State of the Connected Customer Report: Key Findings

Ask a bank customer in Riyadh or a retail shopper in Dubai what "good service" means, and you will not get an answer about friendliness. You will get an answer about speed, relevance and whether the company already knows what they want. That shift — from being served to being understood before you ask — is the entire story behind Salesforce's long-running research into how customers relate to brands, and it is why the newest edition, reframed around artificial intelligence, deserves more than a skim.

The State of the Connected Customer report, now in its seventh edition and published by Salesforce Research under the banner of the State of the AI Connected Customer, is one of the industry's few recurring studies of how expectations move over time. Its throughline this year is not a new metric or a shinier dashboard. It is a trust problem: customers want the convenience AI promises, but they have not yet decided whether they trust the machinery that delivers it. That tension, not any single statistic, is the finding every CX leader should sit with.

What is the State of the Connected Customer report?

It is Salesforce Research's periodic study tracking how business and consumer attitudes toward brands, data and service evolve — now in its seventh edition and reframed to foreground artificial intelligence. Salesforce has not published the full dataset of every prior wave publicly in a form that allows precise year-on-year comparison on every metric, so this article does not attempt to restate numbers that belong in Salesforce's own release. What it does do is read the report's stated focus — AI's growing presence in the customer relationship — through the lens Renascence uses with clients across the region: behavioural economics applied to customer experience design.

That distinction matters. A lot of commentary on research reports simply repeats headline percentages without asking why those numbers exist. The more useful question, and the one this article answers, is what psychological mechanism produces the pattern the report describes — because the mechanism is what you can actually design against.

Why does "connected" now mean something different than it did five years ago?

For most of the last decade, "connected customer" meant omnichannel: a customer who moved between app, call centre and branch and expected the brand to keep up. That bar has been cleared by enough competitors that it no longer differentiates. The newer, harder bar is anticipatory: the customer expects the brand to already know their job-to-be-done before they state it, because every other app on their phone — from food delivery to streaming — behaves that way.

This is a textbook case of a reference point shifting. In behavioural economics, people do not judge a service in absolute terms; they judge it against whatever baseline their recent experience has set. Once a handful of AI-native products train customers to expect prediction rather than reaction, every laggard brand is judged against that new baseline — not against its own industry's historical norm. A bank is no longer compared only to other banks. It is compared to the last genuinely effortless digital experience the customer had, full stop.

Why do customers want AI personalization but still hesitate to trust it?

Because personalization and surveillance are produced by the exact same mechanism, and customers cannot easily separate the two. The same data pipeline that lets a brand pre-fill a form and save someone two minutes is the one that lets it build a profile the customer never consented to in any meaningful sense. Salesforce's decision to rename the study around AI signals that this tension — benefit versus exposure — has become central enough to the connected-customer relationship to rewrite the report's own framing.

Two behavioural concepts explain why the hesitation persists even when the benefit is real:

• Loss aversion. Kahneman and Tversky's well-established finding is that losses loom larger than equivalent gains. A customer weighs the vague, future risk of data misuse more heavily than the immediate, concrete convenience of a personalized recommendation — even when the convenience is the more probable outcome.
• The affect heuristic. People form a fast, emotional judgement about a technology — "AI feels invasive" or "AI feels helpful" — and then reason backward to justify it. That emotional tag, not a calm cost-benefit analysis, is what decides whether a customer reads a personalized offer as thoughtful or creepy.

The practical consequence is that the same AI feature can land as delightful in one brand's hands and alarming in another's, with the underlying technology nearly identical. What differs is the trust capital the brand had already banked before it switched the feature on.

Trust is the currency that makes personalization affordable. Without it, every data request reads as a withdrawal, not a deposit.

What is the behavioural mechanism behind the trust gap, specifically?

It comes down to a mismatch between System 1 and System 2 thinking, the dual-process model popularised by Daniel Kahneman. System 1 is fast, intuitive and emotional; System 2 is slow, deliberate and analytical. When a customer encounters an AI-driven recommendation, System 1 reacts first — a flicker of "how did it know that?" — before System 2 has a chance to evaluate whether the data use was reasonable or disclosed. If that first flicker is unease rather than delight, the brand has already lost the frame, regardless of how compliant or well-intentioned the underlying system actually is.

This is why transparency timing matters more than transparency volume. A long privacy policy read after the fact does nothing for System 1. A short, plain-language explanation offered at the moment of personalization — "we suggested this because you reordered it last month" — gives System 1 a benign story before suspicion has a chance to set the frame. Renascence's behavioural economics work with clients consistently starts here: not with more disclosure, but with better-timed disclosure.

How should CX leaders respond to an AI-trust tension they cannot fully resolve?

