The typical company that trails gives up millions of dollars a year. In our estimate, a treasury agent could earn much of it back.

The short version

  • In the first half of 2026, 64% of the US public companies in our dataset earned less on their cash than 3-month T-bills. The median one that trailed gave up about $2.9 million a year. In 2025, across over 700 companies, it was 69% and $3.6 million.
  • The gap widens when rates rise. It peaked in 2023, right after rates rose fastest, and the Fed raised rates again in September 2026. If increases continue, companies that still approve every cash move by hand will give up more.
  • Treasury agents can close the gap, but only as far as their owners let them act. In our worked example, an agent acting under a cap adds about $3.3 million a year, and one acting within its full policy about $6.1 million.
  • Owners let an agent act alone once they have evidence it stays inside its policy. Building that evidence takes months, so teams that start before rates climb further capture more of the gap.
  • Companies that held half or more of their cash in securities earned a median 4.03% in 2025, against 3.30% for those with none.

Every company holds cash it does not need this week. Some of it sits in operating accounts that pay well below market rates. Moving it into Treasury bills or money market funds is routine work. Someone has to do it every day, and in most companies a person approves each move.

Treasury agents can do this work, but most of them only recommend. Kyriba describes its agent this way: "The agent recommends; you decide." We wanted to know what companies leave behind today and what an agent could add as it earns more authority. So we built a dataset from SEC filings.

We measured the yield on cash at over 700 companies

For each company, we divided 2025 interest income by its average cash, short-term investments and debt securities across the year. We compared the result with the average 3-month T-bill rate for 2025.

Dataset

US public companies that reported interest income in their 2025 XBRL filings, with at least $100 million in revenue and $50 million in cash and investments. We excluded banks, insurers, brokers, REITs and funds. We also dropped yields outside 0.5–8%, which almost always come from inconsistent tagging. We call a company cash-rich when cash and investments make up at least 25% of its total assets.

Seven in ten companies earned less than T-bills

The median company earned 0.62 points less than T-bills. 54% trailed by more than half a point and 36% by more than a full point.

The gap has narrowed in 2026 but not closed. 469 of the companies reported quarterly interest income for the first half of 2026. 64% of them earned less than T-bills, which averaged 3.61% over those six months, and the median one that trailed gave up about $2.9 million a year. Their median gap fell from 0.53 points in 2025 to 0.30 points, as T-bill rates came down.

The gap widens when rates rise

We followed the same 590 companies from 2022 to 2025, and added 2021 and the first half of 2026. The gap was widest in 2023, when rates had just risen fastest, and narrowed as deposit rates caught up and then as rates fell.

T-bills averaged about the same in 2023 and 2024, 5.07% and 4.97%, yet the gap nearly halved. Deposit rates follow market rates with a lag, so the gap opens while rates climb and closes once they stop. Take the median cash-rich mid-market balance of $637 million. The gap on it was worth about $8.6 million in 2023, and $2.2 million a year in the first half of 2026.

Rates are rising again. The Federal Reserve raised its target range to 3.75–4% on September 16, its first increase in over three years, and said inflation "remains elevated". In the minutes released on October 7, officials agreed that inflation was still elevated, and several said rates were too low to restrain the economy. Markets expect no change at the October meeting. If increases continue, the gap should widen again. An agent that already moves cash daily, under limits its owners have approved, would capture it as it opens. A team that starts building that evidence when rates move would spend the first months of the cycle in a shadow pilot.

Companies that put more cash to work earn more

We split the companies by how much of their cash and investments they held in securities, beyond cash and cash equivalents. The median yield rises at each step.

Company size also matters. Companies with more than $10 billion in revenue, which run large treasury teams, trailed T-bills by a median 0.32 points. Mid-market companies, with $500 million to $10 billion in revenue, trailed by 0.73 points, and 72% of them earned less than T-bills.

SegmentCompaniesEarned less than T-billsMedian gapMedian yield given up, among those that trailed
Large (revenue $10B+)14157%0.32 pts$28.7M a year
Mid-market ($500M–$10B)43072%0.73 pts$3.5M a year
Small ($100M–$500M)13471%0.51 pts$1.0M a year
Cash-rich mid-market11669%0.39 pts$5.3M a year

Cash-rich companies have the most to gain from matching their best peers

Among the 116 cash-rich mid-market companies, the top quarter earned 4.21% or more on their cash and the median earned 3.67%. For the median company below that top quarter, matching it would be worth $6.3 million a year. Software and IT services (37 companies) and pharmaceuticals (13) are the largest groups in this segment.

