Right now, the average small business is owed about $17,700 in unpaid invoices, waits nearly a month to get paid, and burns four hours every week chasing money it has already earned. If that sounds like your business, the collections industry has noticed — and its answer is to stop asking your staff to make the awkward calls and start sending software to do it instead.
The latest entry is Daylit, an accounts-receivable platform that launched AI agents for automated collections: software that reads your invoice history, picks which customers to nudge, and sends follow-up emails, phone calls, and texts on its own. Early adopters, the company says, nearly tripled collections on high-risk accounts while cutting manual follow-up work by more than 40 hours a week. Whether or not you ever buy that particular product, the model it represents — autonomous, multi-channel follow-up — is coming to small business AR fast. This guide explains what these agents actually do, where they genuinely help, and the compliance and bookkeeping groundwork you need before you let one loose on your customer list.
The Late-Payment Problem, by the Numbers
Start with why this category exists at all: getting paid keeps getting slower.
- In the second quarter of 2026, small businesses waited an average of 29.3 days to get paid, with payments arriving almost nine days late on average — about a full day worse than the prior quarter.
- One 2026 analysis found that 92% of invoices are now paid after their due date, with the average small business waiting 28.8 days beyond terms.
- Intuit QuickBooks' 2026 small business late-payments report found that nearly 59% of small businesses carry invoices more than 30 days past due, owed an average of $17,700.
- About 29% of small businesses say late customer payments threaten their ability to pay their own suppliers on time, and U.S. businesses collectively hold an estimated $825 billion in outstanding receivables at any given time.
- Owners and staff spend an average of four hours a week actively chasing late payments — more than eight lost business days a month.
The pattern is familiar: the work is done, the invoice is sent, and then the money sits in someone else's account while you draft the third "gentle reminder" and debate whether calling makes you look desperate. Multiply that hesitation across dozens of open invoices and you get the 29-day average. Automation vendors are betting that consistent, unembarrassed follow-up — sent at the right time through the right channel — closes most of that gap without a human having to feel awkward about any of it.
What Just Launched: Collections That Run Themselves
Daylit's launch is representative of the new wave. The platform connects directly to a company's ERP, CRM, and communication channels and positions itself as an execution layer for receivables rather than another dashboard: instead of showing what is owed, it acts on it. Three capabilities define the pitch:
Identify risk early. The system flags accounts showing signs of delayed or missed payment based on payment history, communication patterns, and real-time signals — before a slow payer becomes a write-off. This is the "which invoices should I worry about today" question, answered automatically each morning.
Take action automatically. AI agents handle follow-ups across email, phone, and text, pulling in invoice history, supporting documentation, and prior payment commitments so each message is context-aware. A customer who promised to pay on the 15th gets a different nudge than one who has ignored three emails, without anyone on your team drafting either message.
Improve cash visibility. Cash flow projections update continuously as new payment data, customer communications, and behavior come in, giving whoever runs finance a live picture of where receivables stand rather than a week-old aging report.
The company reports that it has helped more than 200 companies recover hundreds of millions in outstanding receivables, and it backs the launch with a $110 million funding round from September 2025 plus a completed SOC 2 Type I examination in February 2026 — the latter being the baseline security credential you should demand from any vendor that will touch your customer data and your ERP.
Treat the headline metrics as vendor claims, not guarantees: nearly 3x collections on high-risk accounts, 40-plus hours a week of manual work eliminated, AR operating costs down more than 75%, and email reply rates around 50% against an industry average of 15%. Your results will depend on your customer mix, your data quality, and how your current process compares — a business with no follow-up cadence at all will see bigger gains than one with a disciplined collector already in place. But the direction of travel is clear: the follow-up ladder that used to live in a spreadsheet or a collector's head is becoming software that runs itself.
How Autonomous Follow-Up Actually Works
If you have ever set up an email dunning sequence — reminder at 7 days overdue, firmer note at 14, final notice at 30 — you already understand the skeleton. AI agents add three things a static sequence cannot do.
First, they read the account before they write the message. Instead of a one-size template, the agent pulls the invoice, the customer's payment history, open disputes, and any commitments the customer made ("the check goes out Friday"). The outreach references specifics, which is most of why reply rates climb: a message that names the invoice, the amount, and the broken commitment is much harder to ignore than "this is a friendly reminder."
Second, they choose the channel and the timing. Some customers answer texts and ignore email; others do the reverse. Some pay when nudged the morning their AP batch runs and ignore everything else. The agents track which combinations produce responses and payments per customer and adapt over time — effectively A/B testing your collections playbook continuously, per account, without anyone designing the experiments.
Third, they know when to stop and escalate. The responsible versions of these systems route disputed invoices, settlement offers, hardship claims, and key accounts back to a human rather than arguing with your customer. When you evaluate any vendor, the escalation design matters more than the message-writing quality: ask exactly which situations trigger human handoff, how fast, and what the customer experiences in the meantime.
Where the Robots Earn Their Keep — and Where They Don't
Autonomous follow-up is not a universal upgrade. It shines in specific situations and misfires in others.
Best fit:
- High invoice volume with a thin team. If one person owns AR alongside three other jobs, consistent automated follow-up beats sporadic human follow-up every time.
- Repeat B2B customers with chronic slowness. The customer who always pays at 45 days on 30-day terms is a process problem, not a relationship problem. Relentless, polite automation fixes process problems.
- Early-stage delinquency. The first 60 days past due are when internal effort works best and when automation is cheapest. Reserve expensive human attention and outside agencies for older balances.
