Proving It

A Promise Earned 3X More Trust When It Went on the Record

In a money game played by 376 students, a seller’s promise did nothing for trust when nobody tracked it. When buyers could see whether each seller had kept past promises, the same promise won their trust nearly three times as often.

Du, N., & Zhao, Q. (2026). Promise-keeping reputations in an investment game: An experimental investigation. International Studies of Economics, 21(2), 92-110. Open access: doi:10.1002/ise3.70016

A seller on a livestream holds a jar of face cream up to the camera and promises it is the real thing. Viewers have a few seconds to decide whether to believe her and buy.

Ninghua Du and Qun Zhao, two economists at Shanghai University of Finance and Economics, wanted to know what makes a promise like hers worth believing. In their experiment, a promise on its own was worth nothing. The same promise, written down where the next buyer could see whether it had been kept, was worth a great deal.

That is the livestream seller’s position, turned into a game. Before the buyer decided, some sellers could send a promise: ‘If you send, I will split.’ It cost nothing to say. Some buyers had no way to check it. Others could look back at each seller’s last five rounds and see every promise the seller had made, and what the seller did next.

What happened

Nearly nine sellers in ten made the promise, but fewer than half of them kept their promise.

A promise nobody could check won no more trust than silence: buyers sent their tokens about as rarely as when sellers said nothing at all.

Then the same promise went on the record. Buyers handed over their tokens three times in four, nearly three times as often as when nobody could check. And more than four in five of the sellers who promised kept their word.

An unchecked promise can cost you

As the rounds went on, a promise nobody could check did worse than silence. After the fifteenth round, buyers who heard unchecked promises trusted less often than buyers who heard no promises at all. A buyer let down after a promise also pulled back further than a buyer let down without one.

The record did most of the work on its own. In a fourth version of the game, sellers could not promise anything, but buyers could see their last five choices. Buyers trusted them about two times in three, not reliably less than when promises went on the record. What buyers wanted was a look at what the seller had actually done.

What This Could Mean at Your Desk

Put your promises where the client can check them. ‘I’ll call you when the rebalancing is done’ is an advisor’s version of the livestream promise. Written into a follow-up note with a date, and checked off at the next meeting, it becomes a record. In the experiment, the record is what turned a free sentence into one buyers acted on.

Show the record before you ask for more trust. The buyers could see a seller’s last five rounds. A client sees whatever you show them. A short list at the top of each review, what you said you would do last time and what happened, gives the client the same view.

A promise nobody tracks could cost you. Unchecked promises pulled trust down over time, and a letdown after a promise hurt more than a letdown alone. If you won’t track a commitment, it may be better not to make it.

The bottom line

A promise nobody could check did nothing for trust, and over time it did harm. Once buyers could see whether a seller had kept past promises, the same sentence won their trust three times in four, and most sellers who made it kept it.

People Trusted Police Software More When Told Someone Was Checking It

877 Dutch adults imagined calling the police about extra patrols on their street and hearing that a computer program had sent them. Those told an independent agency checks how the police use the program trusted the police more. Those told which law governs it did not reliably trust them more.

Nieuwenhuizen, E. N., Trehan, V., & Porumbescu, G. A. (2026). Does institutional transparency affect citizen trust in predictive policing? Evidence from a survey-experiment in the Netherlands. Public Administration, 104(3), 489-508. Open access: doi:10.1111/padm.70034

The police rarely come down your street. This week a patrol car has driven past every day, and today two officers stopped your new neighbor, who recently moved to the Netherlands, to ask what he was doing there. There have been no burglaries in months. You call the station to ask why.

Esther Nieuwenhuizen, Vishal Trehan and Gregory Porumbescu asked 877 Dutch adults to imagine that call. They wanted to know what makes people trust a police computer program: being told which law governs it, or being told that someone outside the police checks how it is used.

Then the answers split. Asked whether this was legal, some people heard that the Police Data Act applies to the program and sets out when and for which tasks personal information may be used. Asked whether anyone was monitoring the police, some heard that an independent government agency, the Data Protection Authority, watches how they use the program and the personal information. Some heard both. Some heard that the police had not said.

Everyone then rated how far they trusted the police to use the program competently, with good intentions, and honestly.

What they found

The law did not reliably move trust. People told about the Police Data Act rated the police a little higher, but on all three measures the difference was small enough to be chance.

