Fraud Prevention Software: Risk Scoring and Recommended Outcomes

Fraud prevention is the part of Akoben that decides how risky a transaction looks and what outcome it recommends. It is the scoring and decisioning core of the platform: it evaluates each incoming transaction event, calculates a risk score, and produces a recommended outcome. Akoben is not a gateway in the payment path. What the receiving system does with the score and the outcome is that system's decision, and what the investigation team does with the alert is theirs. Transaction monitoring is the alerting, queueing and investigation surface built on top of this core.

What Akoben fraud prevention does

Akoben receives a transaction event and enriches it with the context needed to judge it: the account's recent history, the counterparty relationship, and the device and channel attributes the sending system supplied. The enriched event (the full path is set out step by step) is then evaluated against the active ruleset, which combines thresholds, velocity windows, transaction and account attributes, and behavioural conditions.

The result has two parts. The first is a risk score, a single value that summarises how much of the ruleset the event triggered and how strongly. The second is a recommended outcome, which is the response the configuration says that level of risk calls for. Both are produced for every event evaluated, not only for the ones that look suspicious, so a team can see the full distribution of scores and not just the tail that raised an alert.

Prevention versus detection

Detection is the analytical judgement that an event looks wrong. Prevention (the glossary defines it) is what that judgement achieves when it reaches the right people while it still matters. A system can detect accurately and still prevent nothing, if the finding arrives in a weekly report after the funds have been moved on and the account emptied.

Akoben is built around the second half of that chain. Events are evaluated as they are posted, and the score and recommended outcome are available to the receiving system and to investigators straight away. That is a deliberately bounded claim. Akoben does not sit between a customer and their payment, and it does not execute a response against a transaction. What it does is shorten the time between an event happening and the people and systems that can act on it knowing how risky it was. For a fraud operations team, that gap is usually the difference between recovering a loss and writing it off.

How events reach Akoben

Source systems send events to Akoben in one of two ways. They can post events over a REST interface, sending each transaction as it happens without waiting on a reply. Or they can supply scheduled batch files containing a period of transactions, which suits systems that cannot post events as they occur and suits the historical validation of a new ruleset.

There is no message-queue integration and no call that holds a payment open while Akoben responds. That shape is a design fact worth stating plainly, because it determines what Akoben can promise. A system that posts an event and carries on will not be slowed by Akoben, and equally Akoben will not be the component that decides whether that payment goes through. The integration pages describe each connection type, and the REST interface is the usual route for a product that originates its own transactions.

Risk scoring

Rule outcomes are combined into the risk score, which gives the receiving system a single value to work with and gives investigators a way to order a queue. A binary outcome treats a transaction that barely crossed one loose threshold exactly like one that matched several rules at once. A score keeps the difference, and a graded response is only possible where the signal underneath it is graded.

A score is also what makes review manageable. A queue ordered by score puts the transactions most likely to be fraud in front of the investigator first, which matters when there are more alerts than a team can open in a shift. The same value can be used downstream by whatever system receives it, for example to route a payment for additional checks that the receiving system itself owns.

The recommended outcome

The recommended outcome is the response the ruleset assigns to a given risk level. It is a recommendation in the strict sense: Akoben produces it and delivers it, and the receiving system decides whether and how to apply it. This division of responsibility is intentional. The institution that owns the customer relationship, the payment rails and the regulatory obligations is the one that should decide what happens to a customer's payment, and Akoben's role is to give that decision better information, sooner.

Which outcomes are produced is configured per tenant, which is how the next section's staged operating model works.

Alerting-only mode and acting mode

A new ruleset is a risk. If it is too tight, it raises alerts on good customers; if it is too loose, it misses fraud. The usual response is to tune cautiously, which means weeks of exposure to the old rules. Akoben handles this with a per-tenant operating mode.

In alerting-only mode, a tenant runs a ruleset and sees the alerts and scores it produces, without those results being treated as outcomes to act on. The team studies how many alerts the ruleset generates, which transactions it catches, and which it flags wrongly. When the results have proven themselves, the tenant moves the ruleset to acting mode.

This is a deliberate operating mode, not a limitation. A proof of concept can run entirely in alerting-only mode, and an established tenant can use it every time a ruleset changes. It answers the objection a fraud team actually raises about any new system, which is that it will mishandle good customers on the first day.

Fast enough to act on

A risk score is useful in proportion to how soon it reaches someone who can use it. Akoben evaluates events as they arrive, so the score and the recommended outcome are ready for the receiving system and for the investigation queue as soon as evaluation completes, rather than waiting for a nightly run.

For prevention, this is the claim that matters, and it is a different claim from being in the payment path. An operations team that sees a high-risk transaction soon after it is posted can contact a customer, restrict an account, or alert a partner while the money may still be recoverable. Akoben does not promise to act on the transaction itself. It promises information fast enough for the people and systems that do act to make a difference.

Payment fraud

Payment fraud covers a wide range of behaviour: transfers the account holder never authorised, payments to newly added beneficiaries under pressure, rapid sequences of small payments designed to stay below limits, and cash-out patterns after funds arrive. No single rule catches all of these, which is why the scoring core combines signal families instead of relying on one.

