Fraud and technology: How to handle the double-edged sword
18 November 2026 · 10 min read · Amumpaire

At 8:17 a.m. on a Monday morning, the finance manager of a mid-sized local company received what appeared to be an ordinary instruction from the managing director. The email was short. A supplier payment of UGX 186 million had become urgent, the managing director was supposedly travelling, the supplier was threatening to stop deliveries, and the payment needed to leave before 10:00 a.m.
Attached was an amended invoice containing new bank details, followed by a WhatsApp message. “Please handle this urgently. I am entering a meeting.” The profile photograph was correct, the writing style looked familiar, and the email signature was perfect. Two minutes later, the finance manager received a short voice note that sounded like the managing director. So the payment was approved.
By lunchtime, the real managing director was asking why anybody had transferred UGX 186 million. That is how many modern fraud investigations begin, not with a masked hacker breaking through a firewall, but with something familiar enough to be trusted.
The organisation immediately asked the question everybody asks first. “Who stole the money?” The investigations team, I was leading, asked a different question. “What exactly happened to the evidence?”
Technology has become a magnificent instrument for detecting fraud, reconstructing transactions and identifying relationships that would have remained invisible twenty years ago, but the same technology has also made fabrication cheaper, impersonation easier, evidence more fragile and dishonest explanations remarkably sophisticated. The sword cuts both ways.
The fraud was not in the email alone
Management initially described the incident as a “hacked email.” That conclusion was premature. An email appearing fraudulent does not prove that an email account was compromised, because the message might have been spoofed, a genuine mailbox might have been accessed, a forwarding rule might have been created, an employee's device might have been compromised, credentials might have been stolen through phishing, or somebody inside the organisation might have deliberately assisted the transaction.
Those possibilities have different evidential consequences. We therefore started with the transaction rather than the theory.
a) Who initiated it?
b) Who received the instruction?
c) Which device received it?
d) Which mailbox contained it?
e) Who approved the payment?
f) Was the invoice amended before or after it reached the organisation?
g) When were the beneficiary details changed?
h) Which authentication records existed?
i) Which IP addresses appeared in the relevant logs?
j) Had the supposedly urgent supplier communicated through the same email domain before?
The investigation became tough for everyone quickly. The finance officer had not simply received an email, but also a WhatsApp message and a voice note, all reinforcing the same false story. That matters because fraudsters increasingly manufacture corroboration. One false communication creates suspicion but three apparently independent communications create confidence, except sometimes they are not independent at all.
Screenshots are clues, not evidence strategies
The first thing someone had done after discovering the fraud was take screenshots which was sensible as an emergency reaction, but dangerous as an investigation strategy. A screenshot may show what appeared on a screen at a particular moment, but usually tells you little about what sits behind it.
Consider the email. A screenshot showed the sender as the managing director. It did not establish the actual originating infrastructure, message routing, authentication results, server timestamps or metadata contained in the original message headers. The WhatsApp screenshot showed a profile photograph and telephone number but did not prove who controlled the device at the relevant time. The voice note sounded like the managing director but sounding authentic and being authentic are now very different propositions.
The digital evidence examiner therefore needs the original objects wherever possible, not merely photographs of them.
a) Preserve the email in its native format.
b) Preserve message headers.
c) Preserve audit logs.
d) Preserve the device.
e) Preserve authentication records.
f) Preserve cloud logs before retention periods expire.
g) Record who collected what, when, using which method.
If this eventually becomes litigation or a criminal prosecution, opposing counsel should not be allowed to destroy months of investigative work with one basic question: “Where is the original evidence?”
Then the internal evidence changed the case
The bank details on the fraudulent invoice had been altered which looked obvious.What was less obvious was when a forensic comparison revealed that the invoice received from the genuine supplier contained one account number, while the version attached to the fraudulent email contained another.
But there was another problem. Three weeks earlier, someone inside the company had searched the accounting system for previous payments to that same supplier, which alone proved nothing. Then we found that the same user account had accessed the supplier master file outside normal working hours, which was still not enough.
Then investigators identified a failed attempt to modify the supplier's banking details several days before the fraudulent payment. Now we had a pattern. This is where technology becomes extraordinarily powerful in fraud work. Individual events may be innocent, but when properly reconstructed against time, identity and system activity, the relationship between events begins to speak.
An employee accesses a supplier record, a mailbox receives an unusual login, a fraudulent invoice appears, followed by a payment instruction, a voice note creates urgency, the approval bypasses ordinary confirmation, and then money moves. A case emerges not from one dramatic piece of evidence, but from the alignment of many small facts.
Artificial intelligence has made “I saw it” weaker evidence
The most dangerous assumption in the case concerned the voice message. Three employees listened to it and independently said, “That is definitely the MD.” But human familiarity is not forensic authentication.
