RuncheyResearch

AI Absorption Ledger / C

Citigroup

C · Q2 2026 · reported 2026-07-14 · revenue $24.77bn

Assessment

Citigroup again names AI only on the call, and gives a count: nearly of its people use its AI tools, which the CEO says drive productivity, client experience and faster product launches (claims c1 and c2). The CFO counts more than processes mapped for further automation and names technology and AI automation as a leg of the efficiency push, beside stranded and transformation costs (claim c4). The release, the financial supplement and the 10-Q add no AI statement beyond forward-looking factors (claim c16).

The channels that open are staff productivity from the AI tools, growth from faster product launches, and capital raising for the AI buildout. Only build investment carries a size, the ledger’s own share of revenue, . Staff productivity is directional and unsized: its only measure is the share of staff using the tools, a count that separates no dollars. Faster product launches, process automation and buildout financing credit AI beside other causes and stay unsized, with Services and Wealth revenues, compensation and benefits and Investment Banking revenues as their ceilings.

The cost base moved for reasons the filing names without AI. Total operating expenses rose against revenue growth of ; compensation and benefits rose , with productivity savings and lower transformation expenses as offsets (claim c14); direct staff fell by over the year with severance of in the half. The CFO credits that reduction to past investments and productivity efforts, so the headcount line is a confound on the AI channels, not a measure of them.

Sized channels against the income statement, Q2 2026

1 of 5 channels sized

Each blue mark is one channel's dollars for the quarter; a bar is the range of an estimate. Grey marks are the company's reported lines. The distance between them is the point: how large the AI channel is next to the line it sits in.

New money and old money, Q2 2026

3 expanded2 relabelled

Each channel is tagged once for whether its money existed before language models, from the company's annual report and call at the start of the period. A bar splits one flow's sized dollars by that tag. The incremental total is the part that would not be there without the models: a new channel counts in full, an expanded one only for what AI added, a relabelled one at zero.

Paid for AI$17mn to $235mn sized

expanded $17mn to $235mn

Incremental total $0 to $235mnpoint $0$62mn in 1 channel has no traced baseline

The sized total counts every channel the company credits to AI, including relabelled money that existed before language models and the ledger's own estimates for it. The incremental total counts relabelled channels at zero. A flow is split by layer where its dollars sit at more than one: end use, compute sold to builders, and hardware. The same dollar can be a buyer's spend, a cloud's revenue and a chipmaker's revenue, so the layers are never added together.

Paid for AI1 channel · $17mn to $235mn sized · $0 to $235mn incremental

engineering

Internal investment in AI capability

0.07% to 0.95% of the quarter’s revenue

Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.

our inferencedisclosure: described· motive: exploratory· before LLMs: expanded

The CFO says investment in technology includes AI and the CEO names AI among the cards investments, neither with an amount or parts. Any tool and model cost, including that of the AI tools nearly all staff use, is inside this channel and is not split out, since Citi names no vendor, seat count or price. The size shown is the ledger’s own estimate, , the same share of revenue as in Q1, and the only sized channel this quarter.

Evidence: 3 quotes, 6 figures, 2 confounds, 3 from before coverage

Money Citi spends building AI into the firm: the investment in AI the CEO says falling transformation expenses make room for, the technology investment the CFO says includes AI, and the AI investment the CEO names in the U.S. Consumer Cards plan. It sits in compensation and benefits (technologists) and technology/communication expense, and part may be capitalized software. Citi states no parts of this spending and names no outside model, licence or tool bill, so any tool and model cost, including that of the AI tools staff use, is inside this channel; an unsplit amount stays a single channel and is not divided by judgment shares (the methodology's rule on umbrella metrics with no stated parts).

Why this motive

The CFO says the firm continues to invest in technology including AI and expects productivity saves over time, and the CEO names AI among the cards investments that take time to pay off (claims c3 and c8); no amount, return or measure is given. Investing for later is the exploratory tell, as in Q1.

