Four functions of AI
Today's consumer-facing AI sorts into a ladder of autonomy — Informs → Analyzes → Advises → Acts. Incumbents cluster in the first two; fintechs appear across all four.
Corporate Insight Research · Synthesized for Leadership by XD/Chris Metzner
AI in Financial Services:
The Opportunity, the Risk & the Emerging Reality
Two-thirds of consumers already assume their financial provider uses AI — and their default read is negative. This briefing maps where the industry stands today, what the leaders are getting right, and where the real opportunity lies.
Based on two studies:
- AI in Financial Services Today — June 2026 · 2,000+ consumer survey, interviews & monitoring
- Desktop Virtual Assistants and AI — August 2026 · benchmark of 9 brokerages + 4 mobile leaders
01 · Executive Summary
The gap isn't capability. It's trust — and execution.
AI has powered financial services for a decade, but generative AI reset expectations overnight. Fintechs sprinted ahead; incumbents held back for good reasons. The urgent problem now is that consumers are skeptical, and much of what firms have already shipped underperforms.
0%
assume their financial & insurance providers already use AI moderately or extensively
0%
are more concerned than excited about AI in society — vs. just 19% more excited
0%
have used AI — nearly a third (31%) use it multiple times per day
0%
disconnect from a support chat the moment they believe they're talking to an AI
Fintechs set the pace
Digital-native challengers ship the most sophisticated AI — concentrated in trading-adjacent verticals where data is rich and risk tolerance is high.
Trust must be earned
Consumers use AI heavily but distrust it in financial settings. They read corporate AI as cost-cutting until proven otherwise.
Three comfort drivers
Transparency, an easy path to a human, and human review of consequential decisions consistently raise consumer comfort.
Execution is the real gap
Half of tested automated-investing & retirement VAs failed on a single-word typo. Fix what's deployed before building what's next.
ROI is retention, not acquisition
Only a minority pick a firm because it offers AI. Near-term value is deepening existing relationships and trust.
“They all say, ‘we use AI to better serve you.’ Okay — prove it. Actions speak louder than words.”
— Jessica I., consumer interview participant
02 · The AI Landscape
From information to action: a map of increasing autonomy
This framework measures function, not quality. A tool that informs well isn't inferior to one that acts — the right tier is the one that serves your clients without eroding trust.
1
AI Informs
Information only — the user interprets
- VAs surface account info & navigate to pages
- Market summaries from news, filings & analyst reports
Examples: most incumbent VAs; Robinhood Digests, Coinbase & Webull summaries
2
AI Analyzes
Insights & recommendations — the user decides
- Account insights, spending & forecasts
- Portfolio analysis with market context
Examples: BofA Erica, Wells Fargo Fargo, Schwab Portfolio Insights, Citi Sky; Robinhood Cortex, Public Alpha
3
AI Advises
Specific advice — the user executes
- Tailored investment guidance (with/without guardrails)
- SEC-registered AI advisors emerging
Examples: Origin, Magnifi, AInvest; Vanguard Digital Advisor AI (pilot)
4
AI Acts
Strategy co-pilot — the AI executes
- Parameter-based, semi-agentic trade execution
- Agentic access via API / MCP integrations
Examples: Composer, Public, Interactive Brokers, Robinhood, eToro, Gemini — no incumbents
Incumbents are active here
← →
Fintechs are pushing here
Representative screenshots from the Corporate Insight reports (internal use only). © Corporate Insight.
Why incumbents proceed with caution: deterministic, rule-based VAs give firms control over outputs and materially reduce hallucination and compliance risk — a rational trade-off given fiduciary exposure. In testing, tools in the Advises tier have produced misleading output, and consumer appetite for the Acts tier remains narrow.
03 · Key Trends
What's driving the market right now
Fintechs claimed the frontier
Challengers deploy the most advanced consumer-facing AI — concentrated in crypto and brokerage, where data richness and risk tolerance reward bold moves.
Incumbents move — carefully
The Analyzes tier is heating up: Schwab launched Portfolio Insights, Citi announced Sky, and Vanguard is piloting advice-grade AI for Digital Advisor.
Agentic AI is arriving
“AI Acts” is emerging via API/MCP integrations — Public, Interactive Brokers, Robinhood, eToro and Gemini — but always with a human checkpoint or user-defined rules.
Transparency is table stakes
Popular LLMs like ChatGPT and Claude now set the bar for disclosure, error reminders and reasoning. Consumers expect the same honesty from their providers.
The advisor relationship is a target
AI-forward firms like Origin position their tools as replacements for a human advisor — holistic, on-demand planning and personalized recommendations.
AI wins loyalty, not new logos
AI doesn't win new customers today. Its near-term value is retention — deepening trust with the customers a firm already has.