Generative AI has dramatically reduced the cost of producing persuasive synthetic audio, photographs, documents and video. That creates a new problem for lawyers and investigators because evidence that once carried an intuitive presumption of authenticity may now require technical verification. A familiar face can be generated, a voice can be cloned, a signature can be inserted and a document can retain an apparently convincing layout while the underlying content has been manipulated.
The lawyer handling such a case must therefore stop asking only whether evidence looks genuine and start asking whether its provenance can be demonstrated.
a) Where did the file originate?
b) What is its hash value?
c) Has the file changed?
d) What metadata exists?
e) What application generated it?
f) Are there earlier versions?
g) Can server-side records corroborate it?
h) Does the timeline make technical sense?
Technology has changed the courtroom question from “Can you see this?” to “Can you establish where this came from and what happened to it?”
The fraudster depends on human urgency
The most important control failure was not technological but procedural. The organisation had a requirement that changes to supplier banking details be independently verified but nobody followed it. Why? Urgency. The supposed managing director was travelling, the supplier was supposedly threatening disruption and the finance team did not want to appear obstructive. Everybody had enough information to become comfortable and not enough skepticism to become difficult.
Fraudsters understand organisations better than many control designers do.
a) They know people fear delaying the boss.
b) Junior staff hesitate to question senior executives.
c) Month-end creates pressure.
d) Large organisations confuse email authority with actual authority.
And technology allows them to manufacture the appearance of authority at scale.
The best fraud control is therefore sometimes embarrassingly simple.
a) Call the person using a trusted number already held in your records.
b) Do not use the number provided in the suspicious instruction.
c) Verify changed beneficiary details independently.
d) Separate initiation from approval.
e) Treat urgency involving money as a reason for additional verification rather than reduced verification.
That ten-minute inconvenience can save UGX 186 million.
Investigators must preserve evidence before interviewing everyone
There is another common mistake. Management discovers fraud and immediately starts calling meetings, the suspected employee is summoned, then IT begins examining the laptop. Someone resets passwords, another employee forwards suspicious emails, the affected computer is restarted the WhatsApp conversation is exported and by the time the forensic examiner arrives, everyone has been “helping.”
Unfortunately, helpful people alter evidence. Logging into an account creates new logs, opening files may change metadata, resetting credentials may terminate sessions investigators wanted to examine, deleting malware may destroy evidence of how entry occurre and interviewing suspects too early may tell them exactly what investigators know. Preservation must therefore precede explanation wherever practicable. A good investigator sometimes learns more from thirty silent minutes with logs than from three hours of confident interviews.
What to do before the evidence goes cold
When fraud involving digital systems is discovered:
1) Stop further loss without unnecessarily destroying evidence. Block or freeze transactions, accounts or credentials using controlled procedures and document every intervention.
2) Preserve original messages. Obtain emails in native format with full headers rather than relying only on forwarded copies or screenshots.
3) Secure relevant devices. Do not casually browse through laptops and phones before forensic preservation has been considered.
4) Preserve system and cloud logs immediately. Authentication, VPN, email, ERP, banking and security logs may have short retention periods.
5) Record the timeline. Establish when the instruction arrived, when it was opened, when approvals occurred and when money moved.
6) Preserve transaction evidence. Obtain bank confirmations, beneficiary information, payment approvals, supplier master-file changes and mobile money records where relevant.
7) Document chain of custody. Record who collected each item, when, from where, in what state and what happened to it afterward.
8) Separate facts from assumptions. “The account was hacked” is a theory. “A login occurred from this IP address at 02:13” is evidence.
9) Control interviews. Preserve volatile evidence before alerting people whose conduct may later require investigation.
10) Bring legal and forensic teams together early. Evidence collection, privilege, employment rights, privacy obligations and potential criminal proceedings should not be treated as separate problems after the damage is done.
The real double-edged sword
Technology did not merely enable the fraud in our case but also exposed it. Email records challenged the apparent sender, system logs revealed unusual access, document comparison exposed the changed invoice, and Bank records reconstructed the movement of money. Authentication evidence helped distinguish what employees said they had done from what systems recorded them doing.
That is the paradox practitioners must understand. Technology gives fraudsters speed, scale, disguise and believable fabrication, while giving competent investigators timestamps, metadata, logs, linkages and traces that dishonest people rarely realise they have left behind. But those traces are useful only if somebody protects them. The greatest danger in a digital investigation is therefore not always that evidence does not exist but that the organisation destroys, alters, overwrites or strips it of context before anyone competent examines it.
Modern investigations are not won by collecting the largest possible amount of data, but by identifying what matters, preserving it properly, proving where it came from and connecting it to the human decisions surrounding it. In modern investigations, truth is not found by collecting more data, it is found by preserving the right data, in the right state, with the right context, before people start explaining it away.
Copyright IFIS, 2026. All rights reserved.



Comments
No comments yet. Be the first to share your thoughts.