Automating client reporting: T&T International case study

Automating client reporting means handing the routine of reporting to a system, not the responsibility. The system reads the custodian banks' reports, extracts the relevant client data, consolidates across bank accounts, currencies, and products, and prepares the client report, including the compliance check and the performance calculation. The human reviews and approves. That is exactly how T&T International works today: the time spent on reporting has dropped by 80 percent, and the reporting quality partly exceeds what banks deliver. T&T International has approved that I tell this case with names and numbers.

Why is reporting the case with the fastest visible benefit?

Three properties make reporting the ideal first AI case. First, it is recurring: the quarterly report comes back no matter what else happens. Second, it is rule-based: the steps are the same every time, from reading the bank reports to sending. Third, it is laborious: many sources, many formats, many manual steps. What is recurring, rule-based, and laborious can be automated reliably, and the effect shows up in hours, not in promises.

On top of that comes measurability. On the first run after the build, you compare the same work before and after, step by step. Hardly any other process in the firm delivers such a clean proof, so quickly, of whether the investment pays off.

The industry is still at the beginning here: in the study "AI Use Cases and Crypto Assets in External Asset Management" by the Lucerne University of Applied Sciences and Finup from July 2026, 49 percent of surveyed wealth managers say they use AI; of the 65 reported applications, however, only 11 percent run in production (HSLU/Finup, July 2026). Anyone who automates reporting properly now belongs to the productive minority, with a result clients hold in their hands every quarter.

What was the starting point at T&T International?

T&T International is an independent wealth manager in Switzerland. A very small team serves clients whose wealth is spread across multiple bank accounts, currencies, and products. That spread is exactly what makes reporting expensive: every custodian bank delivers its reports in its own format, with its own logic and its own layout.

Before the rebuild, quarterly reporting was manual work: collecting reports, reading out positions and transactions, transferring them into the firm's own consolidation, calculating performance, assembling each client's report, applying the compliance review, checking everything, sending it out. The same work arrived again in compliance, with the same source data in a different form.

The consequence: a small team spent a noticeable share of its time carrying data from one format into another instead of looking after clients. And every error in that chain fell back on the team, not on the bank's format.

What was built, in plain language?

We built an operations application for T&T International. It performs four steps that people used to do:

1. Reading in: the application reads the bank reports of all connected custodian banks, whatever format they arrive in.
2. Recognizing: it extracts all relevant client data from them: positions, transactions, valuations, references.
3. Bringing together: it consolidates this data across bank accounts, currencies, and products and calculates performance using best-in-class methods.
4. Preparing: it produces the required documents, including a prepared compliance check, and shows each client in understandable form what they hold where.

Two points are decisive here. First, the language of the result: the client does not get bank syntax, but an understandable overall view of their wealth. Second, the location of the data: the application runs in a protected Swiss cloud, hosted on Microsoft Azure in Zurich North. Bank client data is not passed on to public AI models.

Where does the human control and decide?

The application prepares; it does not decide. Three points stay deliberately human:

• Irregularities are flagged, not corrected. If a value does not fit the pattern, the system puts the case forward for review, with the exact location in the original report.
• The compliance check is a preparation. The system checks against the configured rules and documents the outcome. The professional judgment and the sign-off stay with the human.
• No report goes out without approval. Every client report is reviewed before it is sent. In the early days the team re-checked every position; as the measured error rate proved itself over weeks, this became spot checks plus all flagged cases.

This division is not a compromise; it is the actual design: the machine compiles, the human judges. That is exactly what makes the result robust, including towards auditors and supervisors.

What is the result in numbers?

The time spent on reporting has dropped by 80 percent. The most laborious routine of the quarter has become a controlled run in which the team reviews instead of producing. Quality has not suffered, quite the opposite: the performance calculation follows best-in-class methods, and the reporting quality partly exceeds bank level, because the client sees their total wealth across all banks in a single, understandable view.

This is the comparison per work step:

Step	Before (manual)	After (with the application)
Collect and read in reports	By hand, bank by bank	Minutes, automatic
Extract positions and transactions	By hand from each report	Minutes, automatic
Consolidation across banks and currencies	By hand, error-prone	Minutes, automatic
Performance calculation	Calculated by hand	Minutes, automatic
Compliance check and documentation	Documented manually	Prepared, reviewed by a human
Checking and sending	Full review by hand	Spot checks plus flagged cases
Total time required	100%	about 20%

Before-and-after process of quarterly reporting at T&T International.

Where things typically get stuck

An honest case includes this part: not everything runs from day one in projects like this. Three points are typical: a custodian format that changes quietly and has to be reconnected; positions that are misallocated at first and trigger a correction loop; and a review routine that starts stricter than necessary and only relaxes as the measured error rate comes in.

What holds in every reporting project: connecting the formats is the part you must not underestimate. If you work with five custodian banks, you work with five logics, and at least one of them changes every year.

What can other wealth managers take from this?

You need neither the size nor the budget of a bank. Four things from this case transfer directly:

• Start with one quarter, in parallel with the existing method. The old process remains the reference. Only when the system's numbers match across a full quarter do you switch.
• Clarify the formats first. Which custodian banks, which report types, how stable are the formats? This inventory decides effort and duration more than any model question.
• Define the approval points before the build. Who reviews what, at what depth, with what documentation? If you settle this only in operation, you lose half the benefit.
• Budget realistically. Expect a mid five-figure amount for the build and a manageable monthly fee for operations. The concrete orders of magnitude and the payback calculation are in the article What does AI cost for financial service providers?.

Reporting is one of seven use cases that already pay off for independent wealth managers today. The overview with effort and data risk per case is in the pillar article: AI for independent wealth managers.

From my practice: The most visible benefit of this project was not the time saved, but what happened afterwards. Once the team had experienced that the application prepares reliably over weeks and the human only reviews, the next processes landed on the list by themselves. Trust in AI is not created by a presentation, but by measured quality across many repetitions. That is exactly why recurring reporting is the best first case: it delivers this proof anew every quarter.

Conclusion: your next step

Write down your reporting process honestly once: every step, every hour, from collecting the bank reports to sending. That list is your business case, and it shows exactly which steps a system can take over and where your review stays. What such a solution looks like for your firm is shown on AI for wealth managers. With the readiness check you can see where your operation stands in a few minutes.

Note: We are not a law firm. All statements on legal and regulatory topics, such as compliance duties in reporting, are subject to change and do not replace legal advice. For specific questions, please consult a lawyer.

FAQ

Does the application replace compliance responsibility?

No. The system checks against configured rules, flags irregularities, and documents the outcome. The professional judgment, the risk assessment, and every approval stay with the human. What is automated is the preparation, not the responsibility.

What happens to bank client data?

The application runs in a protected Swiss cloud, hosted on Microsoft Azure in Zurich North. Data is not passed on to public AI models. The handling of client data follows the same principles as before: purpose limitation, access only for authorized staff, full logging.

Does this work with our custodian banks?

That is the first question we clarify in a build. Each bank format is connected once; after that, reading in runs repeatedly. If a new custodian bank joins or a bank changes its format, that is a one-time extension, not a rebuild.

How long does such a build take?

That depends above all on the number and stability of the custodian bank formats. In our experience a build of this kind takes a matter of months, with the format connections accounting for most of the time.

What does automated reporting cost?

That depends on the number of custodian banks and their formats. Payback comes from the hours freed up: with 60 to 100 hours of routine work per month, a setup typically pays for itself in 6 to 12 months. The concrete orders of magnitude for setup, pilot and running costs are in the article What does AI cost for financial service providers?.