Do online customers need to trust the app or the company behind it? What an Electronic Markets study found
Anyone using a shopping site or a banking app is weighing up two things: the technology on the screen and the firm that runs it. A meta-analysis across online retail, banking and e-health finds the two closely linked, both related to use, and fed by partly different drivers.
Kuen et al. pool 272 samples on online trust: faith in an app and in the firm behind it go together, track intention to use, and rise with reputation.
Marketing in 3 · 2023 in research · No. 50 · Paper published 5 December 2023
A handshake, the oldest sign of trust between a customer and a provider. Photo: Erika Fletcher on Unsplash.
The short answer
Both, and the two go together. Pooling 272 samples with 105,199 respondents in online shopping, banking and health, the researchers found that trust in the website or app and trust in the company each had a strong link with whether people intended to use the service. When both were modelled side by side, the technology carried slightly more weight. Reputation was the strongest correlate of either kind of trust, while familiarity mattered more for the system than for the firm.
Evidence level
★★★One study
★★★Several studies
★★★Meta-analysis
Key takeaways
Trust in the technology and trust in the provider were strongly related, at a correlation of 0.59 across 25 samples, where 1 would mean they always move together.
Each kind of trust was linked to intention to use at about the same level: 0.50 across 128 samples for the technology and 0.50 across 105 for the provider.
Familiarity from past use went with trust in the system more than with trust in the company: 0.52 against 0.36.
In banking, the sense that safeguards such as guarantees and regulation are in place correlated 0.69 with trust in the provider, against 0.24 in online shopping (5 and 18 samples).
What the researchers did
Leonie Kuen, Daniel Westmattelmann, Maike Bruckes and Gerhard Schewe,
all at the Chair of Organization, HR and Innovation of the University of Muenster in Germany, took up a basic
question for any business that sells or serves online. When customers decide whether to use a website or an app, whose trustworthiness are they judging? It could
be the technology in front of them, or it could be the company or person behind it.
Earlier research gave no settled answer. Some studies measured trust in the technology only, others trust in the
provider, a few measured both, and some blended the two into a single scale. Results disagreed too: a link that was
significant in one study was absent in the next.
The team kept the two apart. Trust in the technology meant believing that the site, app or recommendation tool
works, is reliable and helps. Trust in the provider meant believing that the organisation offering the service is
able, means well and is honest. They then looked at four established sources of trust: propensity to trust, a
person's general readiness to trust others; institution-based trust, the sense that the setting is normal and protected by safeguards
such as third-party seals, guarantees, laws, privacy statements and encryption; familiarity from earlier use; and
reputation, what a person has heard from others.
Evidence base
272 independent samples from 251 articles; 105,199 respondents in more than 50 countries
Settings
Consumer-facing online shopping (171 samples), online and mobile banking (65) and e-health (35)
Which trust was measured
Technology only in 126 samples, provider only in 106, both in 25, a blended scale in 15
Sources
Databases including EBSCOhost, Web of Science and IEEE Xplore, Google Scholar, 12 journals searched by hand and reference lists; 246 samples from journal articles, the rest from conference papers, dissertations and edited volumes; all data collected before 2020
Method
657 effect sizes coded by two coders (91% agreement), pooled under a random-effects model; meta-analytic structural equation models compared four ways to place trust in a model
How to read the figures
The first half of the paper pools correlations, written r. At 0 two measures are unrelated; at 1 they move up and
down together without exception. The authors use a widely cited convention: above 0.10 is small, above 0.30
moderate and above 0.50 large. A correlation shows that two beliefs go together, not that either of them causes the other.
The second half fits path models to those pooled correlations. These report standardised coefficients, written β.
Each shows how strongly one measure is tied to another after the other measures in the model are taken into
account. That is why the same pair can show a smaller number in a path model than as a plain correlation.
Study results differed widely. In almost all of the pooled figures, more than 75% of the spread reflected real
differences between settings rather than chance, so any single business can sit well above or below the average.
Separate beliefs that move together
The pooled correlation between the two was large, with 95% confidence limits of 0.52 and 0.65: high, but short of
the level at which two scales would be measuring the same belief. The technology and the provider, the
authors conclude, are separate targets of trust that depend on each other.
Experience made a difference here. Among respondents with earlier use of a service, the correlation reached 0.64
(13 samples), against 0.44 (5 samples) for first or near-first contact. The setting, whether shopping, banking or
health, made no significant difference. Across studies the spread was wide: in a new setting, the true
correlation could plausibly lie anywhere between 0.15 and 0.83.
The authors explain the link with trust transfer theory. Online, there is seldom any personal contact with the
provider. The site or app is where the customer meets the business, so a system that works well signals a
dependable firm behind it. Pooled correlations fit this account but cannot confirm the direction.
What goes with each kind of trust
All four sources were positively and significantly linked to each kind of trust. Reputation came first for each,
at about 0.60. Propensity to trust came last (0.36 and 0.31 for technology and provider respectively).
Institution-based trust fell between them, at 0.49 for both.
