← back to rankingCFG · Citizens Financial Group, Inc.
Banks - Regional · mkt cap $26.3B · calls: Q1 FY2026 vs Q4 FY2025
57.0 conviction · conf-adj 57
conf 2/10 partial
enthusiasm:27.0 · trend:8 · quantifies:0 · impact:0 · under_radar:0 · credibility:5 · business_impact:8 · disruption:0 · commitment:6 · confirmation:3
Enthusiasm latest 9 / prev 8 (rising)
Citizens' AI thesis is mainly bottomline: automate call center interactions, improve software engineering productivity, and apply analytics to fraud and credit risk. The latest call shows rising credibility because management moved from 2025 pilots and foundations to "real AI use cases in market today," with pilots, production language, and quantified productivity/call-center targets. Topline AI impact is mentioned only indirectly through customer experience, not quantified.
GROUNDED NEXT-FY IMPACT vs CONSENSUS
Grounded on actual base — revenue $11.1B · net income $1.8B · net margin 16.4% · diluted EPS 3.86
These are next-fiscal-year annual uplift estimates, not next-quarter numbers.
Aggregate next-FY est. rev uplift: % · next-FY EPS uplift: % · vs analysts: unclear · priced in: high (model's call-read: medium; verdict above is the hard-data one used for ranking) · confidence: 2/10
| Claim | Figure | Arithmetic | Next-FY Rev % | Next-FY EPS % |
|---|
25% of call-center calls answered by non-humans by YE2026
cost · soft | 25% | Cost-saving lever, but no call-center cost base disclosed anywhere in the claims (no $ opex, no FTE count, no per-call cost). after_tax_saving = 25% * call_center_cost * (1-0.21) is uncomputable. Call-center cost is a small undisclosed slice of total opex (rev 11,148M - op inc 2,328M = ~8,820M), not isolable. Revenue impact = 0 (internal cost initiative); EPS impact cannot be sized. | 0 | |
Ramp to 50% of calls non-human in 2027
cost · soft | 50% | Same missing cost base as above; also 2027 is beyond the next fiscal year, so only a small phasing slice would attribute to FY26 even if sizable. Revenue impact = 0; EPS unanchored. | 0 | |
50% of call-center calls out of human answering (medium-term)
cost · soft | 50% | Medium-term, no call-center cost/FTE/per-call base disclosed. Cannot translate to after-tax saving. Revenue impact = 0. | 0 | |
30% engineer productivity improvement (already seeing)
productivity · soft | 30% | Productivity gain on engineering labor, but no engineering headcount or comp base disclosed in any claim. 30% * eng_payroll * (1-0.21) is uncomputable; eng spend is an undisclosed fraction of ~8,820M opex. Revenue impact = 0 unless it lifts capacity/volume, which is not quantified. | 0 | |
5-10x engineer productivity (some tests)
productivity · soft | 5 to 10x | Reported only for 'some tests', not deployed at scale, and no eng cost base to apply it to. A 5-10x on a tiny test cohort is not a P&L figure. Revenue impact = 0; unanchored. | 0 | |
5-10x engineer productivity (forward use case, high confidence)
productivity · soft | five to 10x | Forward aspiration ('high confidence'), no engineering cost base disclosed. Revenue impact = 0 unless it creates incremental delivery capacity, which is not quantified; EPS cannot be sized. | 0 | |
Assumptions: Current base: revenue $11.148B, net income $1.831B, diluted EPS $3.86, net margin 16.4%, op income $2.328B (implied opex ~$8.82B). Tax rate 21% and incremental margin (default current net margin) prepared but never applicable because no AI claim supplies a dollar base. All claims are adopter-side cost/productivity levers (bottom-line; topline ~0). Phasing: call-center 25% is YE2026 (could touch next FY); 50% ramps in 2027 (mostly beyond next FY); engineer 30% is the only deployed figure, 5-10x is test-stage/aspirational. CFG is a profitable bank, so the loss-maker EPS guardrail does not bite — the blocker is missing cost bases, not a zero earnings base.
Top line: None. Every AI claim is a cost/productivity lever (call-center automation, engineer productivity) that does not add revenue. est_rev_uplift_pct = 0.0% against $11.148B revenue; no supplier-side AI compute/infrastructure revenue exists — CFG is a pure adopter.
