $18bn technology budget.
Sources: [42] Entrepreneur, Oct 2025 (example); [19] [20] Bloomberg and Benzinga, Oct 2025; [7] [4].
Case study · JPMorgan Chase
Company Update · 23 February 2026 · answering an analyst
The closest thing to a method.
"In AI, there are NPVs, some are revenue enhancements, some are cost avoidance, some are risk and fraud ... we can measure it, but we don't give it credit ... 150,000 people use it every week. They think they're saving four hours a [week]. That's not in an NPV."Jamie Dimon, Chairman and CEO. The bank’s transcript corrects "day" to "[week]".
A defensible rule: count project returns, refuse self-reported time saved. What is missing is the breakdown that would let an outsider check the total.
Source: [7] Company Update transcript. The reading is mine.
Case study · JPMorgan Chase
What the annual reports say instead
The formal reports give growth rates for one business, with no base.
2024
Consumer & Community Banking
+35%
"increase in value" from AI/ML
Investor Day 2025. No dollar base given. [4]
2025
Consumer & Community Banking
~+60%
"increase in value from AI and machine learning"
Annual Report 2025. No dollar base given. [3]
Sources: [4] Investor Day 2025 transcript; [3] Annual Report 2025. No firm-wide dollar figure appears in either annual report.
Case study · JPMorgan Chase
06
Jobs, controls and DBS.
What the bank has said about jobs, the penalties, and a second way of counting.
Case study · JPMorgan Chase
What the bank has said about jobs
Each statement on jobs has its own scope.
Marianne Lake · May 2025
Operations headcount to "trend down by about 10% over the next five years or so"
Scope: CCB operations excluding Home Lending, while the business grows over 25%. [4]
Jamie Dimon · Oct 2025
"There’ll be fewer jobs in certain functions"
Scope: with retraining and redeployment. [19] [43]
Jennifer Piepszak · Apr 2026
"The goal is not fewer headcount"
Scope: fewer jobs "will likely be the outcome in certain jobs". [3]
Jamie Dimon · Jul 2026
"Discrete areas where we did reduce jobs by 30% or 40%"
Scope: "most of those people offered jobs elsewhere". [24]
320,000+ employees at the 2025 annual report [3]. No firm-wide figure for AI-related reductions has been published.
Case study · JPMorgan Chase
Regulatory penalties, 2022 to 2024 · none concerned an AI system
The controls record is about data.
Sep 2022SEC and CFTC
$200m
Off-channel messages not preserved
Mar 2024OCC
$250m
Gaps in trade-surveillance data
Mar 2024Federal Reserve
$98m
Same gaps, back to 2014
May 2024CFTC
$200m
Order data missed; $100m offset
Sources: [13] SEC, 27 Sep 2022; [10] OCC, 14 Mar 2024; [11] Federal Reserve, 14 Mar 2024; [12] CFTC, May 2024.
Case study · JPMorgan Chase
Why the penalties matter for AI
Any AI built on incomplete data inherits the gap.
Data
Completeness
Billions of trading events on dozens of venues went unsurveilled for years.
Risk
Concentration
LLM Suite runs models from two providers; no exit plan is disclosed.
Cost
Usage
The CFO expects model usage costs to accelerate in the second half of 2026.
Sources: [10] [11] [22] [24] [25]. The framing is mine.
Case study · JPMorgan Chase
Compared with DBS · measurement only
DBS and JPMorgan count AI value differently.
DBS
JPMorgan Chase
Headline
About S$1bn of economic value, 2025 [46]
About $2bn of benefit for $2bn of expense [19]
Where published
Annual report narrative
Interviews and conferences
Method
Mainly test and control groups [47]
Project NPVs; GenAI time savings excluded [7]
Breakdown or audit
Neither
Neither
Evident 2026
25th
1st [26]
Different currencies and bases: never rank one figure against the other.
Case study · JPMorgan Chase
Reporting
What is often misreported.
Often said
What the record shows
Saving $2bn a year
About $2bn of benefit for about $2bn of expense. [19] [42]
A $2bn AI investment budget
$2bn is AI expense; the technology budget is about $18bn. [4]
At least 10% of staff
About 10% over five years or so, CCB operations ex Home Lending. [4]
AI cut 30–40% of jobs
"Discrete areas", most people offered other roles. [24]
Fined over AI
The 2024 penalties were for trade-surveillance data. [10]
#1 means most value
Evident scores capability and transparency. [26]
COiN replaced lawyers
360,000 hours of review work; no headcount effect reported. [30]
Case study · JPMorgan Chase
07
What boards can take from it.
Seven lessons, a controls checklist and questions to take back.
Case study · JPMorgan Chase
Lessons 1 and 2 · what to copy
LESSON 01
Build the platform before the use cases.