Familiarity was the one source that clearly separated the two. Past use was a stronger correlate of trust in the
system than of trust in the firm. In the authors' explanation, even a well-known company is harder to predict than
a piece of software, so knowing it well adds less, and trust in a provider leans more on outside cues such as
reputation and safeguards.
Studies that used one blended scale for both kinds of trust found a correlation of 0.44 with intention to use,
across 10 samples. Checks for missing studies left every pooled figure positive and significant, and the results held up
when samples were removed one at a time.
How closely each source went with trust (pooled correlation, 0 = no link)
How closely each source went with trust (pooled correlation, 0 = no link)
Group
Value ()
Reputation, technology
0.6
Reputation, provider
0.604
Institution-based, technology
0.493
Institution-based, provider
0.492
Familiarity, technology
0.521
Familiarity, provider
0.356
Readiness to trust, technology
0.363
Readiness to trust, provider
0.309
Pairwise random-effects meta-analyses (Table 2): pooled correlations from 12 to 35 samples per bar; all significant at p < 0.001. Readiness to trust is the paper's propensity to trust.
Where safeguards counted most
The team tested whether the setting or the customer's experience changed these links. Many subgroups held only a
handful of samples, so the authors advise caution with this part.
Neither the setting nor experience significantly changed how either kind of trust related to intention to use. A
new user and a regular one, in shopping, banking or health, showed much the same pattern.
Safeguards were another matter. With technology trust, institution-based trust correlated 0.57 in banking
(16 samples), 0.51 in e-health (4) and 0.40 in online shopping (15). For provider trust the spread was
larger still, with e-health at 0.36 (2 samples) between banking and shopping. The authors regard the result as
preliminary and relate it to what is at stake: money handled online can be lost, so people look for structural
protection before they trust a bank or its app.
When everything is modelled together
The second analysis tested four ways of placing trust in a research model: one blended trust, trust in the
technology alone, trust in the provider alone, and both as separate but linked beliefs. All four fitted the data
well. The model with both, built on 263 samples, fitted slightly better than the rest (CFI 0.998, SRMR 0.031).
Three results stand out. Propensity to trust had no significant path in any configuration.
With only one kind of trust in the model, each had the same path to intention to use. With both included, the
technology carried a little more weight than the provider (see the chart).
The drivers differed too. In the single-entity models, safeguards and reputation had larger paths to trust in the
provider (0.28 and 0.50) than to trust in the technology (0.21 and 0.44), while familiarity had the larger path to
the technology (0.35 against 0.18). In the joint model, safeguards and reputation stayed significant for each,
but familiarity no longer predicted trust in the provider.
Path from each kind of trust to intention to use (standardised β)
Path from each kind of trust to intention to use (standardised β)
Group
Value ()
Technology, alone in the model
0.54
Provider, alone in the model
0.54
Technology, both in the model
0.36
Provider, both in the model
0.31
Meta-analytic structural equation models (Figure 2): isolated configurations II and III, and integrated configuration IV; every path significant at p < 0.001.
Why the transfer path lost significance
The biggest surprise came from that joint model. The direct path from technology trust to
provider trust fell to β = 0.088 and was not significant (p = 0.626), even though the two correlate strongly.
The authors suggest that the other drivers in the model absorb the variation in trust in the provider: reputation
and safeguards matter more for it than the technology does. An earlier study by
Maike Bruckes and colleagues on robo-advisers in the financial industry saw the same pattern: once structural assurance entered
the model, the transfer between trust in the bank and trust in the technology lost significance.
They warn researchers against reading a non-significant path as no relationship, and recommend modelling both
kinds of trust as separate, linked constructs rather than relying on one of them alone.
The authors' advice
For firms, the message is to look after both kinds of trust and neglect neither. A website or app represents the
company, the authors argue: an online banking system that works reliably can raise how trustworthy the bank
seems, and one that fails can damage it.
On the drivers, they recommend building the reputation of the organisation and of its technology, partly by
encouraging customers to review both the system and the firm's conduct. To build familiarity, they advise against
regular major changes to the interface and suggest that a new system should work like the established ones in its
field. Safeguards such as legal compliance and guarantees should fit the risks of the interaction, most of all where
losses can be serious.
They also look ahead to AI chatbots and large language models. When users cannot judge how a system works, they
may trust it because they trust the company behind it, and guarantees or regulation may weigh more. That is a
forecast, not a finding: the pooled studies collected their data before 2020, mostly on websites and apps driven
by menus and buttons.
Marketing in 3 steps
Ready. Set. Market.
1
ReadyDecide and prepare
Measure both kinds of trust. Put six questions into your regular customer survey. Ask three about your website or app: does it work properly, is it reliable, does it help? Ask three about your company: is it capable, does it act in the customer's interest, is it honest? Use the same one-to-seven answer scale for all six, and keep the two sets apart.
Put your safeguards in view. List every protection you offer: money-back guarantee, secure payment, data policy, trade seals, complaint route. Check that each one is one click away from the sign-up and payment pages. Write each in plain words, not legal text.
2
SetBuild it
Make the system dependable. Pull a weekly count of failed logins, error pages and payments that did not go through. Fix the three most frequent problems first. Every breakdown a customer sees also counts against the company.
