Adding an AI billing team to CIS Infinity, without replacing it

How utilities can put AI agents to work on top of CIS Infinity: a secure API layer, narrowly scoped agent tools, and a human review step before anything touches a customer account.

You don't need to replace CIS Infinity to put AI to work in your billing office. Build a secure API layer on top of it, give each AI agent a narrow set of tools that read data and draft work, and route every proposed change to a billing clerk for approval. CIS Infinity stays the system of record. The agents take on the prep work that eats your team's week.

Most utility billing teams we talk to have the same problem. CIS Infinity does its job: it holds every account, meter, read, rate, and payment. But the work around it is still manual. Clerks pull exception reports before every bill run, look up account history to answer "why is my bill so high?", re-key move-in and move-out requests from web forms, and chase the same data-quality issues month after month.

That surrounding work is where AI agents fit. Not as a replacement for the billing system, and not as an autopilot that changes accounts on its own, but as a team of assistants that do the legwork and hand finished drafts to the people accountable for the result.

What an "AI billing team" actually does

Think of it as a few narrowly focused agents, each with one job, rather than one general chatbot. These are the workflows where we'd start, because each is high-volume, repetitive, and easy for a person to check:

Pre-bill exception reviewerScans reads before the bill run, flags likely misreads and unusual consumption, and writes a short note on why each one looks wrong.
High-bill inquiry assistantPulls the account's usage history, rate, and recent reads, then drafts a plain-language reply for the customer service rep to send.
Start/stop service adminTurns move-in and move-out web requests into prepared service orders, with missing information called out, ready for a clerk to submit.
Data-quality monitorFinds likely duplicate accounts, missing contact details, and mismatched mailing addresses, and queues them for cleanup.

None of these agents decides anything about a customer's account. They prepare the work; your staff approve it. That distinction matters to residents, auditors, and council members, and it's what makes the system safe to run.

How it fits together

CIS Infinity system of record API layer auth · scopes audit log AI agents exception review inquiry drafts service orders data quality Human review queue approve · edit · reject Approve approved changes written back and logged
The agents never touch CIS Infinity directly. Everything goes through the API layer, and every change goes through a person.

1. A secure API layer on CIS Infinity

Everything starts with clean, controlled access to billing data. Depending on your version and hosting, CIS Infinity may expose APIs you can build on; where it doesn't cover what a task needs, we build that part of the layer ourselves. We've done this before: for a municipal utility, we built a FastAPI service on CIS Infinity that lets approved vendors check resident eligibility in real time. Lookups that used to take about two days now take under two seconds (read the case study).

The same layer becomes the only door the agents can use. It decides who can see which fields, and it logs every call.

2. Narrow tools, not open access

Each agent gets a short list of tools, and each tool does one thing: "get usage history for an account," "list reads outside the expected range for this meter," "create a draft service order." Read-only tools come first. Write tools come later, and they only create drafts.

This is the most important design choice in the whole system. An agent with broad database access can do anything, including the wrong thing. An agent with five narrow tools can only do those five things, and each one can be tested on its own.

3. A human review queue

Every proposed change lands in a queue where a billing clerk can approve it, edit it, or reject it. The clerk sees the agent's reasoning alongside the data it used, so most reviews are quick. Approved changes are written back through the API layer with a record of who approved them and when.

The guardrails that make it safe

  • Agents draft; people decide. No agent changes a balance, a rate, a service status, or a payment arrangement on its own.
  • Least-privilege data access. Each tool returns only the fields its task needs. The inquiry assistant doesn't need bank details, so it never sees them.
  • A full audit trail. Every agent call, every draft, and every approval is logged, so you can answer "who changed this account, and why?" months later.
  • Tested against history. Before going live, we replay past bill runs and inquiries through the agents and compare their drafts with what your staff actually did.
  • Governance up front. Anything that would make an automated decision about a customer's account is off-limits by design, not by policy reminder. Our free AI Use-Case Scorecard shows the review each idea needs before launch.

Where to start

  1. Pick one workflow. Choose the one that eats the most staff time and is easiest to check, usually pre-bill exception review or high-bill inquiries.
  2. Measure today. Count how many items your team handles and how long each takes. That's your baseline.
  3. Build the read-only API and one agent. Run it alongside your team, drafting only, and compare its work with theirs.
  4. Add the review queue and go live. Once the drafts are consistently right, let clerks approve them directly.
  5. Expand one workflow at a time. Each new agent reuses the same API layer and review queue, so the second one is much faster to build than the first.

Running CIS Infinity and buried in manual billing work? We'll map one workflow with your team and show you what an agent for it would look like, including the review steps.

Book a consultation →

FAQ

Common questions

Do we need to replace or upgrade CIS Infinity to use AI agents?

No. The agents work through an API layer on top of CIS Infinity, so the billing system stays your system of record. Where the version you run doesn't expose the data a task needs, we build that part of the API layer ourselves.

Will an AI agent change customer accounts on its own?

No. Agents read data and draft work: a flagged read, a suggested reply, a prepared service order. A billing clerk approves, edits, or rejects it, and only then does the change reach the account.

Which billing workflow should we automate first?

Pick one that is high-volume, repetitive, and reviewable, such as pre-bill exception review or drafting replies to high-bill inquiries. Measure how long it takes today, then pilot the agent on that one workflow before adding more.

How is customer data protected?

Each agent gets only the fields its task needs, through authenticated API calls that are logged. The agent never has direct database access, and every action it proposes is recorded with the person who approved it.

Let's talk

Have a manual process slowing your team down?

Tell us what your team copies, re-keys, or emails around every week. We'll show you what it would take to automate it.