Executive summary
Scheduled payment reminders automate only half of B2B collections, because the replies customers send back, such as invoice copy requests, wrong PO numbers and disputes, still land in the accounts receivable (AR) team's shared inbox. An overdue invoice with an unanswered customer reply stays unpaid until someone in the AR team responds.
Late payment is common in European B2B trade: nearly four in five Western European companies say their B2B customers pay late, and about a quarter of invoices are paid after the due date, according to the Atradius Payment Practices Barometer published in May 2026.
Paraglide's AI agents automate collections end to end by sending personalised payment reminders, answering customer replies and billing queries, routing disputes and following up in the existing email thread until the invoice is paid.
Paraglide customers see results within weeks: Choco, a wholesale business, reduced its days sales outstanding (DSO) by 34% within 14 days, and Gorilla Gasket, a gasket manufacturer, cut its accounts receivable aging by 38%.
Introduction
Many AR managers, credit controllers and collections managers in B2B companies face the same problem: customers reply to scheduled payment reminders with questions, such as "The PO number on invoice 3320 is wrong", instead of paying. Rule-based dunning software cannot read customer replies, so rule-based dunning software keeps sending payment reminders on schedule even when the customer's question has not been answered. An overdue invoice with an unanswered customer query stays unpaid until someone in the AR team corrects the issue and writes back to the customer.
How collections automation works
Collections automation runs in two layers. Dunning rules in the enterprise resource planning (ERP) system or AR software decide which customer gets a payment reminder and when. AI collections agents read the replies customers send back, answer billing queries from ERP data, log promise-to-pay dates and follow up in the same email thread until the invoice is paid.
How to automate collections in 6 steps
To automate collections, AR teams work through six steps in order: (1) measure current DSO, collection effectiveness index (CEI) and customer reply volume; (2) connect the ERP, CRM, billing system and finance inbox; (3) write dunning rules for each customer segment; (4) make it easy for customers to pay; (5) deploy AI collections agents to handle customer replies, billing queries and disputes; and (6) track the results and give the agents more autonomy where they perform well. The table below shows what each collections automation step involves, who owns it and what the AR team should have at the end of the step.
Step | What you do | Owner | Output |
1. Measure DSO, CEI and reply volume | Pull 3 to 6 months of aged debt, DSO, CEI, customer reply volume and query types | AR manager with finance systems | A baseline and a business case |
2. Connect the ERP, CRM, billing system and inbox | Connect the ERP or accounting system, CRM, billing system and finance inbox | Finance systems or IT | One view of invoices, contacts and customer conversations |
3. Write dunning rules per segment | Segment customers, then choose reminder cadence, tone, channel and escalation for each segment | Credit or collections manager | A reminder cadence per customer segment |
4. Make it easy to pay | Include payment details and invoice copies in every message; add payment links, direct debit or a portal; agree payment plans | AR with treasury | Fewer "how do I pay?" replies |
5. Deploy AI collections agents | Agents handle customer replies, billing queries and disputes, starting with human approval | AR manager | Customer replies answered in the thread; people approve exceptions |
6. Track results and widen autonomy | Track DSO, CEI, promise-to-pay kept rate and forecast accuracy every month | Finance director or CFO | Proof of impact and a decision on where agents act alone |
Step 1: Baseline DSO, CEI and reply volume before you automate collections
A collections automation project needs three baseline numbers: DSO, CEI and the number of customer replies that collections emails generate. Without baseline numbers, an AR team cannot show whether collections automation worked or decide which part of the collections process to automate first.
DSO measures how many days cash takes to arrive after invoicing:
DSO = (Accounts receivable ÷ Credit sales in the period) × Days in the period
CEI measures the share of collectable receivables a company collected in a period:
CEI = (Opening receivables + Credit sales − Closing total receivables) ÷ (Opening receivables + Credit sales − Closing current receivables) × 100
For example, a B2B company starts the month with £1.2m in receivables, invoices £800k of new credit sales during the month, and ends the month with £1.1m in receivables, of which £700k is not yet due. The example company's CEI for the month is (1.2 + 0.8 − 1.1) ÷ (1.2 + 0.8 − 0.7) × 100 = 69%, which means the company collected 69% of the money it could have collected. The example company's DSO is 1.1 ÷ 0.8 × 30 = 41 days, which means the company takes 41 days on average to turn an invoice into cash.