Scale came from cloud, data and one governed platform first.
Board actionAsk what your AI runs on before you ask how many use cases are live.
LESSON 02
Give AI an owner at the top table.
A data and analytics chief on the executive committee made AI a line responsibility.
Board actionName one executive accountable for AI value and AI risk.
Lessons are my reading of the record: [1] [2] [4].
Case study · JPMorgan Chase
Lessons 3 and 4 · reporting value
LESSON 03
Publish named results with their measure.
KYC unit cost and fraud cost against attack rate can be understood and challenged.
Board actionPrefer function-level measures over time to a single total.
LESSON 04
If you give a total, give the method.
A figure given in interviews moves; a figure with a method can be checked.
Board actionState benefit, period and baseline, and report it with the accounts.
Lessons are my reading of the record: [2] [3] [4] [7] [28].
Case study · JPMorgan Chase
Lessons 5 to 7 · controls, people and cost
LESSON 05
Treat data controls as AI controls.
The penalties came from incomplete data.
Board actionAsk for data completeness evidence on any AI in a regulated process.
LESSON 06
Give every workforce statement its scope.
Unscoped numbers became headlines.
Board actionFunction, period and what happens to the people, every time.
LESSON 07
Watch running cost and suppliers.
Usage costs rise with adoption; two providers is a concentration.
Board actionPut both in the board pack each quarter.
Lessons are my reading of the record: [10] [3] [24] [22].
Case study · JPMorgan Chase
For government and regulators
The strongest results came where a measure already existed.
Ask for measures
Fraud cost, unit cost and screening volumes are evidence a supervisor can test.
Ask for the method
Any AI total that appears in investor materials should come with how it was counted.
Treat data as the control
Surveillance and recordkeeping gaps are where AI risk starts in regulated work.
My recommendations, drawn from [2] [3] [4] [10] [28].
Third-quarter results: any AI value figure, usage cost or headcount comment.
Q4 2026
Model usage costs, which the CFO said would accelerate.
2026 to 2027
What the bank discloses about controls on agents that run for hours.
Early 2027
Company Update and Investor Day: a method or breakdown for the value figure.
Apr 2027
2026 annual report: consumer-bank AI value, operations headcount.
Case study · JPMorgan Chase
Evidence and methodology
How strong each claim in this case is.
Claim
Rests on
Grade
Regulatory penalties of 2022 and 2024 for recordkeeping and trade-surveillance failures
OCC, Federal Reserve, CFTC and SEC releases
Measured, regulator findings
Technology expense of about $15.5bn, $17bn, $18bn and $19.8bn, 2023 to 2026
Investor Day 2024 and 2025, Company Update 2026
Company outlooks, primary transcripts
LLM Suite rolled out to more than 200,000 colleagues
Annual Report 2024
Company-reported
About 250,000 with access, half using it near daily
CNBC, September 2025
Press report, bank-sourced
KYC unit costs down nearly 40%; engineers up to 10-20% more productive; screening at double the volume with half the checks
Annual Reports 2024 and 2025
Company-reported, unaudited, no method
Fraud cost flat in basis points while attacks grew 12% a year
Investor Day 2025 transcript
Company-reported, unaudited
AI/ML value in consumer banking up 35% (2024) and nearly 60% (2025)
Investor Day 2025; Annual Report 2025
Company claim, growth rate only, no base
“About $2 billion of benefit” for $2 billion of expense
Dimon, Bloomberg TV and conference remarks, 2025
Company claim, no published method
AI value tracked as project NPVs, generative AI time savings excluded
Company Update transcript, February 2026
Company description of method, no breakdown
The $2bn figure has not appeared on earnings calls
Fortune, October 2026; spot check of three transcripts
Press analysis, partly checked
First on the Evident AI Index in every edition
Evident, October 2026
Third-party rating of capability, not value
Coding-assistant usage feeds engineers’ goals
Business Insider, syndicated copy
Single press report, original not read
The lessons for boards and the controls checklist
My reading of the record cited against each
My call
Case study · JPMorgan Chase
Full timeline · 1 of 3
June 2016 to May 2024.