Before LLMs: expanded

At the anchor Citi already worked with AI and machine learning and had begun using generative AI (claim c-anchor-c1), and explained expense growth by investment in technology and platform modernization (claim c-anchor-c2); technology/communication expense was for FY2024. No AI share of any line was given then or in coverage, so no quarter of the activity before AI can be traced. The size is the whole of an activity that existed before.

“Citi has been working with AI and machine learning for a period of time and has more recently begun using Generative AI, a type of artificial intelligence that uses generative models to create text and other content. Generative AI tools are available to employees within parts of the Company, and in the future Citi may more broadly use, develop and incorporate Generative AI within its technology platform and services, systems and its businesses and functions.”
Filing, risk factors, 10-K periodic report, 2025-02-21
“Services expenses of $10.6 billion increased 6%, primarily driven by continued investments in technology and platform modernization, other risk and controls and product innovation”
Filing, mdna, 10-K periodic report, 2025-02-21
“We're currently deep into a very large body of work, upgrading our data architecture, automating manual controls and processes, consolidating fragmented tech platforms”
CEO, qa, earnings call, 2024-04-12

Figures

  • Technology/communication expense · 2026-CQ2
  • Technology/communication expense, prior year · 2025-CQ2
  • Technology/communication expense against the prior year (negative: a fall) · 2026-CQ2
  • Total revenues, net of interest expense · 2026-CQ2
  • Total revenues growth, year over year · 2026-CQ2
  • Share of Citi’s people using its AI tools, per the CEO (nearly) · as-of 2026-07-14

What else could explain it

  • line composition: The CFO describes technology investment as including AI (claim c3), so technology/communication expense of , against the prior year, is a ceiling on the AI part and never its level.
  • transformation program: The CEO frames the AI integration as applying what the transformation taught as that work winds down (claim c1); transformation spending is falling and is not counted here.

Quotes

“The benefits of our past investments and productivity efforts have allowed us to gain efficiencies across our expense base and reduce our headcount to 219,000, with over $800 million of severance incurred year-to-date. We continue to invest in areas such as technology, including AI, and we would expect an increase in productivity saves over time.”
c3 · CFO, prepared remarks, earnings call, 2026-07-14
“As much of the transformation work winds down, we are not only taking down expenses, but we're applying what we learned about large-scale implementation to integrate AI into our businesses and functions wherever it makes sense. Nearly nine out of 10 of our people are using our AI tools.”
c1 · CEO, prepared remarks, earnings call, 2026-07-14
“We're making investments across the flywheel, so it is investing in our products, marketing for customer acquisitions. It's also in our partnerships. It's in our lifestyle platform. It's also importantly in AI to drive scale economics as well.”
c8 · CEO, qa, earnings call, 2026-07-14

By quarter

  • Q1 2026described · our inference · exploratory · $17mn to $234mn
  • Q2 2026described · our inference · exploratory · $17mn to $235mn

Cost displaced by AI2 channels · 2 not sized

operations · expensive to verify

Process automation with AI in operations

Not sized

The CFO credits the productivity opportunities to technology and AI automation together, counts processes being mapped rather than work done by AI, and expects the saves over time, so nothing separates AI’s part and the AI saving may not have started; the ceiling is compensation and benefits of , and direct staff fell by on an efficiency program not attributed to AI.

described, no sizedisclosure: described· motive: exploratory· before LLMs: expanded

The CFO counts more than processes being mapped end to end for further automation and names technology and AI automation as a leg of the efficiency push, beside stranded and transformation costs. Direct staff was against a year earlier, with severance of in the quarter, credited to past investments and productivity efforts. The channel stays described and unsized.

Evidence: 7 quotes, 9 figures, 4 confounds, 3 from before coverage

Payroll and contractor cost in Citi's manually intensive processes (operations, controls, servicing) that the end-to-end process work the CEO and CFO describe displaces, which they credit to AI and automation together. The CFO counts it as a leg of a structural efficiency push beside stranded costs and transformation costs, and says the firm is mapping processes end to end for further automation. The line it would show in is compensation and benefits, with direct staff as the visible measure; headcount reductions and severance are part of an efficiency program management does not attribute to AI.