AR teams should count how many customer replies their collections emails generate before automating collections. To measure reply volume, take one month of collections emails and tag each customer reply by type: invoice copy request, PO number, payment confirmation, remittance, dispute or promise to pay. The share of collector time spent answering customer replies shows how much of the collections workload scheduled payment reminders cannot automate.
Baseline DSO, CEI and reply-volume figures form the business case for collections automation, because a CFO approving the project will ask how much cash and collector time it frees up compared with today. Gartner's 2026 survey of 160 senior finance leaders found that accounts payable and accounts receivable automation generally returns its cost within nine to 10 months, faster than most finance AI projects.
Step 2: Connect your ERP, billing system and finance inbox
Collections automation software can only act on the data it can read: open invoices and payment status from the ERP, customer contacts from the CRM, and customer conversations from the finance inbox. AR teams should connect all three systems before writing any dunning rules.
Most ERPs already include rule-based collections. Microsoft Dynamics 365 Finance's collections process automation schedules reminder emails, collector activities and collection letters by days relative to the due date, with "quiet days" to limit contact. SAP, NetSuite and Business Central include similar rule-based dunning.
The finance inbox is the system most collections automation projects forget to connect. Customer replies, remittances and PO numbers arrive by email, so collections automation that cannot read the finance inbox cannot tell a customer who has paid from a customer who has not. Paraglide connects to the CRM, ERP, billing system, email inbox and ticketing system, and works with ERPs such as SAP, Oracle, Microsoft Dynamics 365 Finance & Operations, Business Central, Sage, NetSuite, Infor, IFS, Odoo, QuickBooks and Xero.
Step 3: Write segmented dunning rules for each type of customer
Dunning rules should differ by customer segment, because a key account with a PO problem and a small account that forgot to pay need different messages. AR teams should segment customers by value, payment history, country and invoice frequency before choosing reminder cadence and tone.
AR teams should choose a collections channel for each customer segment. Email carries most B2B collections because customer replies, remittances and PO numbers arrive by email, while phone calls and formal letters suit later escalations. Paraglide's AI agents work email-first, in the finance inbox and in existing email threads.
Long tail of small balances: a fixed reminder cadence (pre-due reminder, due date, 7 days overdue, 14 days overdue) with escalation to a formal notice. Small balances are where full collections automation pays back first.
Strategic accounts: fewer and more personal payment reminders, with escalation to the account manager rather than a formal notice.
High invoice frequency: customers with 20 or more open invoices should receive one statement-style reminder per customer, not 20 separate emails. Paraglide's Collections Agent sends one personalised reminder per customer, covering all open invoices.
Habitual late payers: shorter gaps between payment reminders and earlier escalation. Paraglide's AI agents escalate with localised recovery actions such as formal notices.
A comparison of rule-based dunning tools is in Paraglide's guide to dunning software for B2B.
Step 4: Make paying easy so reminders do not trigger "how do I pay?" replies
Every collections message should let the customer pay or act without writing back, with the invoice copy attached, payment details included, and a payment link or direct debit option where the supplier has one. Customers who cannot pay a large balance in one go need a route to a payment plan, recorded as a promise to pay.
Step 5: Deploy AI collections agents for replies, billing queries and disputes
An AI collections agent handles the part of collections that dunning rules cannot: reading a customer's reply and taking the next action. AR teams should deploy AI collections agents after dunning rules are running, starting with human approval on every message.
A collections conversation handled by Paraglide's Collections Agent and Billing Support Agent runs in six stages:
The Collections Agent sends one personalised payment reminder per customer, covering all open invoices and written from the customer's account history, payment behaviour and earlier conversations. A customer who usually pays on time receives a different payment reminder from a customer who pays late across many invoices.
A customer replies, for example "We paid last Friday", "Can you resend invoice 3320 with the corrected PO?" or "We're disputing this charge". The Collections Agent pauses payment reminders as soon as the customer replies.
