AI Absorption Ledger / PGR

Progressive

PGR · Q2 2026 · reported 2026-07-15 · revenue $23.61bn

Assessment

The Q2 2026 call put AI where an insurer's saving would show. The CEO said the company has turned to generative and agentic AI in many areas, counted "dozens" of advanced initiatives producing results, and expected the first effect to be cost reduction on the LAE and expense ratio side, possibly reaching loss costs later. The CEO declined to give numbers that the CEO said exist, on the ground that they are not complete. The 10-Q, the monthly release and the CEO's quarterly letter again carry no AI passage; in writing, the improvement in the non-acquisition ratio is credited to expense discipline.

Neither line moved the way the claim predicts. Loss adjustment expenses rose against premiums up , above the prior-year rate, and the filing explains the loss and LAE ratio by severity. Other underwriting expenses excluding distribution rose , above the prior-year rate, reversing the first quarter, and the non-acquisition ratio was flat in personal vehicle and up in personal property.

The cost channels stay described and sized only by the ledger: claims-handling savings at , non-acquisition savings at , initiative spend at ; advertising production made with AI stays unmentioned. Nothing changed in state, strength or motive from Q1; what changed is that management named the lines and said a number exists.

Sized channels against the income statement, Q2 2026

3 of 4 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

4 expanded

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$1.8mn to $91mn sized

expanded $1.8mn to $91mn

Incremental total $0 to $91mnpoint $0$16mn in 1 channel has no traced baseline
Cost displaced by AI$0 to $104mn sized

expanded $0 to $104mn

Incremental total $0 to $104mnpoint $12mn

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 · $1.8mn to $91mn sized · $0 to $91mn incremental

engineering

Investment in generative and agentic AI initiatives

0.01% to 0.38% 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 CEO said the company has turned to generative and agentic AI in many areas, counted "dozens" of advanced initiatives, described a new AI strategy council and strategy role, and said the CEO is comfortable with what is being invested, without a number. The size shown is the ledger's own estimate, . The counterparty is mixed: the company's own technology and business staff, and the outside model and tooling vendors beneath them, which the company does not separate.

Evidence: 5 quotes, 2 confounds, 7 from before coverage

What the company spends building and running its generative and agentic AI initiatives: the internal technology and business staff on them, the AI strategy council and the new strategy role paired with the technology chief, and whatever model and tooling bills sit underneath. The company does not separate an outside vendor bill from internal effort, so one channel carries both; the line it would sit in is spread across loss adjustment and other underwriting expenses, and nothing is split out.

Why this motive

The CEO describes studying, testing and a pipeline of initiatives, a new AI strategy council and a strategy officer paired with the technology chief (claims c2, c5 and c6): the investing-for-later tell. The comfort the CEO expresses with returns to date (claim c4) comes with no measure.

Before LLMs: expanded

At the anchor the 10-K said the company had developed and used machine learning and other AI for many years, and on the Q1 2024 call the CEO counted models in tests or production, many of them generative. No amount was given then or in the covered quarters. The size is the whole of an activity that existed before.

“We have developed, and used for many years, new technologies, including machine learning and other forms of artificial intelligence (AI), predictive models, algorithms and automated processes, and will in the future develop and use AI and other new technologies in our business.”
Filing, risk factors, 10-K periodic report, 2025-03-03
“We started with machine learning and large language models probably over a decade ago, and we put a lot of those into place. Think of like we have a chatbot that can give you documents without having a human involved. We've continued on those.”
CEO, qa, earnings call, 2024-05-07
“Right now, we have well over 100 different models in different formats. Some are tests, some are full bore, and some are thoughts, including many in generative AI.”
CEO, qa, earnings call, 2024-05-07
“We did an entire several-day session with our board of directors a few months ago on all that we're doing in AI. And I got to tell you, it's pretty exciting. And it's exciting because of the efficiencies we're going to get.”
CEO, qa, earnings call, 2024-05-07
“These new tools also empower our employees with a new text and email communication platform, which includes a customer-facing generative AI assistant for automated tasks, information retrieval, and tailored follow-up actions. We believe these investments enhance both the customer and employee experience, while increasing efficiency.”
Filing, press release, 10-K earnings release, 2026-03-02
“We're gonna, you know, we're gonna have business models, rigorous controls, and we're gonna continue to invest in this, and we're gonna continue to lean into now gen AI. We recently formed an AI Strategy Council, while we're working on sort of the next year and here's where we're at and working with vendors and making sure that we're testing things to make sure it's a really fluid process,”
CEO, qa, earnings call, 2026-03-03
“We've talked, I think, a little bit over the last year or 2 years, maybe even longer, on what we've worked on in predictive AI. A lot of models using unstructured data, voice data to trigger models, and we'll continue to do that.”
CEO, qa, earnings call, 2026-03-03

