By the time an at-risk customer shows up in your numbers, the quarter has closed and the overdue invoice is already headed for write-off. Collections get chased when someone has time, so days sales outstanding (DSO) creeps up while the warning signs sit in two different systems.
What if one agent read your billing data and your product usage, then handed you a ranked list before your next forecast call? The agent in this guide does exactly that: it names the riskiest accounts, why each one is flagged, and the next step worth taking. You can build it in AgentSpot in five steps without writing a line of code.
What Does an At-Risk Customer Agent Do?
An at-risk customer agent answers the question your CFO asks before every forecast review: which accounts could cost you money next quarter, and why? It pulls payment signals and account signals into one view, then scores every account on combined risk. Each time it runs, here's what lands in front of you:
The 10 accounts with the highest combined risk, in ranked order.
The single biggest reason each account was flagged.
One recommended next step: a check-in call, a payment follow-up, or an executive escalation.
Why combine both kinds of signals? A late payer who uses your product daily is a collections conversation, while an on-time payer with stalled usage is a renewal one. When both show up on one account, you're looking at the churn and forecast surprise you want to see early.
Which Signals Should Your Agent Combine?
The agent does three jobs, and each one maps to a system you already run:
Pull payment signals, such as overdue invoices, failed payments, and repeat lateness, from a billing system like Stripe.
Pull account signals, such as annual recurring revenue (ARR), renewal date, and usage trend, from your customer relationship management (CRM) platform and a ThoughtSpot Model.
Rank every account by combined risk so the accounts that need you most rise to the top.
Taken alone, each signal tells you only part of the story: repeat lateness might point to a slow accounts payable (AP) department, and a usage dip might be a holiday. What happens when you put both next to a renewal date 60 days out? The pattern gets much harder to explain away.
How Do You Build Your Risk Agent?
Building it takes five steps in the AgentSpot builder. Since nothing saves until you click Save Agent, you can try things freely along the way.
Step 1: Open the AgentSpot Builder
From AgentSpot home, go to My Agents, where you'll see two paths: Create new lets you describe the agent in plain language, and Create manually lets you set it up field by field. Which one should you pick? Either works, so go with whichever matches how you like to think through a problem.
Step 2: Write the Agent Instructions
Open the Instructions section and paste in the block below. While you're in that panel, you'll notice it also holds Skills and Conversation starters; you'll come back to those shortly.
``` You monitor customer health by combining payment and usage signals.
From the billing system, pull invoices that are open, past due, or uncollectible. Aggregate them by customer: total outstanding, oldest days overdue, and whether two or more invoices went late in the last 180 days.
From the CRM, pull each customer's ARR and renewal date.
From the ThoughtSpot Model, pull each customer's usage trend over the last 90 days.
Score each account on combined risk and rank the top accounts.
For each account, name the single biggest reason it was flagged (for example, "two late invoices in 90 days and usage down 40%"). Recommend one next step: a check-in call, a payment follow-up, or an executive escalation.
Deliver the ranked list to Slack or email, daily or weekly.
Never message or email a customer directly. If a customer-facing email would help, create a Gmail draft for a human to review and send. ```
A ranked list only helps if you can act on it quickly, so each account gets one reason and one next step. Want different windows than 180 days for lateness or 90 for usage? Adjust them to match how your business bills and renews.
Step 3: Connect Your Billing and CRM
In the Connectors section, click + and add your billing system (Stripe) and your CRM (Salesforce). If a connector you expect isn't listed, there's a simple reason: the + picker only shows connectors you've already connected. An admin first adds and shares the connector under Admin Settings > Connectors, and then each person connects it from the profile menu under Connectors.
Why start with Stripe? It already tracks invoice statuses like open and uncollectible, so the agent works from the same records your accounts receivable (AR) team uses every day.
Step 4: Attach a ThoughtSpot Model for Usage
Billing and CRM data tell you who owes what and when they renew, but they can't tell you whether anyone still uses the product. Where does that answer come from? In the ThoughtSpot Models section, click + and pick up to three Models that have Spotter turned on. Choose the Model that tracks product usage, so the agent gets a usage trend for each account.
Why does the trend matter? A 90-day trend shows you which direction an account is heading, even when its login count looks healthy right up until renewal.
Step 5: Save, Run, and Refine
Click Save Agent, then run it on accounts you know well. Does something look off? Correct it in plain words, the same way you'd coach a new analyst: "weight repeat lateness higher" or "drop accounts under $10K ARR."
Once you've made changes, run it again and compare the new list against the accounts already on your mind. If the top 10 include the accounts you were worried about, plus a few you hadn't spotted, your scoring is close.
What Should Your Team Ask the Agent?
Conversation starters give whoever opens the agent a ready-made first question. Add these three in the Conversation starters field you saw in Step 2:
"Which accounts are most at risk of churn this week, and why?"
"Show me every account with a renewal in the next 60 days and a usage drop."
"Which flagged accounts have ARR over $50K?"
Because AgentSpot is Multiplayer AI, you can share the agent with your Controller, your financial planning and analysis (FP&A) lead, and your collections team in one click. So what does each of them actually see when they run it?
What Keeps the Agent Inside Your Guardrails?
Finance can't run on guesswork, and an agent that touches receivables has to earn your trust before it runs unattended. Here's how the guardrails work:
The agent acts as the signed-in person running it, so it sees only the Stripe and Salesforce records that person can see.
Row-level and column-level security still applies to anything the agent pulls from a ThoughtSpot Model.
With a Google Workspace connector, the agent can send Gmail on its own, so your instructions decide whether it sends or drafts.
Every action is logged and replayable, so you can trace exactly what the agent did on any run.
In practice, the agent works inside access you already have, and no new permissions appear along the way. Doing customer-facing work? Add two lines to your instructions: never modify a CRM record, and route every customer message into a Gmail draft. That way, someone on your AR team reads every email before it goes out, and your CRM stays exactly as your reps left it.
When Should the Agent Become a Workflow?
Once you trust how the agent behaves, the next choice is how it runs. A chat agent runs when someone opens it and asks a question; a workflow is a different object in AgentSpot, a scheduled, multi-step pipeline that runs on its own. So which one do you need?
Why start with the chat agent? While you tune the scoring, you want to see every result and push back on it. Once the rankings look right, add the agent as a step in a workflow so the list arrives in Slack or your inbox on schedule.
Workflows are also where multi-agent systems start to make sense for finance. Say you also build an Accounts Receivable agent: both can run in the same workflow, where one finds the accounts and the other drafts the follow-up.
Splitting the job this way, sometimes called multi-agent analytics, gives each agent one slice of the analysis while the workflow passes results between them. What do you get out of multi-agent AI? Focus: each agent stays small enough to check, and every handoff still shows up in the log. For more ideas, the AgentSpot launch post names other finance agents, such as a Budget Anomaly Checker.
Watch AgentSpot Build Agents in a Demo
Whichever version you need, AgentSpot can build the chat agent or the scheduled workflow from the same plain-language description. The signals you feed it decide whether your CFO and customer success team trust it: payments, usage, and renewals, read under your own permissions.
Want to see an agent like this come together? Write down the three signals you'd check by hand for your riskiest account; then book a live AgentSpot demo and watch a real use case go from plain language to a working workflow.