Bottom line: Directionally positive but unsizable. The claims (25%->50% calls automated; 30% and aspirational 5-10x engineer productivity) are real cost levers, but management disclosed only percentages and no cost base — no call-center opex, no engineering headcount or payroll. These are undisclosed slices of CFG's ~$8.82B total opex, not isolable. With no denominator, after_tax_saving = saving * (1-0.21) cannot be computed for any claim, so aggregate EPS uplift is null rather than an invented figure.
[impact n/m (all claims soft/unanchored)] Consensus already embeds strong growth (revenue $7.786B->$8.241B, +5.83%; EPS 3.03->3.84, +27%) and CFG beat it (actual EPS 3.86, NI $1.831B vs ~$1.669B consensus). That trajectory is driven by rate/balance-sheet dynamics, not identifiably by AI. Because no AI claim is dollar-anchored, an AI-specific uplift cannot be computed to compare against the consensus slope, so whether AI savings sit above or inside existing estimates is indeterminate; priced_in defaults to medium rather than flagging an unquantifiable above-consensus surprise.
MODEL CONSENSUS (impact)
partial
Both agree: pure adopter, no revenue claims, EPS uncomputable for lack of cost bases. Reconciled to 0 for revenue/supplier fields and the lower confidence.
Conflicts reconciled
- per-claim rev_uplift_pct: X=0 vs Y=null -> used 0 because cost levers add no revenue (method sets topline ~0)
- supplier_rev_uplift_pct: X=0 vs Y=null -> used 0 because CFG is a pure adopter with no supplier-side AI revenue
- confidence: X=3 vs Y=2 -> used 2, the more conservative value given all bases are missing
| Field | Opus 4.8 | GPT-5.5 |
|---|
| Rev uplift % | 0.0 | 0 |
| EPS uplift % | – | – |
| Priced in | medium | medium |
| vs analysts | unclear | unclear |
| Confidence | 2 | 3 |
| Top line | None. Every AI claim is a cost/productivity lever (call-center automation, engineer productivity) that does not add revenue. est_rev_uplift_pct = 0% by construction; no supplier-side AI compute/infrastructure revenue exists here — CFG is a pure adopter. | No quantified revenue uplift is claimed. All disclosed AI metrics are internal adoption/productivity claims, so estimated revenue uplift is 0.0% against $11.148B current revenue unless management later quantifies capacity-led revenue gains. |
| Bottom line | Directionally positive but unsizable. The claims (25%→50% calls automated; 30% and aspirational 5-10x engineer productivity) are real cost levers, but management disclosed only percentages and no cost base — no call-center opex, no engineering headcount or payroll. Call-center plus engineering cost are undisclosed slices of CFG's ~$8.82B total opex (rev 11,148M − op inc 2,328M), not isolable from the inputs. With no denominator, after-tax savings = saving * (1−0.21) cannot be computed for any claim, so the aggregate EPS uplift is null rather than an invented figure. | Potential bottom-line benefit exists from 25%-50% call deflection and 30% to 5-10x engineering productivity, but the required dollar bases are absent. EPS uplift cannot be responsibly calculated versus $1.831B current net income. |
| Reasoning | Consensus already embeds strong EPS growth (3.03 in FY24 → 3.84 in FY25, +27%) and CFG beat it (actual EPS 3.86, NI 1,831M vs 1,669M consensus). That trajectory is driven by rate/balance-sheet dynamics, not identifiably by AI. Because no AI claim is dollar-anchored, I cannot compute an AI-specific uplift to compare against the consensus growth slope — so whether AI savings sit above or inside the existing estimate is indeterminate (unclear), and I default priced_in to medium rather than flagging an above-consensus surprise I can't quantify. | Consensus revenue rises from $7.786B in 2024 to $8.241B in 2025, a $454.3M increase or 5.83%; consensus EPS rises from $3.02616 to $3.83719, up 26.80%. The AI math here produces 0.0% hard revenue uplift and no calculable EPS uplift, so there is no hard evidence that AI adds above consensus, despite qualitative cost upside. |
Rows highlighted where the two models disagreed.
