AI Agents Are Starting to Log Into Customer Portals

The Bottleneck Hiding After the Invoice Is Created

For years, finance automation has concentrated on creating invoices, sending reminders, applying cash and reporting on receivables. Yet for many B2B companies, an oddly manual step remains between “invoice ready” and “invoice accepted by the customer”: somebody has to log into the customer's procurement portal.

The AR employee may need to find the correct purchase order, enter invoice details, upload supporting documents, satisfy buyer-specific fields, submit the invoice and then return later to find out whether it was accepted, rejected, disputed or paid.

Multiply that workflow across customers using Coupa, SAP Ariba and other portals and a supposedly digital process starts to resemble manual data entry spread across dozens of websites.

On March 27, 2025, Tesorio launched its Supplier Portals Agent, an AI agent designed specifically to attack that workflow.

It is a useful product launch to watch not because supplier portals are glamorous, but because they illustrate where agentic AI may first create value in finance: repetitive work that requires software to navigate systems, gather context and take a bounded action.

What Tesorio Actually Launched

According to Tesorio's launch materials and product documentation available at the time, the Supplier Portals Agent is designed to automate portal-based invoice management.

For supported workflows, it can match invoices to purchase orders in Coupa and Ariba, perform pre-submission checks, verify items such as payment terms and available PO balance, submit invoices and centralize issue handling.

Its monitoring capability is intended to bring portal status information back into Tesorio so AR teams can follow what happens after submission rather than repeatedly checking separate portals.

Max Dame, Tesorio's vice president of finance and operations, described the idea to CFO Dive as an agent that goes into portals for customers, submits invoices and then monitors the submission and approval process.

The agent is not simply answering a question about receivables. It is being asked to perform part of the receivables workflow.

Why Portals Became an AR Problem

Supplier portals solve a real problem for buyers. They standardize purchasing and invoice intake, enforce required information and connect invoices with procurement approvals. But the efficiency can be asymmetric.

Tesorio said at launch that portal-based invoices can account for as much as 70% of invoice volume for some businesses. That figure is a vendor-reported observation, not an industry-wide benchmark.

A rejected invoice can delay the payment process. A missing PO, incorrect field or unnoticed dispute can sit outside the AR team's normal workflow until somebody checks the portal. That makes portal visibility a working-capital issue, not merely an administrative inconvenience.

The First Customer Signals

Tesorio's launch announcement included early customer reports from Plaid, Fastly, Pendo and Workiva.

Fastly Payments Operations Analyst Carla Buggs said the company gained visibility into thousands of portal invoices and reported that manual portal-management time fell by more than 75%.

Plaid Accounts Receivable Lead Jack Spencer described having a unified dashboard for portal invoice statuses as a significant improvement to collections.

Pendo Senior Credit & Collections Manager Melissa Zmuda said the product turned a previously manual portal process into an automated workflow that saves hours each week.

These are customer statements supplied in connection with a vendor launch, so they should not be treated as independent benchmark research. But they provide useful evidence of the operational problem the product is targeting.

From Software That Recommends to Software That Acts

Traditional AR software may identify which account deserves attention or recommend the next collection action. An agentic workflow goes one step further: software performs a sequence of actions toward a defined objective.

In the supplier-portal example, that sequence can include identifying the invoice, locating the relevant PO, validating submission requirements, entering information into the external portal, submitting, capturing the result, monitoring status and surfacing an exception when human intervention is needed.

The realistic near-term model for finance is not unlimited autonomy. It is bounded autonomy: software acts within a defined process, while humans remain responsible for exceptions, controls and decisions outside the expected path.

What an AR Team Should Check Before Automating Portals

1. 01Portal volume — How many invoices actually require portal submission, and which customers create most of the workload?
2. 02Rejection reasons — Identify why invoices are rejected today. Automation should not accelerate bad inputs.
3. 03PO quality — If PO numbers, balances or terms are frequently incorrect upstream, the portal agent will encounter those problems downstream.
4. 04Credential security — An agent that logs into external procurement systems needs controlled credentials, appropriate authentication and clear access ownership.
5. 05Action logging — The team should be able to determine what the agent submitted, when it acted and what happened afterward.
6. 06Exception ownership — When the portal rejects an invoice or a PO is insufficient, somebody needs a clear task, context and escalation path.
7. 07Customer-specific rules — Large buyers often configure procurement systems differently. Supporting a platform does not mean every customer's workflow is identical.

The Product Is Smaller Than the Big AI Story — And That May Be the Point

Much of the 2025 conversation about AI in finance is focused on ambitious concepts: autonomous finance, AI copilots and digital workers.

Supplier-portal automation is much narrower. That is precisely why it is interesting.

Finance teams do not need every AI project to reinvent the function. A product can create value by removing a stubborn operational task that sits between existing systems.

The best test is not whether an agent looks intelligent. It is whether the process becomes measurably better: fewer manual portal logins, faster successful invoice submission, fewer preventable rejections, shorter time to identify disputes, less AR employee time spent checking status and better visibility into invoices awaiting buyer approval.

What This Means for AR Professionals

For an AR analyst, this kind of automation shifts attention away from portal administration and toward exceptions.

Instead of asking “Did we upload the invoice?”, the analyst can spend more time asking: Why was it rejected? Why has approval stalled? Is the PO exhausted? Who at the customer needs to resolve the issue? Which delayed invoice has the largest cash impact?

Automation does not eliminate the work around exceptions. It makes the quality of exception management more important.

Product Watch — What We Would Watch Next

Tesorio's Supplier Portals Agent is new. The relevant questions over the next year are practical:

• How many portal configurations can it handle reliably?
• How does it respond when a buyer changes its portal workflow?
• How frequently does a human need to intervene?
• How are credentials and MFA handled?
• How complete is the audit trail?
• Can automated submission measurably reduce invoice-to-cash delays?
• How does performance differ between standardized and unusual portal configurations?

Those answers will determine whether portal agents become a niche convenience or a standard layer in enterprise AR automation.

The Larger Signal

Finance software is moving from systems that store transactions, to systems that analyze transactions, and increasingly toward systems that can perform bounded operational work.

Supplier portals are an unusually clear example because the task is visible: somebody used to log in and do it manually.

If agents can reliably absorb that work, the AR team does not disappear. The queue changes.

Humans spend less time operating portals and more time resolving the financial exceptions that prevent revenue from becoming cash.