How to Build an AI Collections Workflow Without Writing Code
An AI collections workflow reads open invoices from your ERP, drafts reminders, sorts customer replies and logs every step. How to build one without code.

An AI collections workflow is an automation that reads overdue invoices from your ERP or accounting system on a schedule, decides what each account needs next, drafts and sends the reminder, reads the customer's reply, and records the outcome, with an AI agent handling the steps that need judgment. You can build one without code in a visual workflow builder, on top of the ERP you already run.
The part that needs AI is narrow. Days overdue, amounts and dunning levels are rules, and rules should stay rules. The AI agent earns its place where a person used to read: a customer's email that says "we paid last week", "the price on line 3 is wrong" or "send it to our new AP address".
Where engini fits: engini's Workflow Studio is a visual builder where a collections workflow reads open invoices from NetSuite, QuickBooks, SAP Business One, Priority or a database, applies your rules, and uses an AI agent step to draft reminders and sort replies. It sends through Gmail or Outlook, writes status back to the ERP and keeps a log in Tables, all without code. See the AI Agent step in the engini docs.
Key Takeaways
- An AI collections workflow combines fixed rules (who is overdue, by how much, which step comes next) with an AI agent for the reading and writing a collector does.
- ERPs already send reminders: NetSuite's Dunning Letters SuiteApp and QuickBooks Online's automatic reminders handle schedule-based dunning.
- What they do not do is read and act on the customer's reply; that is where an AI agent adds the most.
- You can build the workflow without code: a schedule trigger, an ERP read, conditions, an AI agent step, an email action and a log.
- Start with one segment and have the agent draft before it sends, then widen the scope as the drafts hold up.
What Is an AI Collections Workflow?
An AI collections workflow is a scheduled or event-driven process that moves overdue invoices toward payment, using an AI agent for the steps that require reading or writing in plain language.
A typical one does five things:
- Finds the work. Reads open, overdue invoices from the ERP each morning.
- Decides the next step. Applies your rules: days overdue, balance, customer segment, last contact.
- Writes to the customer. Drafts a reminder that names the invoices, amounts and due dates, in your tone.
- Reads the reply. Sorts it into promise to pay, already paid, dispute, wrong contact or no reply, and extracts dates and amounts.
- Records the outcome. Writes a status back to the ERP or a log, and sends disputes to the person who owns them.
The difference from a plain reminder schedule is step 4. A schedule sends the next letter whether or not the customer already answered.
What Do ERPs Already Do for Collections?
Most ERPs and accounting systems already automate schedule-based reminders, so an AI collections workflow should add to them, not repeat them.
NetSuite. The Dunning Letters SuiteApp lets you create dunning procedures and assign them to customers, invoices or invoice groups. Dunning levels set thresholds for overdue amounts and days overdue and the letter template for each level, and an administrator schedules the dunning evaluation workflows. Depending on your preferences, NetSuite emails the letters automatically or queues them for you. Oracle's limitations page says the SuiteApp requires the Multiple Currencies feature and cannot detect bounced email messages.
QuickBooks Online. Automatic invoice reminders send up to three reminders, scheduled up to 90 days before or after the due date, for invoices you have already emailed. On QuickBooks Online Advanced, you can use workflows to create invoice reminders.
Tip: If your ERP's dunning already runs well, keep it for the reminders and build the AI workflow around the replies. That is the smallest change with the clearest gain.
How Do You Build an AI Collections Workflow Without Code?
You build it as one workflow with a trigger, a few steps that read and decide, an AI agent step, and steps that act and log. Each step is configured in a visual builder.
Step 1: Pick one segment and write the rules
Choose a slice you can watch closely: for example, invoices 15 to 60 days overdue for customers under a set balance. Write the rules as a short table: which days-overdue band gets which message, who owns disputes, and which accounts are never contacted automatically, such as key accounts or customers in a payment plan.
Step 2: Connect the systems
Connect the ERP or accounting system that holds the invoices, the mailbox that sends reminders, and the chat tool where disputes go. In engini, each connection is one signed-in account. Use an ERP account whose role can read invoices and customers and update only the fields the workflow needs.
Step 3: Read the open invoices on a schedule
Start the workflow on a schedule, for example every weekday morning. The first step reads open invoices from the ERP: on NetSuite, engini's activities get one or more records; on a database, a query or stored procedure returns the list. Group invoices by customer so each customer gets one message.
Step 4: Apply the rules before the AI
Use conditions to sort each customer into a band, skip excluded accounts, and stop if someone contacted the customer in the last few days. Keeping these as rules makes the workflow predictable and easy to audit; the AI agent never decides who is overdue.
Step 5: Add the AI agent step
The AI agent step gets the customer's invoices, the band, the last few emails and your instructions, and returns a draft. Give it a clear brief: your tone, what it may offer (for example a copy of the invoice or a payment link), and what it must never do (change amounts, promise discounts, threaten).