Why this motive

The CFO names productivity from technology and AI automation as a leg of the efficiency push and says saves are expected over time (claims c4 and c3); the processes are being mapped and no result is attributed to AI. AI named with no measure of its own is the exploratory tell, as in Q1.

Before LLMs: expanded

At the anchor the CEO already described automating manual controls and processes as part of the transformation (claim c-anchor-c3), and headcount was being cut through organizational simplification (claims c-anchor-c4 and c-anchor-c5); compensation and benefits was for FY2024 and direct staff at year end. AI put into that work changes its cost per unit, not the existence of the line. The size is the change AI made, not the whole line.

“We're currently deep into a very large body of work, upgrading our data architecture, automating manual controls and processes, consolidating fragmented tech platforms”
CEO, qa, earnings call, 2024-04-12
“As part of these actions, we expect approximately $1.5 billion of annualized run rate saves over the medium term related to a headcount reduction of approximately 7,000.”
CFO, prepared remarks, earnings call, 2024-04-12
“driven by savings related to Citi’s organizational simplification and stranded cost reduction, as well as the lower restructuring charges”
Filing, mdna, 10-K periodic report, 2025-02-21

Figures

  • Compensation and benefits · 2026-CQ2
  • Compensation and benefits, prior year · 2025-CQ2
  • Compensation and benefits growth, year over year · 2026-CQ2
  • Direct staff at quarter end (reported in thousands) · as-of 2026-06-30
  • Direct staff a year earlier (reported in thousands) · as-of 2025-06-30
  • Direct staff, change over the year · 2026-CQ2
  • Severance in the quarter, per the CFO · 2026-CQ2
  • Severance in the first half, per the CFO · 2026-01-01..2026-06-30
  • Processes mapped end to end for further automation, per the CFO (more than) · as-of 2026-07-14

What else could explain it

  • transformation program: Stranded cost and the temporary part of transformation cost are the other legs of the same push (claim c5), and the 10-Q credits lower transformation expenses with offsetting compensation and professional services (claims c14 and c15); none of it is attributed to AI.
  • one time item: Severance of in the quarter and in the half pays for the headcount actions, which the CFO credits to past investments and productivity efforts (claim c3).
  • other: AI is named beside technology and automation as the source of the productivity (claim c4), and nothing separates AI’s part.
  • line composition: Compensation and benefits holds performance-related pay and the compensation of front-office hires the 10-Q names as investments in the businesses.

Quotes

“The third leg of that structural efficiency push is the productivity opportunities that we see from technology and AI automation. I spoke about how we have more than 100 + processes that we're mapping end to end where we see opportunities for further automation. We look at this every week.”
c4 · CFO, qa, earnings call, 2026-07-14
“The benefits of our past investments and productivity efforts have allowed us to gain efficiencies across our expense base and reduce our headcount to 219,000, with over $800 million of severance incurred year-to-date. We continue to invest in areas such as technology, including AI, and we would expect an increase in productivity saves over time.”
c3 · CFO, prepared remarks, earnings call, 2026-07-14
“We talked about transformation cost, and Jane just mentioned that as we reach completion of the programs, we're taking action on the portion of the transformation cost that is temporary.”
c5 · CFO, qa, earnings call, 2026-07-14
“The first quarter was about $500 million. The second quarter just now is another $300 million for an $800 million year to date.”
c7 · CFO, qa, earnings call, 2026-07-14
“In the second half to accelerate some of those productivity opportunities, we may take more severance in the second half, but we're not providing a specific number now.”
c6 · CFO, qa, earnings call, 2026-07-14
“The increase in compensation and benefits expenses was driven by higher performance-related and other compensation and benefits expenses, and higher compensation associated with investments in the businesses, largely offset by productivity savings and lower transformation expenses.”
c14 · Filing, mdna, 10-Q periodic report, 2026-08-06
“The decrease in professional services expenses was driven by lower transformation spend.”
c15 · Filing, mdna, 10-Q periodic report, 2026-08-06

