AI in Finance — Workforce Digest – Fortnight of 3–17 August 2026

The monitor’s first run on a fortnightly cadence, covering 3–17 August. Five items, numbered 80–84. Three fall inside the window: the PwC 2026 Financial Services Workforce AI Survey (3 Aug), July BLS financial-activities employment relayed by Inc. (7 Aug), and a Banking Dive / CIO Dive synthesis of Q2-earnings AI-operations commentary (6 Aug). One item predates the window but enters the log now for the priority wealth subsector — the Mariner / Humanity Labs “AI workforce” partnership (14 Jul). One is filed minor and carried only for completeness — JPMorgan / Dimon’s cross-industry AI-risk coalition (5 Aug), a governance matter with no workforce dimension. No new FCA, PRA, Bank of England, HM Treasury, FSSC, SEC, US Treasury or Federal Reserve workforce output this run. Provenance notes indicate the strength of sourcing, not an endorsement of the claims.

Planning to shrink, paying to stay

PwC 2026 Financial Services Workforce AI Survey (3 August) — 78% of the US financial-services leaders surveyed expect their workforce to shrink by at least 20% over the next five years, while 91% are increasing pay for AI-skilled employees, 58% intend to tie compensation to AI-enabled productivity, and 86% say AI-skills training is now more valuable than an MBA for new hires. The same survey records 77% reporting no measurable return on their AI investment and only 42% having done enterprise-wide workforce modelling; PwC notes most firms have not redesigned the roles involved. PwC’s own gloss is that firms are “planning for a smaller workforce and hoping AI fills the gap”; a commentator in the coverage calls it “headcount reduction with an AI narrative layered on top.”

Estimate/model-grade: every figure is a director-level self-report and a forward-looking intention, not an actual. Scope is 1,004 respondents at US firms with $500m-plus revenue — not the sector or the economy — and PwC sells AI and workforce advisory. The contradiction is held within a single instrument: the same firms planning a fifth fewer people are paying up for the people they keep. It continues the ROI-scepticism thread — F2 Strategy’s “ROI remains elusive” (27 Jul) and the “AI as cover” reading on Chime and Visa (3 Aug) — and the skills-premium thread — PwC’s earlier 62% global wage premium and the 88/12 expert-versus-creator split (last run).

The actual decline the reporting won’t pin on AI

July BLS financial-activities employment (via Inc., 7 August) — US “financial activities” employment fell by 14,000 in July to 9.09m, its lowest since July 2022 and roughly 121,000 below the May 2025 peak, with the losses in credit intermediation (−8,800, including 2,500 in commercial banking) and insurance (−6,700). Set directly against that decline, the source reports 48,859 roles requiring AI skills posted by commercial banks over the past year, up 51% year on year — a juxtaposition it labels a “paradox.” The article attributes the headline fall to pre-AI, interest-rate-driven factors and treats AI as context, not cause.

The BLS figures are confirmed government data for the whole US supersector; the 48,859 / +51% postings figure is reported via the article from a labour-analytics source. This is the honest complication to the section above: an actual decline exists, but its own reporting de-links it from AI. No net figure combines the losses and the new postings, and no role- or seniority-level breakdown is given. It continues the “quiet levers” reading — attrition and slow hiring rather than announced cuts — tracked from “use attrition as our friend” through the Bloomberg 28,000-a-month sector decline.

Adoption at scale, the count withheld

Bank of America and Citi, Q2 earnings commentary (Banking Dive / CIO Dive, 6 August) — Banks describe AI reaching scale inside operations: Bank of America reports more than 200,000 staff using AI, over 400,000 prompts a day and 300-plus approved use cases; Citi puts roughly 90% of its people on its AI tools. No headcount figure is attached to any milestone. The workforce dimension appears only as “effects on their workforces” and “operational changes” — named in the abstract, with the counts reserved for adoption.

Reported via the banks’ own earnings calls — interested parties describing themselves. Whether this adoption reduces operations or support roles or merely augments existing staff is not stated: withheld, not disputed. It continues the pattern of augmentation announced without a workforce figure — Barclays’ 8m AI-summarised calls, Bank of America’s EricaAssist for ~18,000 reps, Citi Sky.

Wealth: an execution layer added, the human tier unaccounted

Mariner / Humanity Labs “AI workforce” partnership (14 July) — A five-year partnership deploys “at least 700 FTE” AI agents — costed at roughly $35m a year, about $50,000 per agent — across client onboarding, account opening, compliance reviews, client reporting, billing and prospect onboarding, paired with a plan to grow adviser headcount from 2,000-plus toward 5,000 by 2027. American Banker frames the result as splitting the firm into an “experience” layer of human advisers and an “execution” layer of AI agents; the partners frame it as breaking a “growth ceiling.”

The partnership is confirmed by joint announcement; the ~$35m figure is derived by InvestmentNews from ~700 agents at ~$50,000. A single firm, its own vendor, its own growth framing — and it predates the window, carried now as the run’s only priority-subsector item. It is the one item pointing up. It earns its own section because the functions the ~700 agents absorb — onboarding, compliance support, reporting, billing — are the same operations tier named as exposed elsewhere in the log, here framed as growth rather than reduction, with no figure for the human support roles added or removed. It continues the wealth “experience/execution” split that has repeatedly named the middle- and back-office and paraplanner layer as exposed while adviser relationships endure. The entry pathway into wealth careers that this tier has traditionally formed is left unaddressed — the run’s only contact with the junior-tier question, and only by omission.

Still open

No executed disaggregated headcount data this run: the Q2 2026 US-bank 10-Qs flagged in prior runs did not appear, and the earnings commentary above carries adoption, not headcount. No named-firm headcount-reduction announcement in the priority wealth, private-banking or asset-management subsectors; the prior run’s payments/fintech cuts (Visa ~2,600, Chime ~150) were not extended, and the only named-firm workforce item here — Mariner — is growth-framed and predates the window. No new regulator or industry-body workforce output, reverting to the long-standing silence after July’s single break, when the HM Treasury Financial Services AI Adoption Plan (14 Jul) addressed skills, talent and visas and the Skills Compact launched. Dimon’s “30–40% in certain areas” was not repeated; Dimon appears this run only via the cross-industry AI-risk coalition (5 Aug), a governance initiative with no workforce dimension. And the entry-level and junior tier — whether a thinning junior and operations layer still forms the senior judgement of the future — remains open, present this run only as an absence: no item gives an entry-level or junior share of any figure.

Assembled from a structured fortnightly monitoring run — the monitor’s first at this cadence. Items reflect what was reported during the window; provenance notes indicate the strength of sourcing, not an endorsement of the claims.

SMA Crown Confidential

Digital Confidantes: Bespoke AI Intelligence for Private Decision-Makers