The Billing Support Agent reads the customer's reply, checks the payment or finds the invoice in the ERP, and answers in the same email thread, with a corrected invoice attached where needed. The Billing Support Agent captures promise-to-pay dates and collects PO numbers from customer emails.
When a reminder email bounces, the Collections Agent identifies a new contact at the customer and updates the contact data.
When a customer goes quiet or a promised payment does not arrive, the Collections Agent follows up in the existing email thread.
When a customer raises a dispute, or a case needs a person, the agent routes the case to the right AR owner with the full email thread, the account history and a summary of where the case stands.
Control: Paraglide's AI agents run in two modes. In suggest mode, also called supervised mode, the agent drafts each reply and a person approves the reply before it is sent. In act mode, the agent sends replies without approval and the AR team monitors the conversations. Before go-live, AR teams should write escalation rules that define what the agent resolves end to end and what the agent always routes to an AR specialist, such as disputes, deductions above a value threshold or high-value accounts. Paraglide customers decide when an agent can act on its own and when human approval is required, and increase agent autonomy gradually in Paraglide's agent studio.
Step 6: Measure DSO, CEI, promise-to-pay kept rate and forecast accuracy
AR teams should measure the same numbers they baselined in step 1 every month, and add two metrics that show whether automated customer conversations are working: promise-to-pay kept rate and cash forecast accuracy.
DSO and CEI: the headline results of collections automation. Comparing like-for-like months removes seasonal effects.
Promise-to-pay kept rate: promises paid on or before the promised date, divided by all promises made. A falling promise-to-pay kept rate means follow-ups on broken promises are too slow.
Forecast accuracy: the cash forecast compared with actual receipts. Logged promise-to-pay dates give the cash forecast specific dates instead of averages.
Why payment reminders alone do not reduce DSO
Collections tools fall into three generations, and the difference between the generations is who handles the customer's reply. Legacy reminder automation and SaaS rule-based workflows send payment reminders and leave the AR team a to-do list of replies to answer. AI-native agents, such as Paraglide's, read each reply, answer it and follow up, so the AR team handles only complex cases. Rule-based tools chase invoices, while AI collections agents collect them.
The table compares how each collections step runs with reminder-only automation and with AI collections agents.
Collections step | Reminder-only automation | AI collections agent (Paraglide) |
First outreach | Template sent at a fixed interval, often one email per invoice | One personalised reminder per customer, covering all open invoices |
Customer replies | Reply lands in the shared inbox; the schedule keeps running | Reminders pause; the agent replies in the same thread |
Billing query | Collector finds the invoice and answers by hand | Billing Support Agent answers from ERP data, or routes with context |
Bounced email | Reminder fails silently | Agent finds a new contact and updates contact data |
Promise to pay | Noted in a spreadsheet or ERP note, if at all | Logged, with a follow-up if the payment does not arrive |
Dispute | Collector logs, investigates and routes | Agent routes to the right owner with the thread and account history |
What is the difference between a Collections Agent and a Billing Support Agent?
A Collections Agent handles outbound collections, while a Billing Support Agent handles inbound customer queries in the finance inbox. Paraglide's Collections Agent sends personalised payment reminders, follows up on overdue invoices, answers customer replies to collections emails and escalates to the AR team when a case needs human judgement.
Paraglide's Billing Support Agent works in the finance inbox 24/7 and in any language, handling invoice copy requests, PO number issues, payment confirmations, statement requests, amount discrepancies and dispute notifications. The Billing Support Agent reads each customer email, finds the relevant data in the ERP, answers routine queries itself and routes complex queries to the AR team with the context assembled.
Collections conversations need both agents, because a customer's reply to a payment reminder is usually a billing query. With reminder-only collections tools, a billing query sits in the inbox until a collector answers it, while the reminder schedule keeps running as if the customer never replied.
Collections automation tools: rule-based dunning vs AI collections agents
Collections automation tools fall into four groups: AI agents that work the collections inbox, ERP-native collections rules, AR platforms with collections workflows, and enterprise order-to-cash (O2C) suites. AR teams should choose a tool group by where payment delay sits: in sending reminders, or in handling what customers send back.