What else could explain it

  • line composition: Technology cost is spread across loss adjustment and other underwriting expenses and is not reported as a line.
  • relabel: Chatbots and predictive AI predate generative AI here (claim c1); part of what is called AI work may be the same technology budget under a new name.

Quotes

“They include chatbots and of course we've had predictive AI in our world for some time now as well on, say like things like progressive.com to make decisions on a package that's good for you. We've now turned to GenAI and agentic AI in many areas of our business.”
c1 · CEO, qa, earnings call, 2026-08-04
“I'd say that we have about a dozen, or dozens I should say, of advanced AI initiatives that are producing meaningful. We'll put some dollars to that at some point in the future, but meaningful results and an exciting pipeline of future initiatives.”
c2 · CEO, qa, earnings call, 2026-08-04
“We'll share more of that. I don't want to make headlines that are out there. I could give you numbers now, but I think they're not as complete as I'd like them to be. Rest assured, we're doing a lot in this area. I feel really comfortable with what we're investing and the returns we're getting to date.”
c4 · CEO, qa, earnings call, 2026-08-04
“I think I talked before that about six months ago or so, we formed an AI strategy council. We'd had an AI council. We had had a strategy council. That's why we thought that it would be good to have an AI council.”
c5 · CEO, qa, earnings call, 2026-08-04
“Of course, a little while ago, about a month ago, we added our first Chief Strategy Officer. He works hand in glove with our Chief Technology Officer. You think of the business side of AI and the technology side of AI, and they're helping manage it across the enterprise so we know what our right hand knows what our left hand's doing.”
c6 · CEO, qa, earnings call, 2026-08-04

By quarter

  • Q1 2026described · our inference · exploratory · $1.8mn to $88mn
  • Q2 2026described · our inference · exploratory · $1.8mn to $91mn

Cost displaced by AI3 channels · $0 to $104mn sized · $0 to $104mn incremental · 1 not sized

back office · expensive to verify

Claims-handling cost displaced by generative and agentic AI

0% to 0.26% of the quarter’s revenue

Incremental total: counts in full.

Matched line moved : the whole line, not this channel.

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

The CEO named the LAE side as where AI would first cut cost and said the numbers exist but would not be given until they are more complete (claim c4): a refusal to size, quoted. Loss adjustment expenses rose to , still faster than premiums. The size shown is the ledger's own estimate, .

Evidence: 5 quotes, 7 figures, 2 confounds, 8 from before coverage

Loss adjustment expense (the cost of adjusting and settling claims, mostly claim employees and their systems) that is lower than it would have been because generative and agentic AI takes work out of claims handling. Management names the LAE side as where an AI cost reduction would show first and has not sized it. Claims errors are costly and regulated (unfair claims practices laws), so the work is expensive to verify.

Why this motive

The CEO now names the LAE side as where AI cost reduction would show first (claim c3) and says the initiatives produce results the CEO will size later (claims c2 and c4). The line did not shrink, so the efficiency tell is not met; testing and a pipeline of initiatives keep the reading exploratory.

Before LLMs: expanded

At the anchor employees handled nearly all vehicle claims and loss adjustment expenses in the segment table were for FY2024 against net premiums earned of ; on the Q1 2024 call the CEO already described investing in technology, people and processes to push the line down. The FY2025 shareholder letter (reference quarter 2025-CQ4) says a customer-facing generative AI assistant was put into the claims communication platform during 2025, for automated tasks, information retrieval and follow-up, and credits the claims tools with increasing efficiency. No AI share of the line was named then or in the covered quarters. The size is the change AI made, not the whole line.