04 · Consumer Sentiment
High usage, high distrust
Consumers use AI constantly, but that doesn't mean acceptance. In financial settings, skepticism is the baseline — and firms must design for an audience that is cautious at a minimum.
Sentiment toward AI in society
Share of respondents by overall attitude
19%
31%
43%
- More excited
- Equally concerned & excited
- More concerned
What comes to mind when a company adds AI?
Top associations — the dominant read is negative
Relative prominence shown; two of the top three associations are negative.
Sentiment by generation
Excitement skews young — but concern persists across every age group
Gen Z
24
34
37
Millennial
25
34
36
Gen X
20
32
41
Boomer
10
26
57
- More excited
- Equally
- More concerned
Preferred level of AI involvement, by task
The higher the stakes, the less AI consumers want
Place investment trades
59
33
8
Pay my credit card bill
56
30
14
Update contact information
45
45
10
- No AI involvement
- AI advises, I decide
- AI handles it completely
73%
of “more concerned” consumers disconnect once they sense an AI — and those who disconnect are far less satisfied (38% top-two-box).
“I see AI as a collaborator: they're the passenger; I'm the driver.”
— consumer interview participant
The bright spot: consumers broadly trust financial institutions to protect their data — a reserve of goodwill fintechs can't easily replicate. Firms that deploy AI transparently can draw on that foundation; those that deploy it in ways that feel self-serving undermine it.
05 · The Three Comfort Drivers
What actually makes consumers comfortable with AI
These are not capabilities — they are execution choices within every firm's control. They matter most for the large middle group who are skeptical but still use AI.
“This would make me more comfortable using AI tools”
Share of all respondents selecting each driver
1
Transparency & disclosure
Whether or not AI is actually present is irrelevant if users perceive it. Disclose AI across all contexts — and be honest about limits.
- Name the tool to signal automation (“virtual assistant”); use icons, not human faces
- Static, plain-language disclosures — not “powered by advanced algorithms”
- Sparkle icons, source links, and error/advice disclaimers beside output
- Show reasoning for high-stakes output (Origin, Public, Composer)
Doing it well: eToro Tori names its OpenAI models; Robinhood labels Digests “AI-powered, not advice.”
2
Human escalation
The single highest-rated comfort driver. Design escalation as a visible feature, not an occasional fallback.
- Surface a static, persistent “contact a human” option
- Embed live chat inside the VA — don't reset the conversation
- Pre-collect context so users never re-explain their issue
- Avoid containment tactics like “Are you sure I can't help?”
Doing it well: Schwab's persistent “Contact a CSR” button; SoFi's 24/7 handoff framing.
3
Human review of decisions
Consumers want to keep decision-making authority — especially as stakes rise. Surface the human element in the loop.
- Show where humans review AI output, especially for high-stakes actions
- Explain AI-driven back-end decisions in plain language
- Emphasize curated, human-approved content libraries
- Insert a mandatory checkpoint before consequential execution
Doing it well: Lemonade auto-routes non-instant claims to humans; IBKR requires review before any AI-drafted order executes.
06 · What Firms Do Well
Best-in-class examples worth emulating
A handful of firms show what modern financial AI looks like — conversational, account-aware, transparent, and reliably connected to a human.
Robinhood
FintechAI VA & Cortex — Analyzes
Replaced a deterministic assistant with a multi-turn, context-aware AI that draws on account holdings, watchlists and market data. Asks for clarification instead of restarting; keeps a full chat history.
SoFi Coach
FintechGenerative VA — Analyzes
Conversational, strong language comprehension, retains context across a session. Surfaces portfolio concentration and a holdings table in-chat, and exposes a summary of its reasoning steps.
Bank of America Erica
IncumbentMobile VA — Analyzes
Among the most capable mobile assistants: understands text and voice, handles informal phrasing, seldom errors, and reliably redirects when a request falls outside its scope.
Wells Fargo Fargo
IncumbentMobile VA — Analyzes
Excels at everyday servicing with text/voice input and accurate speech-to-text. Integrates PFM insights like credit scores and spending, with a full Spanish-language version.
Schwab Portfolio Insights
IncumbentGenAI feature — Analyzes
The first incumbent client-facing GenAI feature at scale: a narrative snapshot of portfolio performance, curated SCFR commentary, and a news recap — with thorough disclosures.
Fidelity Assistant
IncumbentDeterministic VA — Informs
One of the strongest deterministic chatbots: predictive text as you type, clarifying follow-ups, and availability on both the public and secure sites.
Lemonade Maya
FintechClaims VA — trust by design
Dual-routing claims: AI can approve instantly, but anything else automatically routes to a human before a decision — addressing the exact fear consumers voice about claims.