You will not eliminate the trust gap. You can design around it. The following sequence is the one we take clients through when an AI or personalization initiative is stalling on adoption despite sound technology:

1. Map where AI touches the journey today, not where it should. Most organisations overestimate how much of the customer journey is actually AI-mediated and underestimate how opaque those moments feel to the customer. A proper journey map that flags every AI-influenced touchpoint is the starting diagnostic.
2. Score each AI touchpoint for both value and exposure. A recommendation engine that saves five minutes but requires no sensitive data is a low-exposure, high-value touchpoint — expand it. A churn-prediction model that triggers an intrusive-feeling retention call is high-exposure, uncertain-value — redesign it before it costs you more trust than it saves in revenue.
3. Give System 1 a short, honest story at the moment of personalization. One sentence of plain-language context, delivered inline, does more for trust than a policy document the customer will never open.
4. Build an opt-out that costs nothing. Friction on the way out of a data-sharing arrangement is sludge, in Richard Thaler's term — and customers who feel trapped in a personalization scheme distrust the whole brand, not just that one feature. A one-tap opt-out, prominently placed, is cheap insurance against resentment.
5. Measure trust separately from satisfaction. CSAT and NPS can hold steady while trust erodes underneath them, because satisfaction captures the last interaction and trust captures the cumulative relationship. Track it as its own line item, not a proxy.
6. Close the loop publicly when something goes wrong. An AI system that makes a visible mistake and gets corrected transparently builds more durable trust than one that never fails because it is never tested in the open. Silence after an error reads as concealment, which is a far worse outcome than the error itself.

Every one of these steps assumes something the original Connected Customer research series has argued since its first edition: the expectation bar is set by the customer's best recent experience, not by your sector's historical average. Leaders who still benchmark against direct competitors are benchmarking against the wrong reference point.

What should CX leaders stop doing in response to this report?

A few reflexive moves make the trust gap worse rather than better. Avoid these:

• Stop treating AI disclosure as a compliance checkbox. A legally sufficient disclosure buried in terms and conditions satisfies regulators and nobody's System 1.
• Stop personalizing with data the customer doesn't remember giving you. If a recommendation feels like it came from nowhere, it will be read as surveillance even when the data source was entirely legitimate and disclosed at signup months earlier.
• Stop launching AI features without a visible human escalation path. Customers tolerate automation far better when they know a person is one step away if the automation gets it wrong — this is loss aversion again, insuring against the worst-case outcome.
• Stop measuring AI success purely on efficiency metrics. Handle time and deflection rate tell you nothing about whether the interaction built or spent trust.

None of this is an argument against AI in the customer journey. It is an argument against deploying it the way most organisations deploy any new efficiency tool: quietly, from the inside out, optimised for cost before it is optimised for trust. The brands Salesforce's research consistently surfaces as ahead of the curve are the ones that treat AI as a visible part of the service relationship, not a backend efficiency play customers are never meant to notice.

What does this mean for how CX teams should be structured?

If trust is now a measurable, manageable variable rather than a vague cultural aspiration, it needs an owner. In most organisations today, nobody holds that brief. Data science owns the model. Marketing owns the message. CX owns the journey. Trust falls through the gaps between the three. A CX governance structure that explicitly assigns accountability for AI-related trust — not just for AI accuracy — is becoming as necessary as the model-risk function already embedded in regulated industries such as banking and finance.

This also changes what "good AI" means operationally. A model can be statistically excellent and experientially corrosive if nobody is responsible for how its output lands emotionally. Organisations serious about this should run a structured CX maturity assessment that specifically probes AI governance and trust measurement, rather than assuming a high digital-maturity score automatically covers it. The two are not the same thing, and conflating them is exactly how a well-built model ends up producing a badly trusted experience.

Where does this leave the metric trio — NPS, CSAT and CES?

Still useful, still incomplete. None of the standard three was built to capture how a customer feels about the mechanism behind a service, only about the outcome of a single interaction. A customer can rate an AI-resolved query a perfect CES score for effort and still feel, in the same survey, slightly uneasy about how the resolution happened. That nuance never reaches the dashboard unless someone designs a question to catch it. Feeding structured, open-text voice of customer data back into journey analysis is currently the most reliable way to catch it, because customers will say in their own words what no closed-ended scale invites them to say.

What is the single shift worth acting on now?

Treat AI-era trust as a design variable, not a hoped-for by-product. It can be mapped to specific touchpoints, scored for exposure, tested for System 1 reaction, and improved with concrete interventions — timed disclosure, cheap opt-outs, visible human escalation, honest error-handling. None of that requires waiting for a perfect regulatory framework or a flawless model. It requires deciding, today, that the feeling a customer has in the half-second after an AI system acts on their behalf is as much a CX deliverable as the accuracy of the system itself.

That is the real signal buried inside Salesforce's seventh edition: the connected customer was never just asking to be known. They were asking to be known by something they could trust not to use that knowledge against them. The brands that treat that as the design brief, rather than the fine print, are the ones who will own the next decade of the relationship.

If your organisation is still treating AI deployment and trust-building as separate workstreams, that is usually the first thing worth fixing. Renascence's digital transformation and behavioural economics teams work with CX leaders across the region to close exactly this gap — mapping where AI touches the journey, scoring the trust exposure at each point, and building the governance to keep both moving together.