“We have developed, and used for many years, new technologies, including machine learning and other forms of artificial intelligence (AI), predictive models, algorithms and automated processes, and will in the future develop and use AI and other new technologies in our business.”
Filing, risk factors, 10-K periodic report, 2025-03-03
“We're constantly investing to push down both loss adjustment expense and non-acquisition expense, figuring out ways with technology, with people, with processes to reduce those expenses because we can give those back in competitive prices, and that's really important.”
CEO, qa, earnings call, 2024-05-07
“We did an entire several-day session with our board of directors a few months ago on all that we're doing in AI. And I got to tell you, it's pretty exciting. And it's exciting because of the efficiencies we're going to get.”
CEO, qa, earnings call, 2024-05-07
“Our employees handle nearly all of our Personal Lines vehicle and Commercial Lines claims from either physical claims offices throughout the U.S. or through a virtual environment, and are supported by centralized functions at our corporate offices and a nationwide network of about 3,700 third-party repair shops.”
Filing, business, 10-K periodic report, 2025-03-03
“These new tools also empower our employees with a new text and email communication platform, which includes a customer-facing generative AI assistant for automated tasks, information retrieval, and tailored follow-up actions. We believe these investments enhance both the customer and employee experience, while increasing efficiency.”
Filing, press release, 10-K earnings release, 2026-03-02
“In our claims organization, in 2025 we implemented a comprehensive set of digital capabilities aimed at modernizing the claims experience and enabling customers to interact when, where, and how they choose. Customers now have digital options starting at the first notice of loss, throughout the investigation, damage assessment, and repair processes.”
Filing, press release, 10-K earnings release, 2026-03-02
“We've talked, I think, a little bit over the last year or 2 years, maybe even longer, on what we've worked on in predictive AI. A lot of models using unstructured data, voice data to trigger models, and we'll continue to do that.”
CEO, qa, earnings call, 2026-03-03
“As with past innovations, I'm confident we will be a leader in our execution, and we'll do it responsibly. That's where we're at. You know, we are getting. You're seeing efficiencies when you look at our data, and I think that will continue.”
CEO, qa, earnings call, 2026-03-03

Figures

  • Loss adjustment expenses, companywide (segment table) · 2026-CQ2
  • Loss adjustment expenses, companywide, prior year (segment table) · 2025-CQ2
  • Loss adjustment expenses growth, year over year · 2026-CQ2
  • Net premiums earned growth, year over year · 2026-CQ2
  • Loss adjustment expenses as a share of net premiums earned · 2026-CQ2
  • Loss adjustment expenses as a share of net premiums earned, prior year · 2025-CQ2
  • Increase in the total loss and LAE ratio, year over year (points) · 2026-CQ2

Reported line it is matched to

Loss adjustment expenses were against , up while net premiums earned rose : of premiums against , above the prior-year rate. The line again moved against the claim, by less than in Q1, and the filing does not attribute the movement.

2026-CQ2: 2025-CQ2: 2026-CQ2: 2025-CQ2: 2026-CQ2:

What else could explain it

  • line composition: Loss adjustment expense carries claim staff, systems, legal defense and outside adjusting costs; the part AI could touch is not split.
  • other: The 10-Q explains the combined loss and LAE ratio by severity and reserve development (claim c8); claims volume from policy growth also moves the line, and policies in force grew faster than premiums, so premiums understate the claims count.

