Build vs Buy AR Automation: What It Costs to Build Collections In-House in 2026

You can build a basic AR collections loop in a few weeks. Getting to a system that survives real volume, partial payments, remittance mismatches, disputed amounts, unresponsive contacts, and portal rejections that surface days later, takes roughly a year of senior engineering plus $2,000 to $12,000 a month in LLM costs, and then it never stops needing maintenance. For most teams AR automation is infrastructure they depend on rather than a product they sell, so the real question is whether you want to be in the business of building and running it, or whether you want to use it.

What does build vs buy mean for AR automation?

Build means your engineers create and own the system that chases invoices, reads replies, resolves exceptions, and applies cash. Buy means you run that on a platform built for it. The decision looks close on a whiteboard, because a first version is genuinely easy to stand up. It stops looking close the moment the system meets production, where the exceptions are not rare events but the daily norm.

How long does it take to build?

The first build is not the expensive part. A follow-up loop can work within weeks. The gap between that and a production system is six to twelve months of iteration, driven by edge cases that only appear at volume. Broken out, the core engineering looks like this: the workflow and logic layer runs 12 to 18 weeks, integrations with Stripe, NetSuite, QuickBooks, and a CRM run another 12 to 18 weeks, the interface and reporting take 6 to 8 weeks, and edge case handling has no fixed endpoint. Total time to a usable production system is about a year, and that assumes a focused team with no rework.

Why do in-house AR builds stall at production?

Because the hard part is the long tail, and the long tail is most of the work. Three areas get underestimated every time.

State management is the first. An invoice is not simply open or closed. A conversation moves through a lifecycle with branching paths: paused for a promise to pay, escalated to a person, reopened after a bounce. The rules for valid transitions, and the invariants that keep them consistent, are subtle and accumulate for as long as the system exists.

Orchestration is the second. Deciding who gets contacted, when, in what order, and in what tone needs a configurable rules engine rather than a prompt. The targeting logic alone, covering who is in scope, what overrides apply, and how exceptions are prioritized, tends to get rewritten several times even after launch.

Portal automation is the third. A large share of enterprise invoices, 92% in the accounts Monk runs, must be submitted through a procurement portal. Each portal has its own format, PO matching rules, error codes, and login flow. This is browser automation at scale rather than an API integration, and it stays operationally intensive after it is built, because portals change their specs and customers move between them year to year.

What does building cost?

The comparison is not build price against license price. It is a year of senior engineering plus ongoing token and infrastructure cost against a fixed fee and a four-week implementation.

DimensionBuild in-houseBuy (Monk)Time to productionAbout a year to a fragile v1Live in under a weekOngoing LLM cost$2,000 to $12,000 a month, plus infraIncluded in one feeEdge casesHandled manually, never finishedHandled out of the box at scaleSystem of recordStitched across systemsPurpose-built source of truthComplianceYours to design and defendSOC 2 Type II from day oneMaintenanceA permanent internal productOperate, do not maintain

Each invoice needs several LLM calls a month, for drafting, reply classification, exception handling, and cash application, each carrying a large context window of history. At 100 to 10,000 invoices a month, tokens alone run $2,000 to $12,000, before infrastructure, logging, retries, and monitoring.

What about security and compliance?

AR automation touches more sensitive data than it first appears. Billing contacts, payment terms, bank details, and invoice amounts are all in scope, so the system has to meet your data handling standards and your customers'. Submitting to portals means storing and rotating credentials and MFA tokens. Disputes require a complete, append-only audit trail by design, because downstream logic depends on its integrity. Sending financial email at volume is an attack surface for spoofing. And if you sell to enterprises, your customers will ask whether your AR system falls under your compliance posture, which makes that surface yours to defend and audit.

Is an LLM and a Stripe webhook enough?

This is the most common shortcut, and it is where most builds begin. A language model drafts a good reminder, so it feels like the job is nearly done. It is not: the model cannot own invoice state, apply cash, resolve a dispute, or file into a portal. We cover that gap in detail in why an LLM alone is not collections automation, and the full list of what breaks in the AR automation edge cases teams underestimate.

When does building make sense?

When AR automation is your differentiation rather than your infrastructure, building can be the right call. If chasing invoices is a core part of what you sell, you may want to own every part of it. For most companies it is the opposite: collections is critical but undifferentiated, the engineers who build it become the experts you cannot reassign, and the roadmap for the internal tool competes with the roadmap for the product that earns revenue.

How does Monk compare?

Monk is a purpose-built AR system of record, from invoice to cash, that most teams take live in under a week. It resolves 90% of collections with no human intervention, reaches customers with a 24% higher response than standard dunning, and reduces DSO by 40% or more, while cash application matches at 80% and rises to 95% with suggested rules. It handles the 600-plus procurement portals that manual teams dread, files 87% of those submissions autonomously, and is SOC 2 Type II from the start. More than $2B in receivables runs on Monk today, including for Profound and ElevenLabs. See it in Monk's collections automation and ERP integrations.