Jun 2016COiN (Contract Intelligence) goes live. Bloomberg reports in 2017 that it does in seconds a review of about 12,000 commercial loan agreements a year that took 360,000 hours of work.[29] [30]
22 Feb 2023The bank restricts staff use of ChatGPT under its normal controls on third-party software, “not because of a particular issue”.[31]
May 2023A trademark application for “IndexGPT”, a GPT-based tool for selecting securities, is reported.[32]
Jun 2023Teresa Heitsenrether is named the first Chief Data & Analytics Officer. The 2023 letter says the role sits on the Operating Committee.[17] [1]
Nov 2023Heitsenrether says the 2023 target for AI “business value” has been raised from $1bn to $1.5bn, and that AI is starting to generate revenue.[16]
Mar 2024A cash-flow tool is reported to cut some corporate clients’ manual work by almost 90%.[35]
14 Mar 2024The OCC (US$250m) and the Federal Reserve (about US$98.2m) penalise the bank for gaps in trade-surveillance coverage and data controls. Not an AI failure.[10] [11]
Apr 2024The 2023 shareholder letter: “more than 2,000 AI/machine learning (ML) experts and data scientists”, over 400 use cases in production, and AI “may reduce certain job categories or roles”.[1]
20 May 2024Investor Day. Daniel Pinto puts the value the bank assigns to its AI use cases at $1bn to $1.5bn, “difficult to measure”. Technology expense outlook about $17bn.[5] [6]
Case study · JPMorgan Chase
Full timeline · 2 of 3
May 2024 to October 2025.
May 2024The CFTC adds a US$200m penalty for the same surveillance failures, US$100m of it offset by the earlier payments.[12]
Jul 2024LLM Suite, the bank’s in-house generative AI platform, reaches staff in Asset & Wealth Management; about 50,000 have access.[36] [37]
Sep 2024Pinto raises the value figure toward about $2bn and says fraud is a big beneficiary.[18]
Apr 2025The 2024 annual report: LLM Suite rolled out to more than 200,000 colleagues; up to 10-20% productivity for engineers; KYC unit costs down nearly 40%; GPT-Insights cuts research task time by up to 85%.[2]
26 Apr 2025The bank’s CISO publishes an open letter to suppliers warning that software-as-a-service and AI agents amplify cyber risk.[14]
19 May 2025Investor Day. Consumer banking reports AI/ML value up 35% in 2024; operations headcount to trend down about 10% over five years or so. Technology expense outlook about $18bn.[4] [15]
Jun 2025LLM Suite wins American Banker’s 2025 Innovation of the Year Grand Prize.[21]
Sep 2025CNBC: about 250,000 staff have LLM Suite access and about half use it near daily; models come from OpenAI and Anthropic and are refreshed every eight weeks.[22]
7 Oct 2025Dimon: “for $2 billion of expense, we have about $2 billion of benefit”, and fewer jobs in certain functions.[19] [20]
Case study · JPMorgan Chase
Full timeline · 3 of 3
October 2025 to October 2026.
14 Oct 2025CFO Jeremy Barnum says proving AI savings is “a very hard thing to do, hard to prove”, and that the bank relies on expense discipline instead.[8]
23 Feb 2026Company Update. Technology spend about $19.8bn; AI use cases in production doubled in a year; Dimon describes AI value as project NPVs, with generative AI time savings left out.[7]
6 Apr 2026The 2025 annual report: consumer banking AI/ML value up nearly 60%; screening at double the volume with half the manual checks; over 90% of engineers use code assistants; “the goal is not fewer headcount”.[3] [44]
Mar to Apr 2026Business Insider reports that coding-assistant usage feeds engineers’ performance goals.[39]
Jun 2026The bank says it will deploy AI agents that run for hours on their own.[23]
14 Jul 2026Second-quarter call. Dimon: “discrete areas” saw jobs cut 30% or 40%, most staff offered other roles. Barnum: model usage costs to accelerate in the second half; use the right model for the task.[24] [25]
6 Oct 2026Evident AI Index: JPMorgan first for the fifth straight edition, and first on Leadership and Transparency.[26] [27]
9 Oct 2026Fortune notes the $2bn figure has not appeared on 13 earnings calls since mid-2023.[28]
Case study · JPMorgan Chase
For discussion
Six questions to take back to your board.
What does our AI run on, and who controls it?
Who on our executive committee owns AI value and AI risk?
Which AI results can we name, with a measure that existed before AI?
If we quote a total, could an outsider check how we counted it?
Is the data under our AI complete enough for a regulator?
What would it cost to change model provider next year?
Case study · JPMorgan Chase
Take it further
The full case and two tools.
The JPMorgan Chase case study
Full text, motion figures, 47 numbered sources and the FAQ. terencekok.com/research/case-studies/jpmorgan-chase
AI value ledger
Grade an AI value claim by what it rests on before it goes to the board.
Agent risk assessment matrix
Place a planned agent by autonomy and consequence.
