How Autogon AI is cracking a resistant fraud market
Autogon AI's COO and Lead Backend Engineer on building Omniguard, surviving a pricing crisis, and selling fraud detection into institutions that were happy losing money to stay comfortable.
Picture a fraudster or launderer moving money through the Nigerian financial system. They don't move it in one transfer. They break it into dozens of smaller amounts, each one carefully under the ₦5 million threshold that would trigger a regulatory report, spread across multiple accounts fanning the money in and straight back out. Every single transaction looks clean. Every single one passes. Rule-based fraud detection tools see each transfer in isolation, checks it against the threshold, and passes it through. It’s a clever work around that doesn’t fit the expected pattern.
This is structuring. It is not exotic. It is how money moves through Nigerian payment rails every day, and it is what rule-based fraud detection was never built to catch.
In 2024, online fraudsters carted away with ₦52.26 billion, more than triple what they stole the year before — though that figure was partly inflated by a single ₦31.1 billion incident involving one entity. By 2025, that figure had fallen by 51%, down to ₦25.85 billion. Nigeria Inter-Bank Settlement System (NIBSS) and the CBN attribute the drop primarily to BVN and NIN integration, a systemic identity reform that closed verification gaps across banks, fintechs, and agent networks at scale. Identity management closed one gap. Transaction monitoring is the next one.
That’s the gap Autogon AI is solving for. The three-year-old company didn’t set out to build a fraud product. It was running AI readiness assessments for large institutions, the kind of engagement where you audit a business and tell them what’s broken before you tell them what to build. While at that, charge-back fraud kept surfacing as the pain point nobody had solved well. They looked at what existed, decided they could do better, and built something.
A few demos, a first customer, then a customer who passed a CBN inspection partly because of what Autogon AI had put in place. The product has been validated.
That’s the origin story for Omniguard, Autogon AI’s AI-powered transaction monitoring and fraud detection product. That's the origin story for Omniguard, Autogon AI's transaction monitoring and fraud detection product. The company was founded by Obi Ebuka David, who previously built Identity Pass, now known as Prembly.
In this conversation, COO Boluwatife Ashimolowo and Lead Backend Engineer Olumide Olaseyo break down how Omniguard works, what it took to price and sell a new product into institutions built to resist change, and why a rule written by a compliance officer in 2019 still can’t catch what’s moving through Nigerian payment rails in 2026.
Editor’s Note: The conversation below is edited for length and clarity.
Boluwatife, Autogon AI explored a lot of verticals before landing on financial fraud. What changed?
We explored many verticals and that’s what got us here. We were running AI readiness assessments for large institutions, and fraud kept coming up. We looked at what was being used to solve it, specifically charge-back fraud and transaction monitoring, and we thought we could do better. We built, ran demos, got reactions. Then we got our first customer.
After that, it became clearer. We started getting attention from the CBN, from Nigeria Electronic Fraud Forum (NEFF). I held a keynote at the NEFF General Conference, and the conversation in that room, with a lot of CBN people present, seemed to matter. The mandates followed. In May 2025, the CBN issued baseline standards requiring every regulated institution — banks, MFBs, fintechs, payment service providers — to deploy automated, real-time AML monitoring systems. A second directive came in March 2026, setting mandatory minimum requirements for AI-powered detection and reporting. Our solution was already built for exactly that. We didn’t have to retrofit anything.
And the products that didn’t survive?
Mednosis. Our healthcare [focused on radiology] diagnostics platform is one example. We drove to hospitals, ran demos, and showed the product. The adoption wasn’t there. I don’t think it was a product problem or a competence problem. The institutions just weren’t ready to shift like that. We shelved it, kept some resources on it in case the timing improves.
The rule we follow internally: push a product for three months, and if it’s not converting, pull it back. Three months is already generous. In the US, teams are running that cycle in six weeks. We try to move at that speed because the alternative is spending months on something the market isn’t ready for.
Was there a point where this felt genuinely risky, rather than just slow?
Honestly, no, not in the way you’d expect. The shift toward fraud was mostly positive the whole way through. The one moment that actually made us sit back and rethink things was pricing, not adoption. We started out charging what I’d call evidence-based pricing, basically pricing to reflect the actual cost of preventing a loss, which meant we were quoting in the millions. Then a competitor showed up charging ₦350,000 flat. The product didn’t have as many features/depth of solution as ours, but it solved enough of the basic transaction monitoring problem that buyers started comparing us to it on price alone. This forced us to introduce tiers instead of one flat enterprise number. If there were any concerns it wasn’t “is this the right market?” but “are we pricing ourselves out of it?”
Boluwatife Ashimolowo, Autogon AI’s COO
Many fraud detection tools say they use AI now. What’s the real difference from what businesses that process transactions at scale already have?
Boluwatife: The existing systems are rule-based. A bank decides: if someone changes their device, they can’t send above a certain amount for 48 hours. That rule sits in a database. If nobody wrote a rule for a particular scenario, it slips through. If too many rules exist, they start contradicting each other.
So you get banks flagging transactions that aren’t fraud and missing the ones that are. I’ve had my own transaction blocked because I changed devices and sent money to my own account, under my own name. The rule existed for a good reason once. Now it’s just friction.
