← Back to perspectives
/Intellicon FDE Team

Why Can Companies Still Struggle to Access Activity Logs When Their Data Stays Within National Borders?

When a vendor can no longer help, a company may struggle to trace even an invoice’s approval history. Start by retrieving and checking actual records, then simulate a service shutdown to confirm that essential documents remain accessible and work can continue.

AIEgentWrX導入人機協作
Why Can Companies Still Struggle to Access Activity Logs When Their Data Stays Within National Borders?

Suppose a customer asks your company to explain why the AI Agent handling invoices used payment details from a particular email. IT can explain where the data is stored, but still has to request activity logs from the vendor, while finance cannot determine who approved the transaction during processing. Even if the data stays within national borders, the company still needs to answer two questions: Can we independently establish what happened? And how will work continue if the service is discontinued?

At Intellicon Solutions, we believe that when companies assess digital sovereignty, meaning their ability to independently control their data and systems, they should evaluate deployment location separately from actual control. Deployment location establishes where data and systems reside. To establish whether a company genuinely controls them, teams need to retrieve records, verify their contents, and test whether alternative arrangements work. A data center address alone is not enough to pass acceptance testing.

Confidence in Producing Evidence Does Not Mean Control Has Been Verified

In an IBM survey cited by TechOrange, 82% of executives surveyed were highly confident that they could produce evidence of their organization’s control over its data and systems. Yet an assessment based on evidence reported by the organizations themselves placed only 26% at the index’s highest level. These percentages measure different things, so the gap cannot be used to calculate how many companies fall short. Taiwan was also not included in the survey, so the findings cannot be used to estimate the current state of Taiwan’s traditional industries.

The survey also found that 56% of respondents said evidence collection required manual work or was difficult because it took too much time, while 50% said they had not tested their controls, service recovery, or arrangements for exiting a vendor’s services. However, the index relies on self-reported information, has not been independently audited, and cannot establish that IBM products outperform competitors. We recommend that procurement teams turn these issues into acceptance criteria: who retrieves the records, whose cooperation is needed, and what evidence the company can still obtain when the vendor cannot help.

Start with One Task and Check Whether the Records Explain How It Was Handled

For invoice processing, IT, finance, and information security can map the workflow together before acceptance testing. They should identify which documents the AI Agent will read, whose account it will use, which actions it can perform, and which steps require human approval. Pay particular attention to whether reusing an employee account’s existing permissions allows the AI Agent to do more than the task requires.

If the task only involves checking invoices, the AI Agent should not also receive permission to modify supplier records simply because the employee account already has that access. For changes to bank account details, we recommend having a designated person separately verify the source of the information and the proposed changes before approval under company procedures. Acceptance testers should also run actual tests to confirm that unapproved actions are blocked.

Next, complete a task in a test environment and have the person responsible retrieve the relevant records. Check the document versions used, the account under which actions were performed, execution times, approval records, and the results in the business system. Also check whether information such as a document number can establish the sequence of processing across different systems. If the records only show what each system completed separately, list the steps that remain unclear rather than assuming you have a complete account of the process.

EgentWrX supports cloud, on-premises, and hybrid deployments, and offers audit log exports so administrators can obtain the activity records captured by the platform. Hash chain integrity verification can help detect whether those records have been altered. Passing this verification, however, does not mean that every part of a business process has been recorded or that the full sequence of decisions across different systems can be reconstructed. We therefore recommend exporting and verifying the logs, then checking them individually against records from finance and other relevant systems, documenting which questions have been resolved and which remain unanswered.

Being Able to Switch Models Is Not Enough: Check Whether the Work Still Gets Done

We recommend first listing the models, databases, connectors, and platforms required for critical tasks, because replacing one component does not mean the other services can be migrated along with it. For each component, document who is responsible, how it is accessed, and what alternative will be used if it becomes unavailable. This prevents the exit plan from becoming a list of vendor names with no explanation of how business operations will continue.

EgentWrX lets companies choose models from different providers, on-premises models, and open-source models within the same platform, providing a basis for evaluating alternatives. But having other models in a menu does not mean they will produce the same results after a switch. Nor does it mean that the entire platform, its data, and its system integrations can be migrated directly, or that automatic failover is available.

For procurement quote comparisons, for example, we recommend selecting data representative of everyday work and having an alternative model compare quotes using that same data. Procurement staff should then check whether item matching, delivery schedules, and payment terms meet business requirements, while recording which instructions, data, and operating steps need to change. If staff still need to spend substantial time looking up information to complete the results, that additional work should count toward the burden of adopting the alternative. A successful model response alone is not enough to pass the test.

Clearly Identify Untested Arrangements During Acceptance Testing

In addition to testing model switching, we recommend simulating a loss of access to the platform. This helps establish whether business users can obtain essential documents and continue processing work manually or through existing systems. Before the exercise, business owners should define which tasks can be temporarily interrupted, which must continue, and who has authority to suspend work. IT and information security should then check whether the alternatives meet the existing requirements.

The exercise should also cover the handover after service is restored: which documents have already been processed manually, who is responsible for reconciliation, and how duplicate payments or missed tasks will be prevented. If the team has not actually tested an arrangement, mark it as pending verification and specify the owner and what the next test needs to establish.

At your next procurement or implementation meeting, start by selecting one critical AI task. Ask the team to submit a deployment plan, the evidence collected, and records of service recovery exercises separately, so managers can decide which tasks are ready to go live and which need further testing. Even when a vendor promises that data will remain within national borders, the company still needs to perform acceptance testing to establish whether it can control its own data and systems.

FAQ

Does keeping data within national borders mean a company has control?

Data location alone is not enough. Companies also need to confirm that they can retrieve and verify activity logs and keep critical work running when the vendor cannot provide service. We recommend evaluating deployment location, evidence collection, and service recovery exercises separately. Keeping data within national borders is not a reason to skip the other checks.

If records pass integrity verification, can they explain the entire process?

No. Integrity verification can help detect whether records have been altered, but teams still need to check that all required records are present and that the sequence of processing across different systems can be established. Select a test document, trace it from the input data through to the results in the business system, and list each piece of evidence that is still missing.

Does having a choice of models on a platform count as an exit plan?

No. A choice of models only means that replacing a model may be possible. Companies still need to test alternative models on the same task to confirm that the results meet business requirements, and separately rehearse what to do if the platform becomes inaccessible. They must decide in advance who will retrieve essential data, how work will be handed over after manual processing, and who will reconcile records once service is restored.

References

30 minutes to map out which work to hand to AI first

Want every employee to have their own AI teammate?

A consultant will be in touch shortly.

ISO/IEC 27001:2022 Certified ISMS