Step 6: Send, log and write back
Send the message through Gmail or Outlook from a shared collections address. Log every message in a table: customer, invoices, band, date and the draft that went out. Write a status or next-contact date back to the ERP so the collector sees it where they work.
Step 7: Handle replies in a second workflow
Start a second workflow when an email arrives at the collections address. An AI agent step reads the reply and returns a category and the details: a promised date, a remittance reference, a disputed line, a new contact. Then the workflow acts on the category:
- Promise to pay. Records the date and pauses reminders until it passes.
- Already paid. Sends the remittance details to the person who applies cash.
- Dispute. Posts the case to Slack or Teams with the invoice and the customer's words, and pauses reminders.
- Wrong contact. Updates the contact and resends.
Note: Start in draft mode. Have the workflow save each draft to the table or post it to the collector's channel, and let the collector send it. When the drafts consistently match what the collector would write, let the workflow send for that segment.
What Are the Best Practices for AI Collections Workflows?
The best practice is to keep money decisions in rules and people, and give the AI agent the reading and writing.
- One message per customer, not per invoice. Group open invoices so the customer sees the full picture.
- Never contact a disputed account automatically. A dispute pauses reminders until a person clears it.
- Give the agent only the tools it needs. It should read invoices and draft email; it does not need to create credit memos.
- Log the draft and the reply. Anyone reviewing an account should see what was sent and what came back, in one place.
- Measure on outcomes. Track replies sorted correctly, disputes routed the same day, and promises kept, not the number of emails sent.
- Exclude what needs a human voice. Key accounts, payment plans and legal holds stay with the collector.
How Do You Build a Collections Tracker Without IT Help?
You can build a collections tracker as a table in a no-code tool and let the workflow fill it, so the accounting team sees every account's status without asking IT for a report.
In engini, Tables hold typed records, and a record created, updated or deleted can start a workflow. A collections table might have one row per customer with balance, band, last contact, reply category, promised date and owner. The scheduled workflow updates the rows; a collector changing the owner or status can start a workflow too, for example to pause reminders. Apps, engini's no-code app builder, can put a simple list and detail screen on top of the table for the team.
How Does engini Run an AI Collections Workflow?
engini runs the whole workflow in Workflow Studio, its visual builder, on the same connections its developers' agents use, with every run recorded.
- Systems. ERPs and finance systems such as NetSuite, SAP S/4HANA, SAP Business One, Oracle Fusion Cloud ERP, Priority and QuickBooks; CRMs such as Salesforce and HubSpot; Gmail and Outlook; Slack and Teams; databases such as SQL Server, Oracle Database and IBM DB2. Every integration runs both ways, so the workflow reads invoices and writes status back. See the integrations page.
- Legacy and on-premises systems. If invoices live in an on-premises ERP or database, as they often do in banks, credit unions and companies on AS/400, the on-prem agent reaches them from inside your network with no inbound ports opened.
- Your own setup. Custom fields and custom objects are read live, so a "collections status" field you added to the invoice record is available to the workflow.
- The AI agent step. An agent inside the workflow, with the model you connect (Anthropic, OpenAI, Gemini or Azure OpenAI), optional memory and a set of tools.
- Triggers. A schedule, an email received, a record changed in a connected app or in a table, or a webhook.
- Record. Run history shows each run, and every run is recorded under the user who triggered it. Your agent never needs your credentials: engini stores them and adds them when it runs each call.
For the bigger picture of accounts receivable, see AI for accounts receivable; for the cash application step that follows a payment, see the real cost of manual cash application and NetSuite automated cash application. The order-to-cash page covers the flow end to end.
When you do not need engini: if you have a small number of customers on QuickBooks Online and its automatic reminders do the job, keep them. If NetSuite's dunning covers your reminders and your collectors read replies comfortably, the gain is small. engini fits when replies, disputes and updates pile up across the ERP, email and chat, or when invoices sit in a system your current tools cannot reach.
Frequently Asked Questions
What is an AI collections workflow?
An AI collections workflow is an automation that finds overdue invoices, drafts reminders, reads customer replies and records the outcome, using an AI agent for the reading and writing.
How do I automate accounts receivable collections without replacing my ERP?
Build the workflow on top of the ERP. It reads open invoices through the ERP's connection, sends reminders by email, and writes statuses and promised dates back to the same records, so the ERP stays the system of record.
What are the steps to build a collections workflow?
Pick a segment and write the rules, connect your systems, read open invoices on a schedule, apply rules, add an AI agent step to draft, then send, log, write back and handle replies in a second workflow.
Can AI read and sort customer replies to collection emails?
Yes. An AI agent step can classify a reply as a promise to pay, already paid, dispute or wrong contact, and extract dates, amounts and references so the workflow can act on them.
Should the AI agent send collection emails on its own?
Not at first. Start with the agent drafting and a collector sending, then let the workflow send for a segment once the drafts consistently match what the collector would write.