When the Service Knows What You Need Before You Ask

One of the long-standing ambitions of service design has always been to make it easier for people to get what they need.

To reduce the number of steps, shorten the time required, avoid asking for information we already have, and deliver services with as little searching, waiting, and effort as possible.

But what if we could go further?

What if we did not wait for people to request the service in the first place?

What if an organization, using the data it already holds about someone, could identify that they have become eligible for financial support before they even look for it? Or predict that their child may need additional educational support before the parents ask for help? Or infer that a change in their circumstances means they may need a service they did not even know existed?

With the growing availability of data, increasing data integration, and advances in artificial intelligence that enable systems to identify patterns and predict needs, such scenarios are no longer far-fetched.

At first glance, this may seem like the ideal form of service delivery.

A service that does not wait for people to figure out what they need, where to go, how to apply, or what they are entitled to.

Instead, it knows them well enough to reach them at the right time.

But perhaps the question we should ask before celebrating this future is:

Do we always want a service to know what we need before we ask for it?

More importantly:

Do we always want it to tell us what it knows?

When Receiving a Recommendation Is Not a Neutral Experience

Imagine an organization that holds enough information about an individual’s income, financial obligations, and social circumstances.

By analyzing this data, an intelligent system determines that the person may be a suitable candidate for a financial support program or an accessible financing product.

From a service design perspective, this seems like an excellent opportunity.

Instead of requiring the beneficiary to search for the service, the organization could reach out and say:

“Based on your data, you may be eligible for this product.”

The information may be accurate.

The service may be useful.

And it may save the person considerable effort.

But what does that person actually hear when they receive the message?

Perhaps they do not hear:

“There is a service that may be suitable for you.”

Instead, they hear:

“We know your financial situation is not good.”

And perhaps they begin to wonder:

How did you know that?

What data did you access?

And what else have you inferred about me?

The service itself may be desirable, while the way the system arrived at the recommendation may make the beneficiary uncomfortable.

This reveals an important distinction between helping someone based on what you know about them and suddenly making them aware of just how much you know about them.

Sometimes the Prediction Is Correct… and Still Should Not Be Stated So Simply

Consider a different example.

A child has not achieved the expected level in one of their school subjects.

The education system has access to their previous academic records, grades, answer patterns, attendance, and perhaps dozens of other indicators that the parents would never see together in one place.

The system analyzes this data and concludes that the child is at risk of falling behind and would benefit from an intensive educational support program.

The model may be accurate.

Early intervention may genuinely help.

But what happens when the parents receive a message saying:

“Your child’s data indicates a need for intensive educational intervention.”

Will they receive it as a recommendation?

Or as a judgment about their child’s abilities?

Will the child eventually begin to see themselves as the struggling student?

And will their teachers and parents begin treating them differently simply because a system predicted a possible future trajectory?

The problem here is not necessarily that the AI made a mistake.

It may be entirely correct.

But some predictions, simply by being communicated to people, can change how they behave toward themselves and toward others.

That means the impact of a prediction can extend far beyond its statistical accuracy.

The question, therefore, is not only:

Can the system detect the problem early?

It is also:

How do we tell people what we have discovered without turning a probability into an identity, or a recommendation into a judgment?

There Is a Difference Between the Service Knowing… and Me Knowing That It Knows

Most of us have become accustomed, to varying degrees, to organizations holding information about us.

We know that banks retain records of our transactions.

That healthcare providers know about our previous visits.

That schools know how our children are performing.

And that government entities hold a significant amount of information related to us.

But there is a major psychological difference between knowing that this data exists and being confronted with evidence that it is being connected, analyzed, and used to infer something you never explicitly disclosed yourself.

For an organization to know my income is one thing.

For it to infer that I am experiencing financial hardship is another.

For a school to know my child’s grades is one thing.

For it to predict that they may struggle academically in the future is another.

For a healthcare provider to know my medical history is one thing.

For it to tell me, before I even ask, that my data suggests a potential health risk is something entirely different.

So perhaps the privacy question surrounding intelligent services is not simply:

Did the person consent to the collection of this data?

There may be another question that becomes even more important:

Does the person expect their data to be used to reach this kind of conclusion?

Agreeing to the collection of data does not necessarily mean agreeing to everything that can be inferred from it.

We May Be Technically Ready Before We Are Humanly Ready

An organization may have the data.

It may be capable of connecting and analyzing it.

The model may be able to make highly accurate predictions.

And the technical infrastructure may be fully prepared to deliver an entirely proactive service.

But technical readiness does not necessarily mean that the beneficiary is ready for the experience itself.

People do not relate to data in the same way systems do.

A system sees variables, relationships, and probabilities.

A person may see themselves within that data.

Their financial situation.

Their health.

Their children.

Their job.

Their family.

Their choices.

And perhaps things they never expected an organization to be able to know about them.

So, as our ability to make inferences expands, we may need to start measuring something we have not traditionally measured very often:

Human readiness for intelligent services.

Does the beneficiary understand how their data is being used?

Do they expect the kinds of conclusions that may be drawn from it?

Are they comfortable receiving proactive recommendations?

Would they prefer to be asked first?

And does their answer change depending on how sensitive the subject is?