AI agents that work the collections inbox: Paraglide. AI agents send reminders, answer replies and follow up in existing threads. Paraglide is best for high-volume B2B teams whose payment delays sit in replies, queries and disputes.
ERP-native collections: Microsoft Dynamics 365 Finance, SAP, NetSuite. ERP rules send reminders, letters and collector tasks inside the ERP. Paraglide's AI agents handle the replies, queries and follow-ups that ERP reminders generate.
AR platforms with collections workflows: Billtrust, Invoiced. AR platforms give collectors a workspace with invoicing, payments and portals, and suit teams changing how they invoice and take payment. Paraglide automates the collections conversations whatever invoicing platform a company uses.
Enterprise O2C suites: HighRadius. Enterprise suites include collections as one module of a wider finance platform, and suit enterprises putting many finance processes with one vendor.
The table compares each collections automation tool on the same fields, based on each vendor's own website or documentation on 6 October 2026. "Not described" means the pages checked did not cover the capability, not that the capability does not exist.
Tool | Best for | Inbound reply handling | Collections coverage | Dispute workflow | ERP integration | Implementation | Pricing | Main limitation |
Paraglide | High-volume B2B AR teams | Agents read and reply in thread; approval configurable | One reminder per customer, replies, follow-ups, formal notices, promise-to-pay tracking | Detect, route, follow up until resolved | SAP, Oracle, Dynamics 365, Sage, NetSuite, QuickBooks, Xero and others; plus CRM, billing and inbox | Go live in 3 days | Quote-based | Email-first: works in the finance inbox and existing threads |
Dynamics 365 Finance (collections process automation) | Teams already on D365 Finance | Not described | Rule-based emails, collector activities and collection letters by process hierarchy; pre-dunning | Not described on the page checked | Native to D365 Finance | Configured inside D365 | Included in D365 Finance; no separate price published | Replies and queries stay with collectors |
Billtrust | Teams changing invoicing and payments | Customer emails create cases | Collector worklists prioritised by buyer behaviour; invoice delivery to 260+ AP portals | Case-based | 200+ connectors, including SAP, Oracle, Infor and Dynamics | Not published on homepage | Quote-based | Wider platform change across invoicing and payments |
Invoiced (by Flywire) | Mid-market teams wanting a self-serve payment portal | Not described | Reminder cadence adjusted to payment behaviour; customer portal; AutoPay | Not described | NetSuite, Sage Intacct, Dynamics 365, Workday, QuickBooks, Xero | Not published | Not published | Reply handling not described |
HighRadius | Enterprises buying O2C from one vendor | AR email inbox sorts, summarises and drafts responses | Dunning emails, prioritised worklists, in-app calling | Matching against POs and portal data | 50+ ERPs | Vendor cites 3 to 6 months to ROI | Quote-based | Suite purchase with a longer rollout |
What implementation looks like: live in 3 days, autonomy widened over the coming weeks
Paraglide's Billing Support Agent goes live in 3 days. Paraglide's AI agents work with the AR team's existing ERP and email address, so customers keep writing to the same finance inbox and do not need to learn a new portal.
A typical Paraglide rollout runs in three phases:
First 3 days: connect invoices, payment status, customers, contacts and the email inbox; agents go live in suggest mode, with a person approving each reply.
Over the coming weeks: agent autonomy widens to sending documents, confirming payment status, promise-to-pay follow-up and chasing in existing threads, as the AR team reviews the agents' work.
Later, with approval: agents can act in other systems, such as updating customer details in the ERP.
Next steps
Count your customer replies this week. Tag one week of customer replies to collections emails by type. If most customer replies are queries rather than payments, dunning rules alone will not reduce DSO.
Calculate last month's CEI with the CEI formula in step 1, and record CEI next to DSO as the collections automation baseline.
Pick one customer segment for AI agents first, such as the long tail of small balances, and run AI agents in suggest mode for two weeks.
To see Paraglide's AI agents run collections on your own invoices, book a demo.