One agent, and the systems it reads and writes
The shape of it
The platform is one agent and three places it works. Roy is ROIAI One's AI agent for retailer deductions: it reviews every deduction that hits your account, builds the evidence that proves it invalid, drafts the retailer-specific dispute, and files it. It reaches the records that decide a case in your ERP, in the mailboxes where retailer notices land, and in the retailer portals themselves. It reads broadly and writes narrowly, and ROIAI One's own analysts review and release what it prepares.
What Roy does with a deduction
Roy works a deduction in four steps: detect it, pull the evidence, build the dispute packet, and submit it. Done by hand, every deduction is worked one at a time, a different portal, reason code, and evidence pack for each retailer, and most never get filed before the retailer’s window expires.
Detect the deduction
Roy monitors every retailer deduction as it lands, classifies the reason code, and flags the ones worth disputing.
Pull the evidence
Roy matches each chargeback against your invoices, bills of lading, and proof of delivery, assembling the documentation that proves the deduction is invalid.
Build the dispute packet
Roy writes the retailer-specific dispute, formatted to that retailer's reason codes and evidence rules, ready for review.
Submit and track
Packets are filed to the retailer portal or emailed to the right contact, and tracked until the money comes back.
Where Roy works, and what it does there
Roy reads broadly and writes narrowly. An ERP connection is not the only way in: chargebacks arrive by retailer portal, by email, and by EDI, and Roy works from those channels too.
The outer boundary is the permission set your IT team grants the login. Scope that login read-only and none of it can write. There is no data migration, no rip-and-replace, and no change to how your AR team works.
Who checks what gets filed
Roy reviews each deduction, assembles the evidence and prepares the dispute, and ROI-AI’s own analysts review and release it. That review is selective and exception-based, and the scope Roy handles grows as accuracy is established reason code by reason code. This is our AI-native Service-as-a-Software model: the review burden is ours, not yours.
Nothing lands in your queue and nothing waits on your approval.
Roy is also not always a yes. Where the evidence says the charge is valid, Roy says so, and where our published guidance does not yet cover a retailer’s specific rule or window, we say so rather than guess. The security and access overview covers how that access is scoped in practice.
The two platform pages, and what each answers
What is the agent, and how does it decide?
The agent itself: what Roy does with a deduction, how it decides which charges are worth disputing and when to recommend not disputing at all, what it reads, where it works, who checks it, and what it does not do.
ReadERP IntegrationWhat does connecting actually require of us?
The connection: what Roy reads and what Roy writes, record by record, what connecting does not require of you, what happens when the ERP cannot be connected, and what your side actually has to do.
ReadFor what the platform is pointed at rather than how it is built, the solutions pages route by problem, and the retailer intelligence library covers the channel, evidence, and reason codes for a specific retailer.
See it against your own deductions
The platform is easier to judge on your own book than in the abstract. We look at your deduction history and tell you what is worth recovering.