Page 37 - gs260702

Insights and Expertise

When AI pricing then become embedded automation. Another workflow
could start with a customer support team and then
becomes a payments radiate across underwriting, risk, servicing and portfolio
operations. This is where economics starts to have more
impact. If token spend grows faster than user count, agentic
operations challenge workflows are deployed in production, or AI moves from
basic chat into merchant-facing or risk-related automation,
the organization crosses into a whole new cost category.

The risk is not that AI costs rise but rather that the most
useful AI workflows are the ones most likely to become
high-volume, repeatable and expensive under the wrong
pricing model. The more useful the workflow becomes,
the harder it will be to predict its cost. This is not an
encouraging incentive for acquiring banks and processors.
Payments organizations need AI for daily tasks to improve
speed, consistency and decision support. Merchant
onboarding, underwriting support, compliance review,
chargeback documentation, support triage and portfolio
analysis are all logical candidates. And these regular tasks
are also exactly the kinds of workflows where repeatability,
volume and operating cost shouldn’t be throttled.
By David Moscatelli
Go Abacus Understanding AI pricing in payments

cquiring banks and processing organizations David Moscatelli's article focuses on a cost many payments
live in a world of volume, risk, margin pres- professionals may not yet encounter directly: the fees
sure and service-level expectations. They are companies pay to use commercial AI models.
A designed to support merchant growth while
managing a complex web of fraud risks, card network Most organizations do not build their own AI systems from
rules, regulatory obligations, bank partner expectations scratch. Instead, they purchase access to large language
and operational costs. Payment companies already handle models and related AI services from technology providers.
enough unpredictable moving parts; AI pricing should Those providers typically charge either by consumption,
not become another variable they cannot forecast. based on the number of tokens (pieces of text) the AI
processes, or through fixed-capacity arrangements that
AI pricing tends to look quite manageable as long provide predictable computing resources for a set price.
as it stays within a controlled pilot program. But the
economics change when it expands into the routine For acquiring banks and payment processors, AI is
daily work of merchant onboarding, underwriting, fraud increasingly being used behind the scenes rather than
review, chargebacks, compliance, support and portfolio in customer-facing chatbots. It can assist with merchant
monitoring. Once AI starts touching those workflows, onboarding, underwriting, fraud investigations, chargeback
consumption-based pricing becomes part of the operating documentation, compliance reviews and customer support.
economics.
As AI becomes embedded in thousands of routine
During early testing, consumption-based pricing can operational tasks, usage can increase dramatically. Under
appear easy to control. A small team asks a limited set of consumption-based pricing, every document analyzed,
questions; the use case is narrow. Once fully integrated, transaction reviewed or report generated adds to the bill,
though, production works on a different level. While a making monthly costs difficult to predict.
merchant-risk review, chargeback packet or compliance
question appears to the user as a single task, the AI is Moscatelli advises that organizations should match pricing
likely reading multiple documents, pulling records, models to workloads. High-volume, repetitive operational
summarizing evidence, and moving through several tasks may be better suited to fixed-cost AI capacity, while
reasoning steps behind the scenes. premium consumption-based services remain appropriate
Cost profiles change with workflows for specialized work requiring the most advanced models.

That is the awkward part of token pricing in payments: The goal is to gain AI's operational benefits without creating
volume does not always translate into a clean, predictable unpredictable expenses that grow alongside business
cost curve. The cost profile changes with the workflow. success.
One workflow might start as a simple chat interface but
37