Poor fit:
- Disputed invoices. An agent cannot resolve a short shipment, a contested change order, or a quality complaint. Sending payment demands while a dispute is open torches goodwill and sometimes violates the contract's dispute procedure.
- Your largest accounts. A templated nudge sequence — however clever — is the wrong instrument for the customer representing 20% of revenue. Senior humans own those conversations.
- First-ever collections contact after a relationship change. New ownership, a departed AP contact, or a customer in obvious distress deserves a human call first. Automation can take over the routine follow-through afterward.
The practical rule: automate the routine, humanize the exceptions. The businesses that get burned are the ones that automate everything and discover the exception when a key customer forwards an agent's third "final notice" to the owner with a note about finding a new vendor.
The Compliance Fine Print: The Liability Is Yours, Not the Vendor's
Here is the part vendors mention quietly: when an AI agent calls or texts your customer, you are generally the party on the hook for compliance. Outsourcing the keystrokes does not outsource the liability. Three regimes matter most.
The TCPA governs automated calls and texts. The Telephone Consumer Protection Act restricts calls and texts to mobile numbers made with an autodialer or an artificial or prerecorded voice — and a 2024 FCC ruling clarified that AI-generated voices count as artificial voices. That means AI collection calls generally need the recipient's prior express consent, must respect the 8 AM to 9 PM calling window in the recipient's local time, and must honor opt-outs immediately. Statutory damages run $500 to $1,500 per violation, which is how a helpful automation becomes a five-figure lesson. Before switching on voice or text outreach, confirm you have documented consent for each number, a working opt-out mechanism on every channel, and calling windows enforced in the customer's time zone — not yours.
The FDCPA and Regulation F govern consumer collections conduct. The Fair Debt Collection Practices Act and the CFPB's Regulation F — including the 7-in-7 rule limiting call frequency and the required disclosures in each communication — apply to third-party collectors pursuing consumer debts. If your customers are businesses owing trade debts, the FDCPA generally does not apply to your own in-house collection efforts. But mixed customer bases, personal guarantees, and sole proprietors blur the line, and several states impose their own mini-FDCPAs with broader reach. If any of your receivables could be characterized as consumer debt, have the vendor show you exactly how its agents handle disclosures, frequency caps, and cease-communication requests — then have your attorney confirm it.
State laws stack on top. State telemarketing statutes, state debt-collection laws, call-recording consent rules (some states require all parties to consent), and data-privacy laws all apply to automated outreach independently of federal rules. A system configured for federal compliance can still violate a stricter state law. Ask vendors for state-by-state configuration, full interaction logging, and audit-ready records of every contact attempt: what was sent, when, through which channel, under what consent, and what happened next.
None of this is a reason to avoid the technology. It is a reason to buy it the way you would buy a forklift: powerful, useful, and dangerous without training, guardrails, and a clear record of who authorized what.
Before You Switch It On: A Practical Checklist
Whether you adopt a dedicated AI collections platform or just upgrade your accounting software's reminder sequences, the same groundwork determines whether automation helps or embarrasses you.
- Clean your aging report first. Automation amplifies whatever your books say. Unapplied payments, duplicate invoices, and credits never issued become confident, specific, wrong demands. Reconcile AR, apply every open payment and credit memo, and resolve every known dispute before the first automated message goes out.
- Segment ruthlessly. Exclude disputed invoices, customers in workout arrangements, and strategic accounts from automation entirely. Start autonomous outreach on the segment with the clearest facts: undisputed, documented, recently overdue balances.
- Document consent and honor opt-outs. Log how you obtained consent to call or text each number, and make sure opt-outs propagate instantly across email, voice, and text — a customer who texts STOP and then gets an AI phone call the next day is a complaint waiting to happen.
- Review the scripts and the escalation paths. Read every template, test the agent against your trickiest real scenarios (disputed invoice, partial payment, promise-to-pay, angry reply), and confirm a human gets involved fast when conversations leave the happy path.
- Keep humans on disputes and big dollars. Set a dollar threshold and a dispute flag that route accounts to a person automatically, and review the agent's handled conversations weekly at first — then monthly once the patterns prove out.
- Measure what matters. Track days sales outstanding (DSO), percent current, reply rates, promises-to-pay kept, and cost per dollar collected — before and after. If the vendor's 3x claim does not materialize in your 60-day pilot, you want your own numbers to negotiate with.
Clean Books Are the Prerequisite, Not the Afterthought
Every capability above rests on one unglamorous foundation: accurate receivables records. An AI agent is only as good as the ledger it reads. If your AR aging mixes cash- and accrual-basis entries, if partial payments sit unapplied for weeks, if your allowance for doubtful accounts is a guess carried forward from last year, automation will energetically act on fiction. The businesses seeing real gains from these tools share the same boring habits: invoices issued promptly with clear terms, payments applied the day they arrive, credits and disputes logged where the collector — human or silicon — can see them, and a monthly review of the aging that writes off what is truly dead instead of letting it rot.
That discipline also protects you when automation makes a mistake, because the audit trail starts in your books. When a customer disputes an agent's message, the conversation ends quickly if you can show the invoice, the terms, the payment history, and the consent record in one place. It ends badly if the answer is "let me ask the software what it was thinking."
Keep Your Receivables — and the Rest of Your Books — Organized
As collections go autonomous, the advantage shifts to the business with the cleanest data feeding the machine. Maintaining clear, timely financial records is what turns an AI agent from a liability into leverage. Beancount.io provides plain-text accounting that gives you complete transparency and control over your financial data — no black boxes, no vendor lock-in. Get started for free and see why developers and finance professionals are switching to plain-text accounting.