Threshold rules catch the blunt cases, such as an unusually large payment. Velocity windows catch accumulation, where each payment is small but the count or the total in a moving window is not. Behavioural conditions compare the transaction with the account's own history and catch activity that is ordinary for the population but unusual for this account. The risk score brings these together, so a transaction that trips several at once is ranked above one that trips a single weak signal.

Wallet and card channels

Akoben evaluates transactions from payment, wallet, card and bank flows. The scoring logic does not depend on the channel, but the rules do: a wallet cash-out has attributes that a card payment does not, and a bank transfer carries beneficiary information that a card authorisation message lacks. The ruleset expresses channel-specific conditions using the transaction and account attributes the source system supplies.

For mobile money operators, that includes agent and cash-out behaviour. For fintechs, it includes the pace of onboarding and the first transactions on a new account. Fraud prevention for mobile money and for fintechs describe how those flows are monitored.

Tuning the scoring

Scoring is not set once and left alone. Fraud patterns change, and a ruleset that was well balanced last quarter can drift into raising too many alerts or missing new behaviour. Tuning means reviewing which rules contribute useful alerts, adjusting thresholds and windows, and validating each change before it goes live.

Akoben supports that cycle directly. An analyst changes rules without an engineering release, which creates a new ruleset that is applied as a whole and can be reverted to the previous version if the results are undesirable. A ruleset can be validated against historical transactions in batch first. Named reports for rule effectiveness and false-positive rate show which rules are earning their place. The fraud rules engine describes this in detail, including the option of having Akoben design and tune rules as a service.

Integrations

Akoben takes events from the systems that originate transactions: payment switches, core banking systems, mobile wallet platforms, card systems and data platforms, either by posting events over REST or through scheduled batch files. Which connections an institution needs depends on the channels it operates.

A fintech operating through its own product usually posts events directly from its application. An institution with a core banking system and a payment switch typically connects both so that the score reflects account records and transaction flow together.

Deployment

Akoben is delivered as SaaS, in a private cloud, or on-premise, chosen according to an institution's regulatory, data residency and operational requirements. Deployment decides where the platform runs and who operates it; the rules, scoring and operating modes described on this page are how you configure what it does with an event.

Common questions

What is real-time fraud detection?

Real-time fraud detection evaluates a transaction as it happens, rather than reviewing it in a batch hours or days later. The point is not speed for its own sake. It is that a fraud team's options narrow as time passes: a suspicious transfer surfaced within seconds can still be queried, recalled or frozen by the institution holding the account, while the same transfer surfaced the next morning has usually already moved on, often through several accounts.

This is why the distinction between real-time and retrospective monitoring is operational rather than technical. Both find the same transaction. Only one finds it while the institution still has a decision to make about it.

Retrospective review is not a lesser form of the same thing, though. Some patterns — an account slowly accumulating small deposits from many senders, a merchant whose refund rate drifts upward over weeks — are only visible across a window longer than any real-time evaluation can see.

What is fraud prevention?

Fraud prevention is the set of controls that reduce the losses fraud causes by acting on suspicious activity while action still changes the outcome. In transaction systems that means scoring each transaction against rules and the account's own behaviour, producing a recommended outcome, and getting that outcome to the people and systems able to act on it quickly. Prevention differs from detection after the fact, which only explains what already happened.

Prevention is a matter of timing as much as accuracy. A correct finding delivered after funds have left the institution and been moved on is a record of a loss, not a prevented one. The value of a prevention capability is therefore measured by how soon a usable judgement reaches whoever can act on it.

It is also a matter of who acts. A scoring system can recommend an outcome, but the decision to apply it belongs to the system or team that owns the customer relationship and the payment.

See how alerts and investigation sit on top of scoring

What is fraud detection?

Fraud detection is the identification of transactions or account behaviour that show the characteristics of fraud. A detection system compares activity with configured rules, with thresholds and with what the account normally does, and flags what does not fit. Detection produces a finding; whether that finding prevents a loss depends on how quickly it reaches someone able to act, and on what that person or system does with it.

Detection and prevention are often used as if they were opposites, but they describe two ends of the same chain. Detection is the analytical step, deciding that an event looks wrong. Prevention is the operational outcome, where that decision arrives in time to matter.

The analytical step uses the same families of signal whatever the timing: amounts and thresholds, counts inside time windows, attributes of the transaction and the account, and departures from the account's own history.

Read the glossary entry on fraud prevention

What is transaction risk scoring?

Transaction risk scoring converts the outcomes of many rules and behavioural checks into a single value that expresses how risky one transaction looks. Instead of a binary match or no-match result, the score preserves how many signals fired and how strongly. That lets a receiving system act on one value, and lets investigators work the riskiest items first rather than in arrival order.

A transaction that narrowly crossed one loose threshold and a transaction that matched several independent rules at once both count as matches, but they deserve different urgency. A score keeps that distinction. It is also what makes a graded response possible: a recommended outcome can only be proportionate if the signal underneath it is graded.

The score is only as good as the rules and baselines feeding it, which is why tuning the ruleset and reviewing its results are part of running scoring, not an afterthought.

See how the rules that feed the score are maintained

Discuss your transaction monitoring requirements

Tell us which transaction flows you need to monitor, which systems Akoben would integrate with, and how your fraud team works today.

Request a demo