Personal Agents May Spark a Battle to Defend Bank Deposits

Banks have already begun to shore up their defenses ahead of the arrival of personal agents.

In mid-October, the Doubao mobile assistant announced that it would gradually open screen automation operations to third-party apps, and several bank apps were on the list of those it refused to operate.

Almost simultaneously, Citi, Bank of America and JPMorgan Chase also tightened the permissions for external agents to access customer accounts.

Account security is the immediate concern, but how the money in those accounts will flow is a question banks need to consider in advance.

What agents may pry loose is the demand deposits that, during periods of low interest rates, help banks keep their funding costs down;

once an agent obtains the corresponding authorization and can find idle cash for customers, compare returns and complete transfers, balances that once sat in demand deposit accounts because customers had no time to manage them will more easily find a new home.

Depositors may thus earn more, while banks may have to pay more to compete for these funds; and if a large number of agents adopt similar strategies, funds could also be moved in a concentrated way within a shorter period, compressing the time institutions have to raise cash.

This potential battle to defend deposits concerns how cheaply banks can obtain funding, and also how long that funding can stay.

For a long time, the balances customers leave in demand deposit accounts have been an important source of low-cost funding for banks; using these funds to support loans and investments allows them to retain more interest income.

How to keep more balances in accounts has long been an important part of banks' business model.

Customers keep money for payments and emergencies, and they may also leave idle cash in their accounts because finding products and comparing rates is a hassle, making it a steady source of low-cost liabilities for banks.

The rise of personal agents, especially the continuous expansion of AI permissions, is cracking the inertia that commercial banks have maintained for years.

Doubao's mobile assistant, for example, tries to cross the boundaries of developer permissions by recognizing text and buttons on the screen to help users complete clicks, fill in forms and submit tasks that previously required step-by-step human operation;

under this vision, users could query balances, purchase products and even transfer out funds with a single instruction.

The banks' response has been a collective refusal.

When the public comment period for Doubao's screen automation statement and agreement ended, a total of 151 apps had explicitly refused Doubao's operations, including applications from financial institutions such as ICBC, China Merchants Bank and Chongqing Rural Commercial Bank.

Users can still enter the relevant banks through the Doubao phone, but Doubao cannot perform operations within the apps on the user's behalf.

Major U.S. banks have likewise encountered the problem of personal agents extending into financial accounts.

Previously, Meta launched Muse to connect different services, handle emails, arrange travel and shopping for users, and announced an expansion of shopping services and payment connections.

Not long after, Citi banned third-party agents from logging into accounts or trading, retaining only limited read-only access authorized through specific open banking providers; Bank of America studied who bears responsibility for AI payment instructions; and JPMorgan restricted access by external agents.

Read-only access means allowing tools to obtain account information;

if an external assistant acts on the user's behalf, banks need to confirm whether the user has truly authorized it, whether transactions can be blocked in time, and who bears responsibility if something goes wrong.

Banks are building boundaries not only for security, but also because it matters whether agents can turn return comparisons into actual transfers of funds.

Once permissions are breached, agents can continuously compare and transfer funds, moving money into higher-yielding accounts, and banks will inevitably face a more concrete business problem: how much of the cheap demand deposit balance will remain.

Apollo chief economist Torsten Slok suggested that agents could move household cash from demand deposit accounts with an average rate of about 0.1% into accounts yielding 3.3% to 5%, dealing a blow to banks' low-cost deposits.

This is a forward-looking scenario, not an outflow that has already occurred.

Brian Moynihan, chairman and CEO of Bank of America, stressed that checking accounts handle daily income and expenses, while cash for investment purposes is already served by yield products.

Moynihan believes that if chasing higher returns affects bill payments or even incurs overdraft fees, it cannot be called optimization. He is not worried that agents will change this arrangement of funds.

The disagreement thus comes down to what account balances are used for.

How much is working capital that must stay, and how much is simply idle cash that has not yet been managed?

What agents can pry loose is the latter, plus the balances released by more precise management of income and expenses.

Balances begin to move

More than a decade ago, Yu'e Bao showed banks how technology could change where these balances go.