The gap persists in 2026. Of the 77 cash-rich mid-market companies that reported interest income for the first two quarters, 70% still earned less than T-bills. The median one that trailed gave up about $4.0 million a year, annualized.

Payables show the same pattern

Two accounts-payable surveys grouped companies by how automated their invoice processing is. In both, the most automated group captured the largest share of available early-payment discounts.

Ardent Partners' 2025 survey shows the same pattern for approvals. Its best-in-class teams process 51% of invoices with no manual step, at $2.65 per invoice. All other teams process 29% that way, at $12.42 per invoice. A 2% discount for paying in 10 days instead of 30 is worth about 37% a year, so each missed discount is expensive.

What a treasury agent could add at each level of autonomy

A treasury agent can only close as much of these gaps as it is allowed to act on. We describe three levels of autonomy here and a fourth, onchain, in the next section. Treasury owners move an agent up one level at a time, on evidence from the level below.

Most of the value in this example comes from cash. For companies with fewer surplus balances and more supplier spend, payables would carry more of it.

Level 4: an agent that rebalanced onchain earned more

Corporate cash is starting to move onchain. Tokenized Treasury funds such as BlackRock's BUIDL, Ondo's OUSG, Superstate's USTB and Circle's USYC hold T-bills and repo, settle at any hour and charge 0.15–0.20% a year. Onchain lending markets such as Aave, Compound and Sky pay variable rates set by borrowers' demand.

Level 4 is an agent that manages cash across these venues within limits its owners set. We backtested that idea with daily data from 2023 through 2025.

Three results stand out. Tokenized Treasury funds track T-bills less their fees, so they do not beat T-bills; they bring T-bill yields to cash that would otherwise sit in deposits. Lending rates swing with demand: they paid more than T-bills in 2024 and 2025 and less in 2023. A fixed allocation to any one venue would have had a bad year.

The simulated agent followed a simple policy. Each day it put up to 25% into each lending venue that paid more than the tokenized fund, and kept the rest in the fund. It averaged 5.57% over the three years, one point more than the tokenized fund. In 2025, each $100 million of idle cash would have earned $4.65 million with the agent and $3.91 million in a tokenized fund. The median company that held no securities earned $3.30 million.

2026 shows the other side. With rates lower, Aave and Compound have paid less than T-bills so far this year. The agent kept most of its cash in the tokenized fund and earned 3.73%, about 0.2 points more than the fund. A fixed allocation to either lending market would have earned less than the fund. In weak years for lending, 2023 and 2026, the agent's rules kept it ahead of the fund.

Risks the backtest leaves out

USDC and DAI closed near $0.96 on March 12, 2023, after Silicon Valley Bank failed, so the agent's lending positions would have shown a loss for several days. Every lending venue carries smart-contract, oracle and governance risk. Large deposits lower the rate a venue pays, and the backtest assumes they do not. BUIDL and OUSG are open only to qualified purchasers, and USYC only to non-US investors. Most corporate investment policies do not yet allow onchain lending.

Level 4 is a policy decision before it is a technology decision. An agent that moves cash between onchain venues at any hour needs a limit for each venue and a rule for when a stablecoin loses its peg. It also needs a record of every move. Those are the controls treasury owners will ask for as digital assets reach corporate treasury.

Limits of this analysis

  • The patterns compare different companies. Holding more securities goes with higher yields, but other differences between those companies could explain part of it.
  • We also tracked companies from 2024 to 2025. Those that raised their share of securities closed less of their gap than others, so the year-to-year data does not yet confirm the pattern.
  • Some cash cannot be swept. Companies need operating balances for payroll and payments.
  • Cash held abroad earns local rates, which were lower than US rates in 2025. Companies also tag interest income in different ways, so one company's figure can be off; medians across hundreds of companies are more reliable.
  • No published study measures return by level of agent autonomy. The shares closed at levels 2 and 3 are our assumptions, and we intend to measure them with our design partners.
  • The level 4 agent is a simulation on historical rates. It leaves out the risks listed in that section, and past onchain rates are a weak guide to future ones.

AGIS produces the evidence to move an agent up a level

Treasury owners give an agent more authority when they have evidence that it stays inside its policy. That holds at every level, and most of all onchain. AGIS, our security product for agents that move money, produces that evidence. A read-only assessment maps the paths to a loss that the agent's permissions allow. An eight-week shadow pilot then checks the agent's recommendations against the policy before it acts alone.

If your treasury team holds large cash balances and still approves every move, we would like to talk.

Learn more about AGIS

Method and sources