QUANTIFICATIONS
call center calls answered by non-humans: 25% (by the end of the calendar year, bottomline)
“In fact, we expect inside of this calendar year, by the end of the year, we should have 25% of our calls answered by non-humans with the expectation that will ramp in 27% to 50%.”
call center calls answered by non-humans: 50% (ramp after 2026 / 2027 per transcript context, bottomline)
“In fact, we expect inside of this calendar year, by the end of the year, we should have 25% of our calls answered by non-humans with the expectation that will ramp in 27% to 50%.”
engineer productivity improvement: 30% (already seeing / tests done, bottomline)
“We're already seeing a very material productivity improvement and leverage we're getting on our capital investment and deployment ranging from 30% improvement in productivity that in some tests we've done, it's been a 5 to 10x improvement in productivity.”
engineer productivity improvement: 5 to 10x (some tests, bottomline)
“We're already seeing a very material productivity improvement and leverage we're getting on our capital investment and deployment ranging from 30% improvement in productivity that in some tests we've done, it's been a 5 to 10x improvement in productivity.”
call center calls removed from human answering: 50% (medium-term outlook, bottomline)
“The call center as an example, we think that combination of modernizing the tech stack for the call center front to back plus introducing voice AI and other mechanisms, we can get in the range over the medium-term outlook 50% of our call center calls out of a human answering them.”
engineer productivity: five to 10 x (forward use case / confidence statement, bottomline)
“Technology development, productivity of an engineer is a use case that we also have high confidence in that through leveraging AI, we can have, you know, a five to 10 x of productivity with our engineers that the AI is taking the first crack at writing the code where developers are now QA, QC ing the code, adding the last mile and then ultimately having the AI also work on the first round of testing and quality assurance of the code.”
PAST (realized)
- Our AI spend has I would call it backwards looking in 2025 has been a combination of very small targeted pilots and learnings and building the right control infrastructure to get ourselves ready for this, including moving completely to the cloud, in, in '20 in 2025.
- And big investments in data.
- Candidly, of this started back in 2020, 2021 where we're really getting, some bigger investments.
CURRENT (now)
- We are leveraging AI to assist in writing code and expect to have material productivity improvements in software development, cutting down cycle times.
- We are also using AI to improve our interactions with customers, which we expect will materially cut call volumes and improve the overall customer experience.
- We very much have real AI use cases in market today.
- It's already in pilot.
- So this is very real.
- So we are moving very, very fast.
FORWARD (guidance)
- In fact, we expect inside of this calendar year, by the end of the year, we should have 25% of our calls answered by non-humans with the expectation that will ramp in 27% to 50%.
- That really should hit in the summer and into the early fall.
- And so maybe that creates a higher level of benefit, maybe that creates an acceleration and maybe it just creates new work streams that we haven't even thought or possible.
- The call center as an example, we think that combination of modernizing the tech stack for the call center front to back plus introducing voice AI and other mechanisms, we can get in the range over the medium-term outlook 50% of our call center calls out of a human answering them.
- So you'll start to see in the reimagine the bank, effort, there's an overlay of tech spend in addition to our run rate tech spend we're spending on the franchise that will be principally pointed at AI deployment, for reimagine the bank.
TRACK RECORD — PROMISE vs DELIVERY
55/100 track record too-early 6 calls reviewed
CFG's only quantified AI-related promises are the GenAI/Agentic-AI-driven 'Reimagine the Bank' program (~$450M-by-2028 P&L benefit, ~$100M 2026 exit run-rate, plus AI coding and call deflection), all introduced in mid-to-late FY2025 and consistently reaffirmed through Q1 FY2026 but with no 2026-2028 milestone yet reached, so their AI track record is not yet judgeable.