By quarter

  • Q1 2026described · described, no size · exploratory
  • Q2 2026described · described, no size · exploratory

back office · cheap to verify

Staff productivity from AI tools

Not sized

The only measure is a share of staff by count: nearly of Citi’s people use its AI tools (claim c1). A share of users separates no dollars, and no time saved, cost or rate is given, so the channel is directional and gets no ballpark (the methodology's rule for a channel measured only by a count share). The ceiling is compensation and benefits excluding the quarter’s severance, .

described, no sizedisclosure: direction only· motive: exploratory· before LLMs: expanded

The CEO says nearly of Citi’s people use its AI tools and that the tools drive productivity and client experience. A share of users is adoption by count, not a measure of time saved, and nothing is measured, so the channel is read directional and left unsized; compensation and benefits excluding severance, , is the line any saving would sit inside.

Evidence: 3 quotes, 6 figures, 3 confounds, 2 from before coverage

Staff time saved by the AI tools the CEO says nearly all of Citi's people now use, which she says drive productivity and client experience. The saving may show up as fewer hires or more output per head inside compensation and benefits. Separate from the end-to-end process automation in operations, and from the revenue the CEO credits to faster product launches.

Why this motive

The CEO says nearly all staff use the AI tools and that they drive productivity and client experience (claims c1 and c2), with no measured operating result, so the efficiency tell is not met; an implemented tool with no measure is read exploratory.

Before LLMs: expanded

At the anchor generative AI tools were already available to employees in parts of the company (claim c-anchor-c1), and the work is the existing staff cost inside compensation and benefits, for FY2024. No productivity effect from AI tools was measured then. The size is the change AI made, not the whole line.

“Citi has been working with AI and machine learning for a period of time and has more recently begun using Generative AI, a type of artificial intelligence that uses generative models to create text and other content. Generative AI tools are available to employees within parts of the Company, and in the future Citi may more broadly use, develop and incorporate Generative AI within its technology platform and services, systems and its businesses and functions.”
Filing, risk factors, 10-K periodic report, 2025-02-21
“As part of these actions, we expect approximately $1.5 billion of annualized run rate saves over the medium term related to a headcount reduction of approximately 7,000.”
CFO, prepared remarks, earnings call, 2024-04-12

Figures

  • Share of Citi’s people using its AI tools, per the CEO (nearly) · as-of 2026-07-14
  • Compensation and benefits · 2026-CQ2
  • Compensation and benefits excluding the quarter’s severance · 2026-CQ2
  • Compensation and benefits growth, year over year · 2026-CQ2
  • Direct staff at quarter end (reported in thousands) · as-of 2026-06-30
  • Direct staff, change over the year · 2026-CQ2

What else could explain it

  • transformation program: Direct staff fell by over the year, which the CFO credits to past investments and productivity efforts, not to AI tools (claim c3).
  • line composition: Compensation and benefits holds performance-related pay and front-office hiring that move for reasons unrelated to AI.
  • other: The CEO credits the tools with client experience and growth as well as productivity, so time saved may show as service or output rather than lower cost.

Quotes

“As much of the transformation work winds down, we are not only taking down expenses, but we're applying what we learned about large-scale implementation to integrate AI into our businesses and functions wherever it makes sense. Nearly nine out of 10 of our people are using our AI tools.”
c1 · CEO, prepared remarks, earnings call, 2026-07-14
“That's not only driving productivity and client experience, but also growth, helping us bring products to market significantly faster, as we're doing with Payments Express in Services and with our Citi Wealth Advisor Insights platform.”
c2 · CEO, prepared remarks, earnings call, 2026-07-14
“The increase in compensation and benefits expenses was driven by higher performance-related and other compensation and benefits expenses, and higher compensation associated with investments in the businesses, largely offset by productivity savings and lower transformation expenses.”
c14 · Filing, mdna, 10-Q periodic report, 2026-08-06