Most people may have no problem with a service telling them:

“Your document will expire in 30 days.”

But their reaction may be entirely different if it says:

“Based on your financial data, we believe you need this type of support.”

The technology may use similar logic in both cases.

The human experience does not.

Knowing, Sharing, and Acting

Perhaps, then, we should stop treating proactive service as though it exists at only one level.

There are at least three different things a service can do.

First: Know.

Use data to infer that a potential need exists.

Second: Share.

Present that inference or recommendation to the person.

Third: Act.

Initiate a process, change the course of a service, or take an action on the person’s behalf.

Accepting one of these levels does not necessarily mean we should accept the ones that follow.

It may be useful for a system to know that a student is at risk of falling behind so that a teacher can monitor their progress more closely. But that does not necessarily mean the family should receive a definitive message after the first indicator appears.

It may be appropriate for a system to identify that someone could qualify for a particular form of support and then neutrally inform them that certain services may suit their circumstances, without explicitly revealing the classification the system has assigned to them.

And it may be perfectly acceptable for a service to remind me that a document is about to expire — perhaps even preparing the renewal process in advance — while I may not want it to automatically initiate a financial product simply because it inferred that I need one.

Proactivity is not a question of:

Yes or no?

It is a question of:

To what extent?

Not Everything We Know Needs to Become a Message

In the world of traditional services, one of the major challenges was that we did not know enough about the beneficiary.

With artificial intelligence, we may face the opposite problem.

We may know a great deal.

Perhaps even more than is necessary to deliver the service itself.

And this creates a new responsibility in experience design.

Simply because we can reach a conclusion does not mean we should present it to the beneficiary in the same way.

Some insights can be translated directly into recommendations.

Some require neutral wording.

Some require contextual explanation.

Some may warrant human intervention before they are communicated.

And some may be better used quietly in the background to improve the service without becoming a visible label attached to the person.

The goal is not to demonstrate to the beneficiary how intelligent the system is.

The goal is to use that intelligence in a way that helps them.

The Smartest Service Is Not the One That Tells You Everything It Knows

It is easy to measure the advancement of intelligent services by how much they know about the beneficiary.

The more data we have, the better the personalization.

The better the models become, the more accurate the predictions.

And the greater the automation, the less human intervention is required.

But perhaps there is another sign of maturity.

Knowing when to use what you know.

Knowing the difference between ordinary and sensitive information.

Between fact and probability.

Between a recommendation and a judgment.

Between what a system can do and what it should do.

And recognizing that the moment an inference is revealed may be as important to the service experience as the inference itself.

An intelligent service is not merely one that knows you need something before you ask for it.

It is one that also knows:

When to tell you.

How to tell you.

How much to tell you.

When to ask for your permission.

And when to stop.

As artificial intelligence continues to evolve, the question:

“Can organizations predict our needs?”

may become less important over time.

Because, in an increasing number of situations, the answer will be yes.

The more difficult question will be:

What should they be allowed to do with what they know?

Perhaps the best future for services is not one in which organizations know everything we need and act on our behalf.

Instead, it may be one in which they are capable of knowing a great deal about us while still maintaining the distance that allows people to feel that the service understands them without overstepping, helps them without labeling them, and moves toward them without taking away their right to choose.

One Final Paradox… Before the Beneficiary Becomes the Problem!

Imagine an organization launches its new proactive service.

The service analyzes beneficiary data, predicts their needs, and recommends what it believes is most suitable for them before they even ask.

Several months after launch, the team meets to review performance.

All the indicators look promising.

Prediction accuracy is high.

Service delivery time has decreased.

The number of steps has been reduced.

And a significant percentage of beneficiaries have received services they did not even know they were eligible for.

But one metric remains concerning:

The percentage of beneficiaries who accepted what the system inferred about them on the first attempt.

The figure is below target.

Some beneficiaries questioned how the organization had reached those conclusions.

Others felt the recommendations were more intrusive than they should be.

And some simply disagreed with the picture the system had created of them.

And this is where the real meeting begins.

Is the problem with the model’s accuracy?

No. Accuracy is excellent.

Is the problem with the service?

The numbers say it is useful.

Then surely the problem must be the beneficiary.

After several workshops, feedback analyses, and meetings involving the data, experience, and communications teams, a new category finally appears on the dashboard:

Beneficiaries Unreceptive to Proactivity

Then, with greater analytical maturity, they are divided into even more precise segments:

Reluctant to Share Data

Low Trust in Algorithmic Recommendations

Sensitive to Automated Classification

And perhaps, if someone continues to object:

Resistant to Intelligent Transformation

And so, we first used artificial intelligence to infer what the beneficiary needed.

Then we used it to understand why they did not like what we inferred about them.

Then we classified them based on their objection to being classified.

At which point, a new recommendation may appear on the team’s screen:

“We recommend developing a proactive intervention to increase this segment’s acceptance of proactive services.”

And the loop closes successfully.

The service no longer merely knows what you need before you ask…

It also knows why you do not like the fact that it knows.

And on the dashboard, naturally, this will be recorded under:

Deeper Beneficiary Understanding

— An institutional joke… until it becomes an actual KPI.

Thank you.

Knowledge Library Team

ODEL

Riyadh