Quotes

“They include chatbots and of course we've had predictive AI in our world for some time now as well on, say like things like progressive.com to make decisions on a package that's good for you. We've now turned to GenAI and agentic AI in many areas of our business.”
c1 · CEO, qa, earnings call, 2026-08-04
“I'd say that we have about a dozen, or dozens I should say, of advanced AI initiatives that are producing meaningful. We'll put some dollars to that at some point in the future, but meaningful results and an exciting pipeline of future initiatives.”
c2 · CEO, qa, earnings call, 2026-08-04
“In terms of probably the first foray that you're going to see with most companies with AI is going to be more of a cost reduction. I think it'd be more on the LAE side and the expense ratio side. As we get further and further into our AI initiatives, I think it could hit more on loss cost, depending on what route we go.”
c3 · CEO, qa, earnings call, 2026-08-04
“We'll share more of that. I don't want to make headlines that are out there. I could give you numbers now, but I think they're not as complete as I'd like them to be. Rest assured, we're doing a lot in this area. I feel really comfortable with what we're investing and the returns we're getting to date.”
c4 · CEO, qa, earnings call, 2026-08-04
“Our total loss and LAE ratio increased 0.6 points and 0.4 points, for the three and six months ended June 30, 2026, respectively, compared to the same periods last year, primarily due to higher severity, partially offset by greater favorable prior accident years reserve development.”
c8 · Filing, mdna, 10-Q periodic report, 2026-08-03

By quarter

  • Q1 2026described · our inference · exploratory · $0 to $60mn
  • Q2 2026described · our inference · exploratory · $0 to $61mn

back office · cheap to verify

Non-acquisition operating expense displaced by generative AI

0% to 0.18% of the quarter’s revenue

Incremental total: counts in full.

Matched line moved : the whole line, not this channel.

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

The CEO named the expense ratio side alongside LAE as where AI would first cut cost. The non-distribution line rose , faster than premiums, and came in against its prior-year share, the reverse of the first quarter. The size shown is the ledger's own estimate, .

Evidence: 7 quotes, 7 figures, 3 confounds, 8 from before coverage

Other underwriting expense outside policy acquisition (mainly employee compensation in customer service, policy servicing and administration, the base of the non-acquisition expense ratio) that is lower than it would have been because generative AI tools do part of the work. Management ties its plan to keep lowering the non-acquisition expense ratio to technology, in an answer about generative AI, and names the expense ratio side as a place AI cost reduction would show. Servicing and document work is quick to check.

Why this motive

The CEO names the expense ratio side as the other early place for AI cost reduction (claim c3), with no measure. The non-acquisition ratio was flat in personal vehicle and rose elsewhere, so no displaced line shrinks; the exploratory reading is kept.

Before LLMs: expanded

At the anchor other underwriting expenses in the segment table were for FY2024, and the non-acquisition expense ratio of the personal vehicle business fell that year, a decline the CFO said had run for at least a decade and that a rising average premium helps. The CEO already described a chatbot that sends documents without a human. No AI share of the line was named. The size is the change AI made, not the whole line.

“We have developed, and used for many years, new technologies, including machine learning and other forms of artificial intelligence (AI), predictive models, algorithms and automated processes, and will in the future develop and use AI and other new technologies in our business.”
Filing, risk factors, 10-K periodic report, 2025-03-03
“We're constantly investing to push down both loss adjustment expense and non-acquisition expense, figuring out ways with technology, with people, with processes to reduce those expenses because we can give those back in competitive prices, and that's really important.”
CEO, qa, earnings call, 2024-05-07
“We started with machine learning and large language models probably over a decade ago, and we put a lot of those into place. Think of like we have a chatbot that can give you documents without having a human involved. We've continued on those.”
CEO, qa, earnings call, 2024-05-07
“We did an entire several-day session with our board of directors a few months ago on all that we're doing in AI. And I got to tell you, it's pretty exciting. And it's exciting because of the efficiencies we're going to get.”
CEO, qa, earnings call, 2024-05-07
“In 2024, our NAER decreased 0.4 points and 0.6 points in our personal vehicle and core commercial auto businesses, respectively, compared to 2023, and increased 0.3 points in our personal property business.”
Filing, mdna, 10-K periodic report, 2025-03-03
“I'd just add that we have continued to make progress on our non-acquisition expense ratio, and we've been doing so over at least the past decade. So structurally, as you say, as we increase average premium, not only the efficiencies we plow into our business, but the denominator is a tailwind for sure.”
CFO, qa, earnings call, 2024-05-07
“3 Primarily consists of employee compensation and benefit costs, and the increase in the allowance for credit loss exposure on our premiums receivable.”
Filing, notes, 10-K periodic report, 2025-03-03
“As with past innovations, I'm confident we will be a leader in our execution, and we'll do it responsibly. That's where we're at. You know, we are getting. You're seeing efficiencies when you look at our data, and I think that will continue.”
CEO, qa, earnings call, 2026-03-03