Reimagine the Bank: fully phased-in run-rate benefits greater than TOP six's >$400M, with net benefits beginning in 2027 and accelerating into 2028 — promised Q3 FY2025
too-early Reaffirmed and refined to a $450M P&L target by end of 2028 in Q4 FY2025 and Q1 FY2026; program only just launched, so the 2027-2028 timeframe has not arrived
$450M Reimagine-the-Bank P&L benefit by end of 2028 (GenAI/Agentic AI, AI coding, AI call deflection) — promised Q4 FY2025
too-early Explicitly reaffirmed in Q1 FY2026 ('reaffirm our $450 million P&L target by the end of 2028'); cannot be judged this early
~$100M annualized exit run-rate pretax benefit from Reimagine the Bank by end of 2026 — promised Q4 FY2025
too-early Reaffirmed in Q1 FY2026 ('exit 2026 with an annualized run rate of about $100 million'); year-end 2026 milestone not yet reached as of mid-2026
2026 Reimagine-the-Bank one-time costs of ~$50M effectively offset by ~$45M of benefits realized later in the year — promised Q4 FY2025
too-early Q1 FY2026 reported ~$6M of implementation costs ramping with saves expected to benefit 2H 2026; full-year outcome still pending
AI used to write code for material software-development productivity and to materially cut call volumes (within the $100M 2026 exit run-rate) — promised Q1 FY2026
too-early Stated as in-progress work streams in Q1 FY2026 with no quantified delivered result yet; too soon to verify
PRICED-IN (REFINED)
HIGH (already in)Est. revisions rising · Fwd P/E 22.4 · EV/Sales 2.3x
AI claim maps to Service Charges and Fees, Card Fees
Estimate signals lean rising: recent price targets are above the last-year average and forward EPS/revenue estimates show consensus already baking in growth, while ratings remain strongly Buy-heavy. The 22.4x forward P/E is rich for a mature regional bank, even though P/B is near 1.0 and EV/Sales is moderate. AI-related efficiency or revenue upside would most plausibly show through Service Charges and Fees and Card Fees, but the combination of rising estimates and a rich forward multiple indicates that upside is already substantially priced in.
COVERAGE — ENTHUSIASM TRAJECTORY + CATALYSTS
1Q4 FY20241Q1 FY20257Q2 FY20257Q3 FY20258Q4 FY20259Q1 FY2026
AI enthusiasm across 6 calls — trend ↗ rising
AI moved from absent to a named bankwide transformation with specific initiatives, cost-benefit timing, and a $450M 2028 P&L target.
RECENT AI CATALYSTS & NEWS
BUSINESS IMPACT - QUALITATIVE MATERIALITY
7/10 qualitative impact material medium-term · mixed evidence
Where AI matters: call-center automation, software productivity, fraud/credit workflows
Citizens has moved beyond vague AI pilots into deployed call-center automation and AI-assisted engineering, with targets for 25% of calls handled by non-humans by year-end and roughly 50% over time. The upside is mostly cost and productivity, not revenue transformation, and the AI-specific P&L contribution remains hard to isolate despite the broader $450M efficiency target.
Caveats: AI savings may be embedded in broader Reimagine the Bank targets rather than incremental upside; Execution risk around customer experience, compliance, model governance, and fraud controls; Call-center and engineering cost bases are undisclosed, limiting confidence in EPS materiality; Larger banks may deploy AI at greater scale and widen technology gaps
AI DISRUPTION / CANNIBALIZATION RISK two-sided · 3/10
AI does not automate away Citizens' core balance-sheet model of deposits, lending, credit risk, and regulated banking relationships. The risk is mainly that AI-native fintechs or larger banks commoditize customer service, underwriting, and digital banking UX, pressuring regional-bank differentiation rather than destroying the model.
OPTIONS / MARKET STRUCTURE
option liquidity: good
proxy inputs — dollar-ADV $286M · beta 1.041 · px $62.31
source: proxy (no options chain on FMP)
FMP /stable/ exposes no options-chain endpoint on this key, so ATM IV, bid-ask spread and open interest are unavailable. Liquidity below is a PROXY from dollar-ADV, beta and price level (a stand-in for option depth), not measured option-market data.
CONFIRMATION — INSIDERS · 13F · LANGUAGE
Mixed — insiders selling, institutions adding, management language 7/10 committed.