Origin
FintechAI Advisor — Advises
The clearest example of the leading tier: an SEC-registered RIA whose multi-model AI gives specific recommendations grounded in linked accounts — with visible reasoning and disclaimers.
Product screenshots captured from the Corporate Insight reports (subscriber content, internal use only). © Corporate Insight.
The leading deterministic VA playbook
For firms not ready for GenAI, these six practices define the best version of a rule-based assistant:
Navigational support
A smart routing layer with NLP that interprets typos, slang and varied phrasing.
Self-service tasks
Highest-volume tasks (password resets, updates) reachable in a single prompt.
Embedded instructions
Pair every redirect with a brief in-assistant, step-by-step sequence.
Live-agent redirection
Frictionless escalation with context pre-collected so users don't repeat themselves.
Account information
Balances, holdings and activity shown directly in the chat window.
Personalized insights
Timely, tailored content built on top of the foundational capabilities.
07 · Desktop VA Benchmark
How the brokerages actually performed
Corporate Insight tested a consistent set of prompts across nine desktop brokerage VAs plus four leading mobile assistants. The pattern is stark: two AI-powered tools pull far ahead of the deterministic field.
Y — Handled
P — Partial
N — Failed
— N/A (no true VA)
| Prompt | Ally | Ameriprise | Schwab | E*TRADE | Fidelity | JPM | Robinhood | SoFi | TIAA | BofAM | RobinhoodM | U.S. BankM | Wells FargoM |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| What can you help with? | — | N | Y | N | Y | — | Y | Y | N | Y | Y | Y | Y |
| Do you employ generative AI? | — | N | N | N | N | — | N | Y | N | Y | Y | N | N |
| Buy three shares of GE | — | P | P | N | N | — | P | P | N | P | Y | N | N |
| GE quote | — | N | Y | N | Y | — | Y | Y | N | Y | Y | N | N |
| Show me my holdings | — | N | Y | N | P | — | Y | Y | Y | Y | Y | P | N |
| Market performance resources | — | N | Y | N | P | — | Y | N | N | Y | Y | N | N |
| Show me my watchlist | — | N | Y | Y | P | — | N | N | N | N | Y | N | N |
| Help me create an alert | — | N | Y | Y | Y | — | Y | Y | N | Y | Y | Y | Y |
| Show me my tax documents | — | Y | Y | Y | Y | — | Y | Y | Y | Y | Y | Y | Y |
| Speak with a representative | — | Y | Y | Y | Y | — | Y | Y | Y | Y | Y | Y | Y |
| Update my email address | — | Y | Y | Y | Y | — | Y | Y | Y | Y | Y | Y | Y |
| Update my password (typo) | — | Y | P | N | Y | — | Y | Y | Y | Y | Y | Y | Y |
M = mobile example. Ally funnels straight to live chat; J.P. Morgan Self-Directed Investing has no full VA (a Secure Messages tool instead). Highlighted columns are the AI-powered standouts.
2
firms — Robinhood & SoFi — replaced deterministic VAs with AI that holds multi-turn, account-aware conversations.
0
incumbent brokerage VAs could fully place a trade — and only two could route a client to their watchlist.
48%
of firms place the VA in the main navigation — a fundamental findability gap even for capable tools.
44%
of qualifying firms offered multi-modal responses — and none did so consistently.
08 · Areas of Opportunity
The biggest gaps — and where to win
The most urgent problem isn't missing advanced capability. It's that much of what's already deployed underperforms on the basics that build trust.
50%
Comprehension basics
Half of tested automated-investing and workplace-retirement VAs failed on a single-word typo. Robust NLP is achievable at scale — and quietly erodes trust when it's missing.
48%
Findability
Only about half of firms put the VA in the main navigation. A capable tool nobody can find delivers no value — Robinhood's strong VA is curtailed by poor findability.
44%
Multi-modal responses
Pairing an in-assistant answer with a link and brief instructions accommodates more users — yet fewer than half of qualifying firms do it, and none consistently.
0%
Goal-based guidance
No automated-investing VA delivers personalized goal-related insights in-chat — a glaring gap for a model built entirely around reaching financial milestones.
0–6%
AI disclosure
Almost no insurance or retirement VAs disclose they're AI-powered (6% in banking). Disclosure is a low-cost, high-impact comfort driver being left on the table.
⚠
Context-losing handoffs
Users repeatedly cite re-explaining an issue after a handoff as a top frustration. Direct, in-assistant transitions that preserve context are strongly preferred.