Figures

  • Other underwriting expenses excluding distribution, companywide (segment table) · 2026-CQ2
  • Other underwriting expenses excluding distribution, companywide, prior year (segment table) · 2025-CQ2
  • Other underwriting expenses excluding distribution, growth year over year · 2026-CQ2
  • Net premiums earned growth, year over year · 2026-CQ2
  • Other underwriting expenses excluding distribution as a share of net premiums earned · 2026-CQ2
  • Other underwriting expenses excluding distribution as a share of net premiums earned, prior year · 2025-CQ2
  • Increase in the personal property non-acquisition expense ratio, year over year (points) · 2026-CQ2

Reported line it is matched to

Other underwriting expenses excluding distribution were against , up against premiums up : of premiums against , against the prior-year rate. The non-acquisition expense ratio was flat in personal vehicle and rose in personal property. The line moved against the claim this quarter.

2026-CQ2: 2025-CQ2: 2026-CQ2: 2025-CQ2: 2026-CQ2: 2026-CQ2:

What else could explain it

  • operating leverage: The ratio's long decline before AI rested partly on a rising average premium (anchor claim pgr-anchor-c8); premium growth slowed this quarter.
  • line composition: The line also carries technology staff, which AI investment adds to, and the allowance for credit losses on premiums receivable.
  • transformation program: The CEO's letter credits the year-to-date personal vehicle improvement to expense discipline and operational efficiencies and does not mention technology or AI (claim c10).

Quotes

“They include chatbots and of course we've had predictive AI in our world for some time now as well on, say like things like progressive.com to make decisions on a package that's good for you. We've now turned to GenAI and agentic AI in many areas of our business.”
c1 · CEO, qa, earnings call, 2026-08-04
“I'd say that we have about a dozen, or dozens I should say, of advanced AI initiatives that are producing meaningful. We'll put some dollars to that at some point in the future, but meaningful results and an exciting pipeline of future initiatives.”
c2 · CEO, qa, earnings call, 2026-08-04
“In terms of probably the first foray that you're going to see with most companies with AI is going to be more of a cost reduction. I think it'd be more on the LAE side and the expense ratio side. As we get further and further into our AI initiatives, I think it could hit more on loss cost, depending on what route we go.”
c3 · CEO, qa, earnings call, 2026-08-04
“We'll share more of that. I don't want to make headlines that are out there. I could give you numbers now, but I think they're not as complete as I'd like them to be. Rest assured, we're doing a lot in this area. I feel really comfortable with what we're investing and the returns we're getting to date.”
c4 · CEO, qa, earnings call, 2026-08-04
“For the second quarter 2026, our NAER was flat in our personal vehicle business compared to the same period last year, while increasing 1.5 points in personal property and 1.1 points in core commercial auto.”
c7 · Filing, mdna, 10-Q periodic report, 2026-08-03
“For the second quarter and first half of 2026, our underwriting expense ratio increased 0.5 points and 0.4 points, respectively, compared to the same periods last year. The increase was primarily attributable to higher advertising spend.”
c9 · Filing, mdna, 10-Q periodic report, 2026-08-03
“As always, we remain focused on expense discipline and operational efficiencies, delivering a 0.2 point improvement in the Personal Lines vehicle (personal auto and special lines) non-acquisition expense ratio (NAER) YTD versus the same period in the prior year.”
c10 · Filing, press release, 10-Q earnings release, 2026-08-03

By quarter

  • Q1 2026described · our inference · exploratory · $0 to $41mn
  • Q2 2026described · our inference · exploratory · $0 to $43mn

marketing · cheap to verify

Advertising production cost displaced by generative AI

Not sized

The call, the letter, the release and the 10-Q say nothing about AI in advertising production this quarter, and advertising production is not split from advertising cost in any filing; the only statement, on the Q4 2025 call, has no figure.

inscrutabledisclosure: not mentioned· motive: exploratory· before LLMs: expanded

The only evidence for this channel is the reference quarter: the FY2025 letter and the Q4 2025 call describe the first AI-generated TV ad, made for much less time and money (anchor claims pgr-anchor-c13, pgr-anchor-c14 and pgr-anchor-c18). Neither covered quarter mentions it; it is carried with no reading of its own. The counterparty is mixed: outside production companies and shoots, and the company's own in-house creative staff.