INSIDERS selling 2 open-market sell(s) vs 0 buy(s) — net distribution
INSTITUTIONS (13F) adding as of 2026-03-31: 141 new / 98 closed positions; 571 increased / 306 reduced; institutional ownership -3.23pp; +43 net 13F holders
MGMT LANGUAGE 7/10 committed Limited AI discussion, but management uses ownership language, concrete productivity areas, and program targets tied to business benefits.
commit “we reaffirm our $450 million P&L target by the end of 2028.”
commit “we are leveraging AI to assist in writing code and expect to have material productivity improvements in software development”
commit “We are also using AI to improve our interactions with customers”
VERBATIM AI QUOTES
“We are leveraging AI to assist in writing code and expect to have material productivity improvements in software development, cutting down cycle times.”
— Aunoy Banerjee, Q1 FY2026
“We are also using AI to improve our interactions with customers, which we expect will materially cut call volumes and improve the overall customer experience.”
— Aunoy Banerjee, Q1 FY2026
“I think that's a really good call out, Ryan, is that the adoption curve, the innovation curve that we're seeing in AI is really -- it's almost mind boggling.”
— Bruce Van Saun, Q1 FY2026
“And so maybe that creates a higher level of benefit, maybe that creates an acceleration and maybe it just creates new work streams that we haven't even thought or possible.”
— Bruce Van Saun, Q1 FY2026
“So we won't create a lot of science fair projects and kind of use some of this new technology in ways that actually don't deliver real benefits.”
— Bruce Van Saun, Q1 FY2026
“We very much have real AI use cases in market today.”
— Brendan Coughlin, Q1 FY2026
“In fact, we expect inside of this calendar year, by the end of the year, we should have 25% of our calls answered by non-humans with the expectation that will ramp in 27% to 50%.”
— Brendan Coughlin, Q1 FY2026
“We're already seeing a very material productivity improvement and leverage we're getting on our capital investment and deployment ranging from 30% improvement in productivity that in some tests we've done, it's been a 5 to 10x improvement in productivity.”
— Brendan Coughlin, Q1 FY2026
“Our AI spend has I would call it backwards looking in 2025 has been a combination of very small targeted pilots and learnings and building the right control infrastructure to get ourselves ready for this, including moving completely to the cloud, in, in '20 in 2025.”
— Brendan Coughlin, Q4 FY2025
“But as we turn the page to 2026, the dial turns a little bit.”
— Brendan Coughlin, Q4 FY2025
“The call center as an example, we think that combination of modernizing the tech stack for the call center front to back plus introducing voice AI and other mechanisms, we can get in the range over the medium-term outlook 50% of our call center calls out of a human answering them.”
— Brendan Coughlin, Q4 FY2025
“Technology development, productivity of an engineer is a use case that we also have high confidence in that through leveraging AI, we can have, you know, a five to 10 x of productivity with our engineers that the AI is taking the first crack at writing the code where developers are now QA, QC ing the code, adding the last mile and then ultimately having the AI also work on the first round of testing and quality assurance of the code.”
— Brendan Coughlin, Q4 FY2025
“And then maybe lastly or just analytics, fraud, credit risk.”
— Brendan Coughlin, Q4 FY2025
“So you'll start to see in the reimagine the bank, effort, there's an overlay of tech spend in addition to our run rate tech spend we're spending on the franchise that will be principally pointed at AI deployment, for reimagine the bank.”
— Brendan Coughlin, Q4 FY2025
ANALYST QUESTIONS ON AI
Q (Q1 FY2026, Ryan Nash): In the slides, you highlighted some of the things that you're doing with reimagine the bank, including corporate and the LLMs and a handful of things. I guess, given the pace of change we're seeing in the markets in areas like AI, are there opportunities to accelerate any of these initiatives or adjust the timing given, again, just the rapid pace of change that we're seeing?
A: I think that's a really good call out, Ryan, is that the adoption curve, the innovation curve that we're seeing in AI is really -- it's almost mind boggling. It's very significant. And so maybe that creates a higher level of benefit, maybe that creates an acceleration and maybe it just creates new work streams that we haven't even thought or possible.
Q (Q4 FY2025, David Shibarini): And my follow-up is on AI. You've been front-footed on AI some of your peers. Can you talk about your AI spend and some of the use cases you're seeing?
A: Our AI spend has I would call it backwards looking in 2025 has been a combination of very small targeted pilots and learnings and building the right control infrastructure to get ourselves ready for this, including moving completely to the cloud, in, in '20 in 2025. But as we turn the page to 2026, the dial turns a little bit.