Machine Learning in Transaction Monitoring
Fraudspect learns what "normal" looks like for each of your customers, then flags transactions that don't fit.
Standard rules check transactions against fixed limits, such as a maximum amount per transaction or per day. Machine learning adds a second check. It asks a simple question: does this transaction look like something this customer would normally do?
This helps you catch fraud that stays under your rule limits, such as a hijacked account, a sudden burst of activity, or a customer whose behavior suddenly changes.
How It Works
Every time a transaction is processed, Fraudspect automatically checks it in the background:
- Learns what's normal. Fraudspect looks at the customer's past transactions to understand their usual spending, the times they are usually active, where they usually transact, and how they usually pay.
- Spots what's different. It compares the new transaction to that usual pattern and to what is typical across the platform.
- Watches for slow build-ups. It looks at transactions over time, so it can notice when a customer's activity is gradually changing.
- Explains the result. It gives you the reasons behind each score in plain words.
What Fraudspect Looks At
| What we check | What it means |
|---|---|
| Amount | Is this much larger or smaller than the customer usually spends? |
| Frequency | Is the customer making more transactions than usual in the last 24 hours? |
| Recipient | Is the money going to someone this customer hasn't paid before? |
| Timing | Is this happening at an odd hour or on an unusual day for this customer? |
| Location | Is this far from where the customer normally transacts? |
| Device and channel | Is the customer using a device or method we haven't seen before? |
The Risk Score
Each transaction gets a score from 0% to 100%. The higher the score, the more unusual the transaction.
| Score | Risk level | What it means | Suggested action |
|---|---|---|---|
| 70% – 100% | High | Very unusual. Several things look wrong at once, for example a much larger amount, from a new place, on a new device. | Hold the transaction, ask the customer to verify it's them, or send it to a senior reviewer. |
| 50% – 69% | Medium | Somewhat unusual. For example, a larger amount than normal or an odd time of day, but not enough to be a serious concern. | Let it go through, or flag it for an analyst to review later. |
| 0% – 49% | Low | Looks normal for this customer. | Approve automatically. |
The score ranges above are Fraudspect's defaults (70% for High, 50% for Medium). You can adjust them to match how much risk your organization is comfortable with.
Understanding Why a Transaction Was Flagged
Open a transaction and look under ML Risk Analysis. You will see the main reasons behind the score, each written in plain language. For example:
- Amount: "Transaction amount is 4.5x higher than the customer's usual spending."
- Location: "Transaction was made 420 km from where the customer usually transacts."
- Time: "Customer transacted at an unusual time compared to their usual habits."
- Device: "Transaction was made from a device we haven't seen before."
You can see why an alert was raised without any data science knowledge.
New Customers
A new customer doesn't have enough history for Fraudspect to learn their habits yet (fewer than 3 past transactions). In that case:
- Fraudspect compares them to typical customers on the platform instead.
- The dashboard shows this notice: "This customer has not made enough transactions yet, so they were evaluated using baseline metrics."
- As the customer makes more transactions, scoring becomes more personal and more accurate.
Catching Account Takeovers
Fraudsters often avoid fixed limits by starting small and building up. For example, they may make a few small purchases before trying a large transfer.
Fraudspect watches how a customer's activity changes across a series of transactions. If it moves quickly away from their usual pattern, Fraudspect flags it as Active Behavioral Drift, so your team can step in before significant money is lost.
Spotting Money Laundering Networks
The checks above look at one customer at a time. Money laundering often involves several accounts working together, so each transaction can look harmless on its own.
To catch this, Fraudspect also looks at how money moves between accounts and flags suspicious patterns. Each one is raised as a case for your team to review, with a confidence level and a map of the accounts involved.
| Pattern | What it looks like | Why it's suspicious |
|---|---|---|
| Pass-through chain | Money moves from account A to B to C to D, with each account passing on almost all of it within a day. | The money is being moved along to hide where it came from. |
| Round trip | Money leaves an account, passes through a few others, and ends up back where it started. | Legitimate payments rarely go in a circle. |
| Many into one | Four or more different accounts send money to the same account within a day. | The account may be collecting money from many sources. |
| One into many | One account sends several small amounts to four or more different accounts within a day, kept under reporting limits. | The amounts are split up to avoid attention. |
Fraudspect gives each case a confidence level so your team can review the most likely ones first.