A cautionary note on racing ahead: in the Advises tier, minimally-guardrailed tools like AInvest's Aime have produced hallucinations, spelling errors, stray foreign-language characters and even inappropriate content — a brand-liability risk. A failed or disorienting AI interaction can do more reputational harm than no AI at all.
09 · Industry Snapshots
Where each vertical stands today
AI deployment is nascent and uneven. VAs that surface account information remain the predominant form; analysis is the next frontier.
| Industry | AI Informs | AI Analyzes | AI Advises |
|---|---|---|---|
| Consumer banking | |||
| Brokerage | |||
| Automated investing | |||
| Insurance | |||
| Workplace retirement | |||
| Workplace financial wellness |
- None
- Early / limited
- Moderate
- Substantial
- Widespread
Consumer banking
94% have a web VA6% disclose it as AI
Gap: chatbots handle basic queries only; AI browsers can now act on authenticated sites.
Priority: reframe AI as a complement to human service; extend into budgeting & cash management.
Brokerage
38% have a web VA100% offer live-chat escalation
Gap: client-facing GenAI nearly nonexistent; “advice” regulation raises the cost of missteps.
Priority: roll out GenAI in stages — contained use cases first, with transparent disclosure.
Automated investing
57% have a VA50% failed a one-word typo
Gap: goal-related information is nearly absent — the core of the robo model.
Priority: connect VA experiences to goal tracking and portfolio data.
Asset management
2 of 20 deployed advisor AI49% of advisors excited
Gap: most AI investment is internal; advisor-facing deployment is early and narrow.
Priority: convert portfolio data into client-ready observations that free advisor time.
Insurance
61% P&C / 50% life / 39% annuity have a VA0% disclose AI
Gap: customer-facing chatbots vary widely; internal AI is far more mature.
Priority: guide coverage decisions and execute policy & claim functions with trust.
Workplace retirement
44% have a VA0% disclose AI
Gap: chatbots redirect rather than embed info; only four offer a mobile VA.
Priority: personalize with account-level data; make escalation consistent.
Workplace financial wellness
71% have a VA80% indicate it's AI-powered
Gap: execution is uneven; many link to content rather than generate guidance.
Priority: conversational coaching that builds context and ties to the benefit package.
10 · The Path Forward
Getting there first matters less than getting it right
Incumbents' AI problem is execution. Trust is their advantage — and it must be earned on the experiences they've already shipped before building the next ones.
Fix what exists before building what's next
Building new capability on top of experiences that erode trust is a losing sequence. Improving existing VAs is a precondition for new features earning consumer confidence — not a maintenance task.
Measure success as retention, not acquisition
Only a minority choose a firm because it offers AI. The right internal metric is whether the customers already open to a good experience are getting one.
Calibrate to your actual client base
The mainstream consumer isn't the early adopter on an agentic-trading waitlist. 59% want no AI in placing trades. Build the capability calibrated to that reality — not the most advanced option available.
Compete on execution quality
As firms converge on similar GenAI features, the contest shifts to how well the tool handles failure, discloses limits, and hands off to a human. That advantage is harder to copy than the technology.
“The firms best positioned to define what AI in financial services becomes are not those chasing the most advanced capabilities. They are those that have earned — through execution quality and attentiveness to how customers actually feel — the trust required to take the next step.”
— Corporate Insight, AI in Financial Services Today
Appendix
Glossary, sources & methodology
Key terms
- Deterministic VA
- A rule-based assistant with pre-written responses. Interprets intent but never generates novel text — eliminating hallucination risk.
- Generative AI (GenAI)
- LLM-powered tools that create dynamic, personalized responses rather than returning scripted answers.
- Agentic AI
- AI that can take actions on a user's behalf (e.g., execute trades), typically with pre-authorization or a review checkpoint.
- NLP
- Natural language processing — how a tool interprets typos, slang and varied phrasing.
- MCP
- Model Context Protocol — a standard that lets external AI agents (e.g., Claude, ChatGPT) connect to a platform's data and actions.
- RIA
- Registered Investment Advisor — an SEC registration status some AI-advice fintechs (Origin, Magnifi) have secured.
Sources
- AI in Financial Services Today: The Opportunity, the Risk and the Emerging Reality
Corporate Insight, June 2026. Draws on a February 2026 survey of 2,000+ U.S. adults, January 2026 in-depth consumer interviews, and ongoing Monitor competitive intelligence across 21 verticals. - Desktop Virtual Assistants and AI (e-Monitor Industry Overview)
Corporate Insight, August 2026. A benchmark of desktop VAs across nine brokerages plus four top mobile examples, testing a consistent prompt set for comprehension, output, findability and design.
This briefing is an internal synthesis of the two Corporate Insight reports for leadership discussion. Figures and examples are drawn from the source studies; some chart values are shown at report-level precision.