AI assistants such as Claude and ChatGPT can now connect directly to business systems through MCP, an open standard for giving an assistant controlled access to another application. For a telecoms reseller, that includes the billing platform: an assistant can answer questions about revenue, collections and individual accounts from live data, and with the right permissions carry out routine tasks.
Billing data is among the most sensitive information a reseller holds, so the value of such a connection depends on how access is controlled. This post covers what an assistant connected to billing can do, the risks of a poorly designed connection, and the questions to ask any billing provider that offers one.
Key Takeaways
- Through MCP, an assistant can query a billing platform directly instead of working from copied data
- Useful tasks include reporting on revenue, collections and aged debt, and retrieving a customer’s account history
- The main risks are over-broad access, shared credentials, unrecorded changes and connections that cannot be revoked
- Each connection should use one person’s permissions, be approved with 2FA, keep write access separate and off by default, and be logged
- SAFE’s AI assistant connection is live, built on those controls
Key terms in this article
What is MCP?
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to business software in a controlled way. Instead of data being copied into a chat window, the assistant queries the system directly, within the permissions it has been given.
What is OAuth?
OAuth (open authorisation) is the standard way web applications grant each other access without sharing passwords, as used when a site offers “sign in with Google”. A well-designed AI connection uses it so that no password or API key is handed to the assistant.
What an Assistant Can Do With Billing Data
Many questions asked of billing data are one-off: who owes the most, how last month’s collections went, which customers have grown or reduced their services. In most businesses, each needs a report or an export prepared by someone who knows where the data is.
An assistant connected to the billing platform answers them on request, from current data, and can take a follow-up question about the same figures. It can also retrieve a customer’s account, invoices and payment history when someone is answering a billing query. With write access, it can carry out routine tasks such as emailing an invoice or generating a payment link.
The Risks of a Poor Connection
Connecting an assistant casually to billing data creates specific risks:
- Over-broad access: an assistant that can see every customer’s data regardless of who is using it.
- Shared credentials: a single company-wide key placed in chat tools, which cannot be traced to a person or revoked cleanly.
- Unrecorded changes: an assistant that edits records without confirmation or a record of what it did.
- No way to remove access: connections nobody can list, review or disconnect when a member of staff leaves.
There is also the question of where the data goes. Responses are produced by the assistant provider’s service, so the data in each conversation is processed by a third party, and that needs considering under your data-protection obligations.
Questions to Ask a Billing Provider
If your billing provider offers AI assistant access, these questions show how carefully it has been designed:
- Whose permissions does the assistant use? It should use those of the individual who connected it, never a shared account. Someone who cannot see dealer margins in the platform should not see them through an assistant.
- How is the connection approved? Look for a proper sign-in with 2FA using OAuth, rather than an API key pasted into the assistant.
- Is write access separate, and off by default? Reading a report and raising a refund are different levels of authority, and changes should need a deliberate grant.
- Is there an audit trail? Each question and action should be logged, marked as assistant activity and traceable to a named person.
- Can access be revoked? Each person should be able to disconnect their own assistant, and an administrator should be able to switch off access for the whole company.
A provider that cannot answer these clearly has not finished the feature, however well it demonstrates.
AI Assistant Access in SAFE
SAFE’s AI assistant connection is live and works with Claude on claude.ai and ChatGPT on chatgpt.com. Each connection belongs to one person, who approves it with their login and a two-factor code. Reporting, reading records and making changes are granted separately, with changes off by default. Actions go through the platform’s own workflows, significant ones need confirmation, everything is recorded in the audit trail, and access can be revoked by the individual or switched off company-wide.
With it, staff can ask for collections, aged debt, growth figures or a customer’s history in plain English, and, where write access has been granted, email an invoice or generate a payment link. The AI assistants service page summarises the feature, the API access page covers code-based integration, and the product site has the technical detail.
Should Resellers Act Now?
Billing runs perfectly well without an assistant, and a cautious approach is reasonable. MCP is, however, an open standard supported by the major AI providers, and business software is adopting it, so the question of how assistants connect to billing data is likely to come up whichever provider you use. The questions above apply equally to any of them.
To try the connection on your own platform, or to discuss how it would fit your team’s work and data-protection policies, contact us.