Yu'e Bao was initially linked to the Tianhong Zenglibao money market fund; when users transferred in spare change, they were actually buying fund shares, and the fund invested in bank deposits, short-term bonds and other assets to earn returns, with relatively quick redemption.

In 2012, the average yield of Class A money market funds was 3.6 percentage points higher than that of demand deposits, but the number of holders was less than 3 million; the following year, Alipay connected money market funds to Yu'e Bao, bringing spare-change wealth management into daily consumption, and by the end of that year Yu'e Bao's customers had soared to 43.03 million households, with scale reaching 185.3 billion yuan.

Liabilities that leave banks' retail accounts will come back, but at a more expensive price.

At the end of 2013, bank deposits and settlement reserves accounted for 92.21% of Tianhong Zenglibao's assets; low-interest demand deposits had become concentrated fund allocations that banks had to compete for at higher rates; moreover, after fund managers freely chose where to allocate, the funds would not necessarily flow to the account-opening bank, and every bank faced a new round of liability migration.

To cope with platforms' competition for customers and funds, banks have learned to put money market funds and convenient subscriptions and redemptions into their own channels, sacrificing part of their demand deposit scale in exchange for the chance to keep participating in follow-up services.

The lesson competition has taught banks is that the more efficiently customers manage their idle money, the harder it is for low-cost balances to accumulate.

But technological evolution will not stop because of this.

Putting products into accessible, inclusive everyday wallets was the 1.0 era of mass cash management led by Yu'e Bao;

while the 2.0 era that personal Agents may open involves further intelligent needs analysis, cross-institution comparison, and continuous adjustment of fund arrangements.

One can imagine that in the future, after a salary arrives, the Agent, combining bills and user settings, sets aside enough for living expenses, repayments and emergency funds, then compares where the remaining balance would be better placed;

at first, it may only remind the customer or generate a plan; after obtaining the corresponding authorization and interfaces, it can submit transfer-in and subscription applications; with account openness and transaction verification support, it can further make automatic allocations.

In an ideal scenario, salaries can continue to be paid into a familiar bank, and repayments can also be completed through the original account, while idle balances may be moved elsewhere.

Even if customers do not leave the bank, low-cost funds may continue to flow.

If the Agent can track income and expenditure plans and compare products across institutions, a single transfer-in is merely the starting point for the next round of choices;

changes in returns, arrival times or spending plans may all trigger new suggestions, and whether a switch is worthwhile must also factor in fees, redemption restrictions and risk differences; transferring into a product no longer means staying for the long term.

Nikhil Lele, EY's Americas Banking and Capital Markets Consulting Leader, and Akshata Udiavar, an EY partner, believe that Agents may accelerate deposit repricing and weaken customers' dependence on brands.

Even after a bank wins a transfer-in, it still faces the next comparison and the next choice; balances accumulated through user inertia are bound to become more expensive and harder to predict.

Retaining customers VS retaining money

However, even if today's customers compare returns again, the calculation they make is different from 2013.

According to banks' posted rates and fund disclosure data, the demand deposit rate at major banks such as ICBC is now 0.05%; as of October 9, the 7-day annualized yield of some funds linked to Yu'e Bao was about 0.90% to 0.97%.

Calculated statically at this level, 10,000 yuan would earn about 85 to 92 yuan more a year; in 2013, the average money market fund yield was about 3.57 percentage points higher than the demand deposit rate, corresponding to about 357 yuan.

The narrowing interest spread has weakened customers' incentive to move money, but Agents may reduce the hassle of moving it. Whether they can drive migration still depends on whether the returns can cover operating, service and other costs.

But helplessly, banks at this moment also need to defend cheap funds more than before.

In the second quarter of 2013, the net interest margin of commercial banks was 2.59%; in the second quarter of this year it was 1.41%, 1.18 percentage points lower than that year. As personal demand deposits continue to flow out, banks need more supplementary funds to maintain asset scale.

This cost can be measured with a static scenario.

Taking every 1 trillion yuan of personal demand deposits as an example, assuming 1% flows out and replacement funding costs 1 percentage point more, the annualized additional cost is 100 million yuan; if the outflow ratio rises to 10%, the additional cost is 1 billion yuan.