Omniguard monitors transactions in real time and builds a behavioural profile for each user. Someone who uses one bank for small, frequent transfers and another for bigger purchases has a pattern. A transaction that breaks that pattern sharply gets flagged, but instead of blocking it outright, we score it against a threshold. Below a certain risk level, it still goes through. We’re also building toward face verification for borderline cases, so the user can clear themselves instead of hitting a wall.
Olumide, what makes that meaningfully different from a smarter rulebook?
Olumide: Rules are static. A human has to predict every scenario in advance and write a rule for it. What we do is combine rule-based logic with AI models that learn from behaviour over time, so the system builds context around an event instead of judging it in isolation. A late-night login from an unfamiliar device reads differently depending on whether it’s a one-off or whether it’s happening across multiple accounts from the same source at the same time. A static rule can’t make that distinction. Our model can, because it’s reading the combination of signals, not just the single event.
A lot of what we built is aimed at cutting false positives. Here’s an example, “I’ve moved two or three million naira in transactions in a single day without issue, then tried to send a smaller amount to someone and been told to do a liveness check, turn my head to the side, confirm it’s really me, for a transaction that should have looked completely ordinary against my own history.”
From a customer’s standpoint, that’s just friction with no payoff. Rule-based systems do that constantly, because the rule doesn’t know any better. It kills the experience for the customer and buries the fraud team in alerts that don’t mean anything.
Olumide Olaseyo, Autogon AI’s Lead Backend Engineer
Fraudsters are always exploiting new avenues to steal. How does the system actually improve over time? Does someone have to feed it new fraud patterns by hand?
Olumide: It learns from its own data. There’s a concept in computer science called case-based reasoning: once the system has seen an outcome, it can apply that knowledge to a new, similar case. We also draw on other data sources to keep improving the model. Nobody on the fraud team has to manually write a new rule every time attackers shift their approach. The model adapts on its own.
Give me one real case. A specific transaction or pattern Omniguard caught that a rule-based system would have missed.
The clearest example is structuring. Nigerian AML rules require a report on any transaction above 5 million naira for individuals, 10 million for corporates. So launderers don’t cross it. They break a large sum into smaller transfers, each one comfortably under the line, spread across several accounts that fan money into one beneficiary and straight back out.
A rule-based engine checks each transaction on its own. Every transfer is under the threshold, so every one passes. The problem is that the pattern is the crime, and a rule engine never sees the pattern.
Omniguard looks across accounts and across time. The anomaly and clustering modules see the shape: many small inbound transfers, multiple accounts pointing at one beneficiary, money in and straight back out, all sitting just below the threshold. It pulls them together as one coordinated case.
In a 90-day pilot with a Nigerian fintech: the old rule-based engine flagged nothing. Omniguard surfaced 6 mule and structuring networks [coordinated fraud networks], 240 linked accounts, ₦310 million ($230,000) moving through them in structured below-threshold amounts, and generated 6 suspicious activity reports ready for filing. If this got flagged by regulators, it’ll have meant tens of millions in potential fines, plus licence risk.
Boluwatife, you’ve called the sales process difficult. What’s the actual objection you run into?
I’ll be honest. I don’t think most of these financial institutions actually want to solve the fraud problem. I think they just want to be CBN-compliant. NIBSS publishes fraud data, and I remember a point in 2025 where the number was up something like 13% quarter on quarter, and almost nobody reacted to it. Most institutions aren’t treating fraud as urgent until a regulator forces the issue.
The second problem is that they’ve already paid for whatever they’re using, sometimes locked in for years. Ripping that out, even when it’s clearly the source of their losses, is a real organizational fight. People here do not like change. I’ve never seen resistance like it. Even when the existing solution is obviously the problem, closing a deal with a bank still takes around six months.
So how did you actually start converting?
Fintechs first, because they move faster than banks. And we changed the pitch. We stopped telling institutions to rip out what they had. Now the offer is: we integrate with whatever you’re already running. That single shift did more for our conversion than anything else. We also started white-labeling Omniguard to KYC providers, so they can offer it as an extension of their own service without building it themselves.
We are onboarding commercial banks now, a few MFBs, and some white-label partners. We had to introduce tiered pricing once we realised charging a commercial bank and a microfinance bank the same rate doesn’t work, especially after a competitor’s flat low-cost offering forced our hand on that.
What does pricing look like in practice?
For commercial banks, there’s a one-time setup and configuration fee, around twenty thousand dollars, covering staff training and onboarding. Then a recurring monthly fee that scales with transaction volume, in the range of 2.5 million naira for a commercial bank. MFBs sit on a different, more accessible scale.
The CBN mandates have worked in your favour. Is there a version of this that turns against you?
Not one we’ve hit yet. We were compliant before the mandates landed, so that wave has been entirely upside for us. The May 2025 directive gave institutions 12 months to comply after the final framework was published. That clock is running now. The real pressure now is speed of conversion. The market’s getting crowded. KYC providers are starting to claim they do fraud management and AML/CFT Compliance and reporting too. Some genuinely do. Some are stretching the term. Either way, it means we have to be sharper about what we are and what we’re not.
We partner with KYC providers rather than compete with them. We’re not trying to become a KYC company. We sit after verification, watching what happens once some-kyc-company is already inside the system. That’s a different problem, and if anything, it’s getting harder as fraud gets more creative.
The market went exactly where we thought it would. We just need to close faster than it’s growing.