What Fraud Detection Software Really Costs a Bank With 1.5 Million Accounts
A hypothetical five-year model for a 1.5-million-account bank puts the full cost near $14.6M. Licenses are only about a third of it. Analyst headcount is the line most buyers underprice.
Why the license quote is the smallest number you'll negotiate
A vendor's quote for a fraud platform usually arrives as one annual figure: a license band, a few channel modules and a services estimate. Procurement compares that number across bidders. For a bank with around 1.5 million accounts, it's often well under 40% of the five-year program cost. Most of the rest pays for the analysts who work the alerts, and for the people who tune the models so there are fewer alerts.
That's why fraud detection software pricing can't be judged from the license alone. A platform that costs $150,000 a year more but cuts false positives meaningfully can be the cheaper choice once headcount is counted. A cheap platform that floods the queue can cost the most.
Below is a hypothetical five-year cost model for a bank of this size. It isn't based on any real institution or vendor price list. It uses typical 2026 market ranges that vary widely by contract, region and negotiating position, so check every figure against real quotes.
The hypothetical bank
- Accounts: 1.5 million (current, savings, card and loan accounts) across about 1.1 million customers.
- Digital users: roughly 700,000 active in online or mobile banking in a given year.
- Monitored events: about 120 million a year across card authorizations, domestic transfers, instant payments, wires and high-risk digital events such as payee changes and logins from new devices.
- Deployment: vendor-hosted SaaS, replacing an older rules engine.
- Operations: a 24/7 card fraud queue plus business-hours queues for payments and scams.
Banks with 1.5 million accounts can look very different. A card-heavy bank may score three times as many events; a savings-led bank far fewer. The method still holds. Swap in your own volumes before trusting any total.
How vendors meter a bank of this size
Licensing cost follows the metric the vendor charges on, and most will quote more than one if you ask. The metric matters more than the headline rate, because each grows at a different speed as your business changes.
| Pricing metric | Typical 2026 range (indicative) | Hypothetical annual cost for this bank | What makes it expensive |
|---|---|---|---|
| Per account per year (core transaction monitoring) | $0.25–$0.75 per account at this scale | $375k–$1.1M | Counting dormant or closed accounts; account growth after a merger |
| Per scored transaction or event | $0.002–$0.008 per event in volume bands | $240k–$960k | Instant payment growth; scoring logins and non-monetary events |
| Per active digital user (device and behavioral analytics) | $0.20–$0.60 per user per year | $140k–$420k | A loose definition of "active"; per-session overage clauses |
| Enterprise or platform fee (flat, with volume ceiling) | $500k–$1.2M per year for a bank this size | $500k–$1.2M | Volume ceilings set too low; true-up penalties |
These are indicative ranges, not quotes. Confirm them with vendors. Large incumbents such as FICO, NICE Actimize, SAS and ACI Worldwide tend to sell enterprise or account-band licenses. Cloud-first vendors such as Feedzai and Featurespace (now part of Visa) more often quote on volume or platform tiers. Device and behavioral specialists such as BioCatch and LexisNexis ThreatMetrix usually price per user or per session. These are tendencies, not rules, and any of them will build a custom structure for a deal this size.
Per-transaction fraud scoring fees and the instant payments problem
Per-transaction fees look attractive when volumes are steady. They stop looking attractive when instant payments grow quickly, as they have in many markets: UPI in India, Pix in Brazil, Faster Payments in the UK, SEPA Instant in the euro area, FedNow and RTP in the United States. If 30% of your events are instant payments growing at double digits, a per-event contract reprices itself upward every year without anyone signing anything. Ask for volume bands with a per-event price that steps down as you grow, and a cap on annual true-ups.
Channel add-ons
Core monitoring usually covers cards and standard payments. These modules are commonly sold separately:
- Authorized push payment (APP) scam detection for instant payments. Close to essential where reimbursement rules put the loss on the bank.
- Digital banking and device intelligence, including behavioral biometrics for account takeover.
- Check fraud and image analytics, which matters a lot in some markets and barely at all in others.
- Application and identity fraud at onboarding, often a different vendor entirely.
- Consortium data or network intelligence feeds, sometimes priced apart from the scoring engine.