Revenue arriving through AI2 channels · 2 not sized

product revenue · cheap to verify

Growth from products brought to market faster with AI

Not sized

The CEO credits AI with faster launches and new vectors of growth (claims c2 and c9) with no measure, and the call and the 10-Q credit Services and Wealth growth to deposits, cross-border volumes, assets under custody, fees and mandates; AI is named beside those causes and nothing separates its part (the rule for AI named beside another cause). The ceiling is Services and Wealth revenues of .

described, no sizedisclosure: described· motive: exploratory· before LLMs: relabelled

The CEO says the AI tools help bring products to market significantly faster, naming Payments Express and the Citi Wealth Advisor Insights platform, and that AI opens new vectors of growth in Services; nothing is measured. Services and Wealth revenues were together, the ceiling on any AI part. The channel is read described and left unsized.

Evidence: 2 quotes, 3 figures, 3 confounds, 2 from before coverage

Revenue the CEO credits to AI helping Citi bring products to market significantly faster, naming Payments Express in Services and the Citi Wealth Advisor Insights platform, and the new vectors of growth she says AI opens in Services. The products are unpriced as AI: the AI part is in how they are built, and the revenue is existing Services and Wealth revenue. Payers are corporate and institutional clients (Services) and wealth clients (Wealth), so the counterparty is mixed.

Why this motive

The CEO credits AI with bringing products to market significantly faster and with opening new vectors of growth (claims c2 and c9), with no measured movement in price, conversion or revenue, so the offensive tell is not met; AI named as a cause with no measure is exploratory.

Before LLMs: relabelled

At the anchor the CEO already credited Services growth to product innovations including Citi Payment Express (claim c-anchor-c6), and Services expense to product innovation (claim c-anchor-c2); Services revenue was and Wealth revenue for FY2024. Management gives no measure of the AI-attributed demand, so the tie-break reads it as existing product growth described with AI named.

“Services expenses of $10.6 billion increased 6%, primarily driven by continued investments in technology and platform modernization, other risk and controls and product innovation”
Filing, mdna, 10-K periodic report, 2025-02-21
“We have market-leading product innovations, and with those continue to drive good returns, good growth. If it's Citi Token Services, Citi Payment Express, 24/7 Clearing,”
CEO, qa, earnings call, 2024-04-12

Figures

  • Services total revenues, net of interest expense · 2026-CQ2
  • Wealth total revenues, net of interest expense · 2026-CQ2
  • Services and Wealth revenues together · 2026-CQ2

What else could explain it

  • relabel: Payments Express was already among the product innovations the CEO credited with Services growth at the anchor; the AI part is in how the products are built.
  • mix shift: Services and Wealth growth is credited in the call and the 10-Q to deposits, cross-border volumes, assets under custody, investment fees and new mandates, without AI.
  • other: AI is named beside the deposit, volume, custody and fee causes the call and the 10-Q give for Services and Wealth growth, and nothing separates its part.

Quotes

“That's not only driving productivity and client experience, but also growth, helping us bring products to market significantly faster, as we're doing with Payments Express in Services and with our Citi Wealth Advisor Insights platform.”
c2 · CEO, prepared remarks, earnings call, 2026-07-14
“As we lean into disruption, AI, blockchain, digital commerce, they're all opportunities for us to lead. They're not threats to us. AI is opening up many new vectors of growth and also competitive edge.”
c9 · CEO, qa, earnings call, 2026-07-14

product revenue

Capital raising and financing for the AI buildout

Not sized

The CEO names AI beside tech, data centre, energy and defence capital spending and gives no AI share of Banking revenue, so nothing separates AI’s part (the rule for AI named beside another cause). The ceiling is Investment Banking revenues of , above the prior year.

described, no sizedisclosure: described· motive: exploratory· before LLMs: relabelled

Asked about the pipeline, the CEO said AI dominates client conversations and that capital spending across tech, data centres, energy and defence is accelerating, and named an offering the bank led for a memory maker. Investment Banking revenues were , up . The funding is mixed: borrowers raise from public equity, bond markets and bank debt. The channel is read described and left unsized, as the buildout’s money moving through a bank.