Evidence: 0 quotes, 5 from before coverage

The cost of producing advertising creative (production companies, shoots, in-house creative time) that is lower because generative AI makes the spot. The CEO said on the Q4 2025 call that the company's first AI-generated TV ad was made in much less time and for much less money than a regular commercial, and that it worked. The media spend that buys the airtime is not part of the channel. Neither covered quarter mentions it: the channel was first seen in reference quarter 2025-CQ4, and firstSeen names the first covered quarter because the schema requires an exhibit there.

Why this motive

No earlier exhibit reading exists to carry; the motive is read from the reference quarter. On the efficiency side the CEO said the AI-generated commercial took much less time and money than a regular one (anchor claim pgr-anchor-c13), but no cost line visibly shrank and advertising cost rose. The deciding tell is the letter's framing of a first ad made to learn new creative tools (anchor claims pgr-anchor-c14 and pgr-anchor-c19); with tells in conflict the less durable motive is taken. Silence this quarter is not evidence about motive.

Before LLMs: expanded

At the anchor the company already made its own television advertising; production sat inside total advertising costs of for FY2024, with no production line split out. The anchor call's AI tests in media are about media buying, not production. On the Q4 2025 call (reference quarter) the CEO credited the first AI-generated TV ad with less time and money, without a figure. The size is the change AI made, not the whole line.

“In fact, we have a board meeting this week, and we're going to go over some AI tests in media that have proven to be really successful.”
CEO, qa, earnings call, 2024-05-07
“I did talk in my letter about a marketing commercial that we did called Drive Like an Animal. That was all AI-generated with the exception of Flo's voice, and we did that in so much less time than a regular commercial and so much less money, and more importantly, it worked.”
CEO, qa, earnings call, 2026-03-03
“We continued to push creative boundaries by producing our first AI-generated TV ad.”
Filing, press release, 10-K earnings release, 2026-03-02
“We complemented these proven strategies by embracing innovation, notably through the integration of artificial intelligence, which enabled greater efficiency and unlocked new creative opportunities.”
Filing, press release, 10-K earnings release, 2026-03-02
“The spot helped us learn how to use new creative tools more effectively while gaining several efficiencies and keeping our very talented creative teams at the forefront of innovation.”
Filing, press release, 10-K earnings release, 2026-03-02

By quarter

  • Q1 2026not mentioned · inscrutable · exploratory
  • Q2 2026not mentioned · inscrutable · exploratory

Reported lines, year-over-year growth

Revenue +7.3%

Q2 2026. Growing slower than revenue: losses and loss adjustment expenses (+7.1%), policy acquisition costs (+3.7%). 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.

Losses and loss adjustment expensesPolicy acquisition costsOther underwriting expensesTotal expensesRevenue
0%10%20%30%40%50%Q1 2025Q2 2025Q3 2025Q4 2025Q1 2026Q2 2026Other underwriting expensesTotal expensesRevenueLosses and loss adjustment expensesPolicy acquisition costs
Reported values and filings
LineQ1 2025Q2 2025Q3 2025Q4 2025Q1 2026Q2 2026
Total revenues$20.41bn$22.00bn$22.51bn$22.75bn$22.19bn$23.61bn
Losses and loss adjustment expenses$12.80bn$13.61bn$13.45bn$14.11bn$13.83bn$14.57bn
Policy acquisition costs$1.46bn$1.51bn$1.55bn$1.57bn$1.54bn$1.57bn
Other underwriting expenses$2.72bn$2.69bn$3.02bn$2.91bn$3.05bn$3.01bn
Total expenses$17.17bn$18.02bn$19.19bn$19.06bn$18.62bn$19.40bn