The more expensive the supplementary liabilities, the greater the cost pressure, but shrinking assets may in turn reduce corresponding interest income.

It is worth noting that the channel rules for cash management products reaching customers are also changing.

The Measures for the Administration of Online Marketing of Financial Products, effective September 30, restrict non-bank payment institutions from listing asset management products as payment tool options and from marketing them.

Yu'e Bao and Lingqian Tong can still be used for payment, but must be displayed separately; banks' proprietary channels are not subject to the above specific provisions targeting non-bank payment institutions.

This may improve banks' relative conditions for competing for cash management customers;

but after the space for non-bank payment institutions to market asset management products through payment scenarios narrows, whether customers turn to banks still depends on product returns and user experience.

Improving customer experience and providing more competitive deposit-taking conditions also require banks to make concessions proactively.

Banks need to weigh whether the benefits of retaining customers can make up for the losses caused by higher liability costs.

JPMorgan Chase's cash management plan provides a concrete example.

In his 2025 annual shareholder letter, JPMorgan Chase Chairman and CEO Jamie Dimon proposed that this year the bank plans to use Smart Cash to let customers automatically allocate funds between higher-yield products in checking accounts and brokerage accounts.

According to Dimon's description, customers can save multiple steps of trading, clearing and transferring; further ahead, AI will also be used to predict cash needs, anticipate bills and assist with budgeting.

The same bank, on one hand, restricts external Agents from accessing accounts, while on the other hand preparing to help customers pursue returns more conveniently;

what banks are competing for is no longer just where balances stay, but also who continuously arranges this money for customers.

The cost of liquidity

For depositors, this is an account of returns; for banks, it is an account of liability costs.

And for financial markets, it is also an account of liquidity.

When a large number of Agents adopt similar strategies, respond to the same market signals and obtain automatic execution authority, transfers or redemptions originally spread across different times may occur in a concentrated manner.

When customers concentrate on withdrawing deposits, banks need to raise money promptly;

when investors concentrate on redeeming funds, funds need to prepare cash, and when cash is insufficient they may need to sell assets;

And when funds withdraw bank deposits, that may also increase the funding pressure on the banks concerned.

For example, if demand deposits flow from Bank A to Bank B, the money is still within the banking system, but Bank A must promptly replace the funds that flowed out; if it lacks sufficient cash on hand, it will have to raise financing at a higher cost or even sell assets that have not yet matured.

Similarly, if multiple funds sell similar assets at the same time, the selling may push down prices and affect the valuations of other holders;

if the price decline further triggers redemptions, the pressure to raise cash may continue to expand.

Fast redemption services for money market funds have previously exposed timing gaps in the flow of funds.

Users demand immediate crediting, but the assets held by the fund may not yet have matured; fast redemption services often require institutions to advance funds first, and once redemptions occur in a concentrated manner, the advancing institution may face pressure to raise a large amount of cash within a short period.

The policy implemented in June 2018 set the daily T+0 fast redemption withdrawal limit for a single money market fund held by a single investor through a single sales channel at 10,000 yuan. This limit applies to fast-crediting services; normal redemptions are not subject to this cap.

Fast redemption services shorten the wait for funds to arrive, and now personal Agents will further shorten the time for decision-making and operations;

as fund movements become more concentrated, the room left for institutions to arrange cash may also shrink accordingly.

The Financial Stability Board has also warned that shared models and data, and automated responses, may amplify the correlation of market behavior and liquidity pressure.

Banks therefore need to reassess how quickly balances managed through Agents may flow out, and arrange cash reserves and backup financing accordingly.

Personal Agents make it easier for each depositor to arrange their own money, and may also make it harder for banks to arrange the long-term use of that money;

what this battle to defend deposits ultimately has to answer is how, as money becomes increasingly easy to move, the financial system can continue to provide stable support for loans and investments that require time. Risk disclosure and disclaimer

Markets carry risk, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial conditions or needs of individual users. Users should consider whether any opinions, views or conclusions in this article are suitable for their particular circumstances. Investment based on this is at the user's own risk.