Evidence: 2 quotes, 4 figures, 2 confounds, 1 from before coverage

Investment banking and financing revenue from companies raising money for AI-related capital spending. The CEO says AI dominates the conversations with clients, names tech, data centres, energy and defence capital spending together, and names an offering the bank led for a memory maker. No AI share of Banking revenue is given. Underwriting for builders predates LLMs; this is the buildout's money moving through a bank, not absorption by the borrower.

Why this motive

The CEO names AI as dominating client conversations and a bottleneck ecosystem where activity is strong (claim c10), with no AI share of the business; AI named as part of a cause with no measure is exploratory, and the sources do not contradict each other.

Before LLMs: relabelled

Underwriting and financing for companies raising capital was existing Banking work at the anchor (claim c-anchor-c7), with investment banking fees in Banking of for FY2024; the anchor names no data centre or AI borrower. Management gives no measure of the AI part, so the tie-break reads it as existing underwriting described with AI named.

“Banking includes Investment Banking, which supports clients’ capital-raising needs to help strengthen and grow their businesses, including equity and debt capital markets-related strategic financing solutions and loan syndication structuring”
Filing, mdna, 10-K periodic report, 2025-02-21

Figures

  • Banking Investment Banking revenues · 2026-CQ2
  • Banking Investment Banking revenues, prior year · 2025-CQ2
  • Banking Investment Banking revenues against the prior year · 2026-CQ2
  • Banking Investment Banking revenues growth, year over year · 2026-CQ2

What else could explain it

  • other: Tech, data centre, energy and defence capital spending are named together as the source of the activity (claim c10), and nothing separates the AI part.
  • mix shift: The 10-Q credits Investment Banking growth of to a strong wallet across debt and equity underwriting (claim c13), and does not name AI or data centres.

Quotes

“AI is dominating a lot of the conversations. Tech, data center, energy, defense, CapEx is accelerating. Companies are accessing the public equity and bond Markets alongside bank debt in size. SK hynix, for example, last week is another testament to that. An offering, by the way, that we led. Wherever there's a bottleneck in that whole energy power compute memory ecosystem, we're seeing a lot of activity.”
c10 · CEO, qa, earnings call, 2026-07-14
“Investment Banking revenues increased 44%, reflecting a strong wallet, driven by growth in DCM and ECM, partially offset by a decline in Advisory.”
c13 · Filing, mdna, 10-Q periodic report, 2026-08-06

Reported lines, year-over-year growth

Revenue +14.3%

Q2 2026. Growing slower than revenue: total operating expenses (+4.7%), compensation and benefits (+4.7%), technology/communication (−0.9%), professional services (−14.1%). A displaced cost shows up as a line that stays under the dashed revenue line. These are the audited lines, as first reported; nothing here is attributed to AI by the filing.

Total operating expensesCompensation and benefitsTechnology/communicationProfessional servicesRevenue
-20%-10%0%10%20%Q1 2025Q2 2025Q3 2025Q4 2025Q1 2026Q2 2026RevenueCompensation and benefitsTotal operating expensesTechnology/communicationProfessional services
Reported values and filings
LineQ1 2025Q2 2025Q3 2025Q4 2025Q1 2026Q2 2026
Total revenues, net of interest expense$21.60bn$21.67bn$22.09bn$19.87bn$24.63bn$24.77bn
Total operating expenses$13.43bn$13.58bn$14.29bn$13.84bn$14.31bn$14.21bn
Compensation and benefits$7.46bn$7.63bn$7.47bn$7.07bn$8.38bn$7.99bn
Technology/communication$2.38bn$2.29bn$2.33bn$2.43bn$2.33bn$2.27bn
Professional services$476mn$510mn$514mn$573mn$441mn$438mn