Keep the screens familiar. Bundle design changes into no more than two releases a year. Leave the main menu, search box and checkout button where regular users expect them. When you launch something new, follow the layout of the best-known services in your field.
3
MarketLaunch and measure
Collect reviews of both. After each purchase or session, send a two-question review request: one rating for the app or site, one for the company. Show recent reviews on your landing page and reply to every negative one within two working days.
Track trust next to use. Every quarter, rerun the six survey questions and set the two scores beside sign-ups, active users and repeat purchases. If either score falls two quarters in a row, find out why before the next release.
Found this useful?
Marketing in 3 is free and ad-free, with new issues every Monday and Thursday. A coffee keeps it that way.
Work on the app shows up in views of the firm. On the authors' reasoning, a smoother, more reliable system should also lift how dependable the company seems, though the pooled studies cannot prove which comes first.
Reputation is the main lever. Of the four sources tested, what customers hear from others showed the largest association with trust in the platform and in the business, so reviews and word of mouth deserve more effort than general reassurance.
A trusting nature is not a target. How trusting customers are by nature had no unique link with trust once other sources were counted, so, on our inference, segmenting audiences by it would add little.
Your own figures may differ. For provider trust and safeguards, the plausible range for a new setting ran from slightly below zero to about 0.80, so test what holds for your customers.
Where it does not work
Associations, not causes. The pooled studies reported correlations between measured beliefs, and the path models assume directions rather than prove them.
Intention, not behaviour. The outcome was people's stated intention to use a service, not recorded use.
Before 2020 and before generative AI. Data came from before the pandemic and mostly covered sites and apps with menus, not chatbots, voice assistants or immersive tools.
Four sources of trust only. Information quality, privacy and perceived risk, all known to matter, were left out.
Small subgroups. Many comparisons by setting or experience rested on fewer than ten samples, some on one or two.
Two analyses, two answers. Technology trust and provider trust correlated strongly, yet the direct path between them was not significant once both entered one model. Why remains to be tested.
Questions marketers ask
What builds trust in a website or app?
Reputation, above all. Across the pooled studies, what people had heard about a service showed the closest association with how far people trusted the technology and the company. Protective structures like guarantees and legal rules came next, familiarity mattered mainly for trust in the system, and an individual's general tendency to trust mattered least (Kuen et al., 2023).
Is trust in a website the same as trust in the company behind it?
No, though the two are closely related. They correlate strongly, yet not so strongly that they count as one belief, and different drivers feed each. People trust a site when they believe it functions well, can be relied on and is useful; they trust a company when they think it is capable, well-meaning and honest.
Does a reliable app make people trust the company more?
Possibly. Because online customers rarely deal with the provider directly, the system acts as its stand-in, and a smooth experience can signal a dependable firm. The pooled correlations fit this view, but with reputation and safeguards also in the model, the link running from confidence in the system to confidence in the company was no longer significant.
Does trust matter less once customers know a service well?
Not for their willingness to use it. The tie between trust and intention to use was much the same for new and experienced users, and in shopping, banking and health. Experience did make confidence in the app and confidence in the firm more closely related.
Do these findings apply to AI chatbots?
Only with caution. The studies collected data before 2020 and mostly covered websites and apps with menus and buttons. In the authors' view, complex AI tools may reverse the usual flow, so that trust passes from the company to the technology, and guarantees and regulation may weigh more.
Companion apps that offer an AI friend to talk to now run at scale, and no one can predict exactly where their conversations will go. Field data from two companion AIs showed mental health crises in a small but real share of user conversations, and a test of several commercial apps found they often failed to recognise distress or responded poorly. In an experiment, people reacted badly to risky or unhelpful chatbot replies, a reputational risk for the firms behind them. A brand that puts a chatbot in front of customers should decide in advance how it will deal with conversations the bot was never built for.
Journal of the Academy of Marketing Science · 2024
AI lets services fit each customer closely, but it can also treat vulnerable consumers unfairly or shut them out of the market, often without the firm realising. These are people who lack access to resources or control over them, for a while or for good. Drawing on work on consumer vulnerability and AI for social good, the authors propose the AID framework: AI that makes services more accessible, adapts customer journeys interactively and improves decisions over time. Before an AI service goes live, it helps to ask which customers it could leave behind.
Source
Kuen, L., Westmattelmann, D., Bruckes, M., & Schewe, G. (2023). Who earns trust in online environments? A meta-analysis of trust in technology and trust in provider for technology acceptance. Electronic Markets, 33(1), 61. https://doi.org/10.1007/s12525-023-00672-1
This brief is written for practitioners from the published article. Open access (CC BY 4.0): free to read at the publisher. Numbers are as reported by the authors; read the article for full methods.
Fang et al. tested keep-it partial refunds on 853,521 online shoppers: the first return cost more, but later spending left a net gain of $1.44 per customer.
People who read bad reviews before listening paid $17.29 to see a show, against $20.45 for people who read none. The review changes the experience, not just the purchase.
Bruno Schivinski is Associate Professor of Marketing at Gdańsk University of Technology. He studies how consumers engage with brands in digital and social media.