In our hypothetical model the bank buys the APP scam module and digital banking analytics, and leaves check and onboarding fraud to existing tools.
The five-year model, line by line
Here's the full hypothetical fraud prevention total cost of ownership for the bank above. Figures are in thousands of USD, rounded. License lines include a 5% annual uplift, common in multi-year contracts unless you negotiate it out.
| Cost line (USD thousands) | Year 1 | Year 2 | Year 3 | Year 4 | Year 5 | 5-year total |
|---|---|---|---|---|---|---|
| Core monitoring license | 600 | 630 | 662 | 695 | 729 | 3,316 |
| Channel add-ons (APP scams, digital banking) | 340 | 357 | 375 | 394 | 413 | 1,879 |
| Vendor implementation and change requests | 800 | 75 | 75 | 75 | 75 | 1,100 |
| Internal project and platform team | 400 | 220 | 220 | 220 | 220 | 1,280 |
| Model build, tuning and data science | 250 | 150 | 150 | 150 | 150 | 850 |
| Data feeds, non-production environments, storage | 60 | 60 | 60 | 60 | 60 | 300 |
| Fraud analyst operations | 1,360 | 1,190 | 1,105 | 1,105 | 1,105 | 5,865 |
| Total | 3,810 | 2,682 | 2,647 | 2,699 | 2,752 | 14,590 |
That's about $14.6 million over five years, or around $1.95 per account per year. License fees (core plus add-ons) come to about $5.2 million, roughly 36% of the total. Analyst operations come to about $5.9 million, roughly 40%.
Implementation
The $800,000 vendor services figure assumes SaaS. It covers integration with card authorization, the payments hub and digital banking, plus a parallel run alongside the legacy engine. Typical 2026 ranges for a bank this size are $400,000 to $1.2 million for SaaS. On-premises or private-cloud deployments often run $1 million to $3 million and add $150,000 to $300,000 a year in infrastructure. The internal team line covers a project manager, integration engineers and fraud specialists pulled off their day jobs. Banks routinely leave it out of the business case.
Model tuning
Out-of-the-box consortium models get you started. The first-year $250,000 covers adapting them to your portfolio, rebuilding rule sets and running champion-challenger tests. After that, models drift as fraud patterns shift, so budget for ongoing tuning, either as vendor services or an in-house data scientist. Skip this line and a good platform turns noisy by year three.
Analyst headcount, the line that decides the business case
The model assumes an alert rate of about 0.1% on 120 million events: roughly 120,000 alerts a year, or 330 a day. An analyst clearing 45 to 55 alerts a working day handles roughly 11,000 to 14,000 a year. Add 24/7 coverage for the card queue, leave and team leads, and you reach about 16 people in year 1, at an assumed fully loaded cost of $85,000 each. That's plausible for Western Europe or North America and far too high for many other markets (see the next section). As tuning lowers false positives, the model drops to 14 people in year 2 and 13 from year 3.
Here's the sensitivity that matters. Each analyst you don't need saves about $85,000 a year in this scenario. A vendor that credibly gets you to 11 analysts instead of 13 saves about $510,000 over years 3 to 5, enough to cover a much higher license. Ask every bidder to model your alert volumes on your own historical data, not a reference customer's.
How quotes and cost structures change by region
The same platform produces very different totals depending on where the bank operates. Four things move the numbers most.
Labor cost changes which line dominates
In our model analysts are the biggest line because of the $85,000 loaded cost. Run the same 13 to 16 people at $20,000 to $35,000, a plausible range for many banks in India, parts of Southeast Asia and parts of Africa, and the license becomes the dominant line. Latin American and Gulf markets often fall in between, though Gulf pay for experienced fraud staff can match Western levels. Where labor is cheap, a premium for lower false positives is harder to justify, so push harder on license price. Where labor is expensive, do the opposite.
Data residency and hosting
Data-protection and banking secrecy rules shape deployment options. The EU's GDPR, India's Digital Personal Data Protection Act and RBI expectations on payments data storage, Saudi Arabia's PDPL, and banking regulators in markets such as Indonesia and Nigeria all affect where data can sit and how it moves. If a vendor has no in-country or in-region cloud, you may be pushed to private-cloud or on-premises deployment, which raises implementation and infrastructure costs. Check where scoring, model training and consortium data processing physically happen, not just where the primary database lives.