The watchdog did. People told that the Data Protection Authority was checking trusted the police more on competence, on good intentions and on honesty. Hearing about the law as well added nothing reliable on top. The effect of one sentence on the phone was modest, but it was the only sentence that moved trust reliably.

Why the law fell short

The researchers’ reading is that people may not see a law as a safeguard on its own. A rule can exist on paper with no one enforcing it. An agency checking the police is evidence that someone would notice if the rule were broken. The Netherlands had recently lived through a scandal over a government algorithm used for childcare benefits, which the authors say has made calls for that kind of accountability louder.

What This Could Mean at Your Desk

Name who checks, not just what the rules are. When a client asks who makes sure their money is safe, ‘we follow the regulations’ is the Police Data Act answer. Naming who checks is closer to the answer that moved trust in the experiment: the custodian that holds the assets and sends its own statements, the auditor, the regulator that examines the firm.

Point to the check the client can see. A statement that arrives from the custodian, not from you, is a watchdog the client can hold in their hand. It may be worth saying so out loud the first time one arrives.

Keep it to one sentence. The researchers kept each answer to a sentence or two, because longer versions they tested earlier left people remembering less. A client asking who watches their money may need one clear sentence more than a copy of your Form ADV.

The bottom line

Dutch adults told that an independent agency monitors how police use a crime-prediction program trusted the police more on competence, good intentions and honesty. Being told which law governs the program did not reliably change their trust, alone or alongside the agency.

Trust in New Software Rose and Fell With How Well It Worked

Employees at four German companies rated their trust in new business software from before launch to eight months after. Trust tracked how well each system worked: it climbed where the software ran well and fell where it failed. The support around it added nothing once people had judged the software itself.

Müller, L. S., Nohe, C., Reiners, S., Becker, J., & Hertel, G. (2026). Building trust in workplace information systems: A four-company study. Behaviour & Information Technology, 45(3), 388-408. Open access: doi:10.1080/0144929X.2025.2518236

In August 2022, an IT services company with 50 employees switched to a new customer-relationship system that it had picked and set up without outside help. A few key users had tried it about four weeks before launch. Nobody had tested it under a full day’s load. Then, parts of it failed.

Eight months later, its employees trusted the system less than they had before it arrived. Lea Müller and four colleagues in Germany followed that company and three others through the same kind of change, and found that trust followed how well the software worked.

What they found

At the two traders, which shared a system and trained their users before launch, trust rose over the eight months. At the importer, it held steady. At the IT firm, trust fell a full point on a seven-point scale in the first month, and kept falling.

Those employees had trusted the system before they had used it. The IT firm’s drop in the first month was the largest in the study, which the authors read as disappointed expectations.

What separated the companies was how people rated the software itself. At every survey after launch, the more reliable, credible, and easy to use people found it, the more they trusted it. Once that was taken into account, the support and training, how involved employees had been, and their own trust in technology added nothing reliable. The authors called the result for support and involvement surprising.

Trust also tracked how people felt about their work. Employees who trusted the system more reported doing their jobs better and feeling less strain.

What the companies did differently

The researchers interviewed the project manager at each company afterward. Support made no difference once people had judged the software. But the project managers at the companies whose software worked credited the training and consultants with getting it right. The wood and seed traders trained their users before launch. One of their project managers described the new system as something ‘we have to form together ourselves.’ The importer called its on-site consultants ‘worth its weight in gold.’

Looking back, the IT firm named choosing the system without outside help as one of its key failures. Most employees never got to try it before launch, the software was bent to fit old processes at the cost of features, and on launch day it failed under full load.

What This Could Mean at Your Desk

This study is about software, not advisors. But clients meet your firm more and more through systems: the portal, the performance report, the onboarding forms, the transfer paperwork. The study suggests that trust in those follows how they perform.

Test the client’s experience before launch. The company whose trust collapsed let a few insiders try its system and never tested it under load. Before a new portal, custodian or reporting package reaches clients, run real client tasks through it: logging in, finding a statement, moving money.

Service didn’t add trust on top of performance. Support and training added nothing reliable once employees had judged the system. A warm call about a late or wrong report may do less than a report that arrives on time and right.

Set expectations before launch day. Employees trusted the new software before they had used it, and the biggest early drop came where the authors saw disappointed expectations. When a client moves to a new platform, tell them in advance what may be rough in the first month.

The bottom line

At four German companies, employees’ trust in new software tracked how well the system worked, climbing where the software ran well and falling where it failed. The support around the system and people’s own trusting nature made no reliable difference once the system’s performance was taken into account.