9JPMorgan Chase. (2026, 13 February). Annual report on Form 10-K for 2025. US Securities and Exchange Commission.www.sec.gov
10Office of the Comptroller of the Currency. (2024, 14 March). OCC assesses $250 million civil money penalty against JPMorgan Chase Bank, N.A. (NR 2024-25).occ.gov
11Board of Governors of the Federal Reserve System. (2024, 14 March). Federal Reserve Board announces it has fined JPMorgan Chase & Co. about $98.2 million.www.federalreserve.gov
12Commodity Futures Trading Commission. (2024, May). J.P. Morgan Securities LLC to pay $200 million for supervision failures (Release 8914-24).www.cftc.gov
13US Securities and Exchange Commission. (2022, 27 September). SEC charges 16 Wall Street firms with widespread recordkeeping failures (Press release 2022-174).www.sec.gov
14Opet, P. (2025, 26 April). An open letter to third-party suppliers. JPMorgan Chase.www.jpmorganchase.com
15JPMorgan Chase. (2025, 19 May). Investor Day 2025: full presentation.www.jpmorganchase.com
16Doherty, K. (2023, 29 November). JPMorgan says AI tech is starting to generate revenue. Bloomberg, via American Banker.www.americanbanker.com
Case study · JPMorgan Chase
Sources · 2 of 3
Filings, regulators and named press.
17FinTech Futures. (2023, June). JP Morgan Chase appoints new chief data and analytics officer.www.fintechfutures.com
18Wilkinson, L. (2024, 11 September). JPMorgan Chase LLM Suite generative AI employee tool. CIO Dive.www.ciodive.com
19Bloomberg. (2025, 7 October). JPMorgan’s Dimon says AI cost savings now matching money spent.www.bloomberg.com
20Renner, C. B. (2025, 28 October). Jamie Dimon declares JPMorgan Chase’s $2B AI investment “paid for itself”. Benzinga, via Yahoo Finance.finance.yahoo.com
21Arizent. (2025, June). JPMorganChase’s LLM Suite wins American Banker’s 2025 Innovation of the Year Grand Prize.www.arizent.com
22Son, H. (2025, 30 September). Here’s JPMorgan Chase’s blueprint to become the world’s first fully AI-powered megabank. CNBC.www.cnbc.com
23CNBC. (2026, 9 June). JPMorgan Chase plans to deploy more powerful AI agents this year.www.cnbc.com
25Smith, P., & Doherty, K. (2026, 14 July). JPMorgan urges staff to avoid deploying pricey AI for easy tasks. Bloomberg, via Bloomberg Law.news.bloomberglaw.com
26Evident. (2026, 6 October). Evident AI Index for banks, fifth edition.evidentinsights.com
27FF News. (2026, 6 October). JPMorganChase extends lead as world’s most AI-advanced bank.ffnews.com
28Maricic, M. (2026, 9 October). Listen for the AI number on next week’s bank calls. Fortune.fortune.com
29ABA Journal. (2017, March). JPMorgan Chase uses tech to save 360,000 hours of annual work by lawyers and loan officers.www.abajournal.com
30Bloomberg, via Data Center Knowledge. (2017, 6 March). JPMorgan marshals an army of developers to automate finance.www.datacenterknowledge.com
31CNN. (2023, 22 February). JPMorgan restricts employees from using ChatGPT.edition.cnn.com
33American Banker. (2024). The Most Powerful Women in Banking 2024: Teresa Heitsenrether, JPMorgan Chase.americanbanker.com
34FXC Intelligence. (2024). How are banks using AI in 2024?www.fxcintel.com
35Bloomberg. (2024, 4 March). JPMorgan’s AI-aided cashflow model can cut manual work by 90%.www.bloomberg.com
36Fortune. (2024, 26 July). JPMorgan rolls out LLM Suite chatbot to staff, reporting the Financial Times.www.fortune.com
37NBC News. (2024). JPMorgan Chase is giving its employees an AI assistant powered by ChatGPT maker OpenAI.www.nbcnews.com
38Anand, N. (2025, May). JPMorgan says AI helped boost sales, add clients in market turmoil. Reuters, via AOL.www.aol.com
39Business Insider, via AOL. (2026). Big companies are turning AI into a scoreboard (syndicated copy; original not retrieved).www.aol.com
40Wilkinson, L. (2024, 28 May). JPMorgan Chase AI training strategy. Banking Dive.www.bankingdive.com
41FStech. (2025). JPMorgan Chase allows staff to use in-house AI to draft performance reviews.www.fstech.co.uk
42Entrepreneur. (2025, October). JPMorgan Chase is saving $2 billion a year using AI, according to CEO Jamie Dimon (cited as an example of a misreport).www.entrepreneur.com
43Fortune. (2025, 8 October). Jamie Dimon on AI and jobs: no head in the sand.fortune.com
44CNBC. (2026, 6 April). JPMorgan CEO Jamie Dimon in annual letter cites risks in geopolitics, AI and private markets.www.cnbc.com
Connect on LinkedIn