Liability rules decide which add-ons you can skip
Who pays for a scam loss differs by market. The UK's mandatory APP reimbursement regime makes scam detection on outbound instant payments close to non-negotiable for UK payment firms. EU instant payment rules, including verification of payee, push the same way. Brazil's Pix has its own return mechanism for fraud cases. Where customers carry more of the loss, banks sometimes defer the scam module. That saves money now and leaves a gap if the rules change. Have your compliance team or a licensed advisor confirm your exposure in each market.
Channel partners and currency
In much of the Middle East, Africa and parts of Asia, large vendors sell through local partners or systems integrators. That can add a margin to license and services, often hidden in a blended price, so ask for the split. USD-priced contracts expose non-dollar banks to currency risk across five years, and withholding taxes on cross-border software payments, which vary by country, can raise the effective cost further. Get quotes for your specific entity and currency.
Red flags in fraud platform quotes and how to negotiate them
Most overruns come from contract terms, not the headline rate. Watch for these in any proposal:
- Volume definitions that grow by themselves. If "event" includes logins, balance checks or every API call, billable volume can double without any growth in payments.
- Uncapped true-ups. Overages reconciled annually at list price, with no right to move into a higher band at the discounted rate.
- Escalators above 5% a year, or escalators tied to an inflation index with no ceiling.
- Tuning sold only as professional services, with no tools that let your own team adjust rules and thresholds.
- Fixed-price implementation that excludes integration with core banking, the card processor or the payments hub, which is where the real work is.
- No performance commitments. A vendor that won't model detection and false-positive rates on your data during evaluation probably can't hit the numbers in its pitch.
- Undefined exit costs. Find out what it costs to get your case history, labels and model features back in a usable format.
A five-step process for comparable bids
- Publish a volume schedule. Give every bidder the same five-year projections for accounts, digital users and events by channel, including instant payment growth. Ask for pricing under at least two metrics.
- Run a data-based proof of value. Provide six to twelve months of labelled historical data and have each shortlisted vendor report detection and alert rates at a fixed review capacity. This needs data-sharing agreements that meet your local privacy rules.
- Convert alert rates to headcount. Use your own analyst throughput and loaded cost. That turns a model-quality claim into a dollar figure.
- Build the full five-year table for each bidder using the cost lines above, including internal staff and tuning.
- Negotiate the terms, not just the rate: escalator caps, band step-downs, true-up pricing, a tuning allowance and exit provisions.
The decision rule: if analysts make up more than a third of your modeled five-year cost, pick the vendor with the lowest proven alert volume at your target detection rate, even at a higher license. If they make up less than a fifth, which is common in lower-cost labor markets, pick on license price and contract flexibility. Before the first vendor call, fill in your own volumes and loaded analyst cost in the table above.
Quick answers
Is per-account or per-transaction pricing cheaper for a mid-sized bank?
It depends on how fast transaction volumes grow relative to account count. Per-account pricing is more predictable when instant payments or card usage are rising quickly. Per-transaction pricing can be cheaper for savings-led banks with low activity per account. Ask for both and model them against a five-year volume forecast before choosing.
How much should we budget for fraud platform implementation?
Typical 2026 market ranges for a bank with about 1.5 million accounts are $400,000 to $1.2 million in vendor services for SaaS, and $1 million to $3 million for on-premises. Add internal staff costs, often $300,000 to $500,000 in the first year. These ranges vary a lot by contract and integration complexity, so confirm with vendors.
Why does analyst headcount matter so much in fraud prevention total cost of ownership?
Every alert the platform raises needs a reviewer, and in higher-wage markets five years of analyst salaries can exceed the license. A platform that produces fewer false positives at the same detection rate cuts headcount directly. That's why evaluations should measure alert volumes on your own data.
Do data residency rules change what a fraud platform costs?
Often, yes. If local rules require data to stay in-country and the vendor has no local hosting, you may need a private-cloud or on-premises deployment, which raises implementation and infrastructure costs. Rules differ across the EU, India, the Gulf states, Southeast Asia and Africa, so have compliance or legal counsel confirm the requirements for each market you operate in.
Written by the Smartnode Suite editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.