How to Identify Unbilled Revenue with AI and Your ERP

The question from finance leadership seems simple: “What work have we already delivered but not yet billed?” Answering it usually requires reviewing contracts, emails, acceptance records, and ERP transactions. When identifying unbilled revenue with AI, the challenge is not generating a list of completed deliverables: it is proving which ones meet the billing requirements and remain unbilled.

The topic is timely following NetSuite's SuiteSkills announcement on October 26, 2026. Its Revenue Leakage capability is designed to investigate potentially unbilled deliverables and charges below agreed amounts, distinguishing them from accounting exceptions, according to the Oracle NetSuite press release. This is a starting point for evaluating the problem, not evidence of guaranteed recoveries.

For a service, maintenance, or project-based business, the opportunity lies in connecting information that is currently reviewed separately. You do not need to replace your ERP: you need traceability between what was agreed, delivered, authorized, and billed, with human review before any charge.

1. What identifying unbilled revenue with AI means

In this context, we are looking for services delivered but not billed that could become confirmed unbilled items. We are not automatically determining revenue recognition for accounting purposes or looking for overdue invoices: an issued but unpaid invoice belongs in collections.

Some deliverables are not yet billable, either. The contract may require customer acceptance, completion of a phase, or waiting until a billing cutoff date. That is why each finding must answer:

• Which contractual obligation does this deliverable correspond to?
• Is there sufficient evidence of completion and authorization?
• Has the billing milestone or date been reached?
• What portion is covered by previous invoices or applied advance payments?

AI helps interpret documents and connect supporting evidence. Business rules determine eligibility and calculations; Finance validates the conclusion. A probable match should not turn directly into an invoice.

2. Connect four sources around each deliverable

The unit of review should be a deliverable, service order, or milestone, not just the customer. The same company may have multiple contracts and projects underway simultaneously.

Source	Information it provides	What must be verified
Contracts and amendments	Scope, rates, milestones, currency, and terms	Current version and approved changes
Operational evidence	Reports, tickets, completion records, or logged deliverables	Work performed, date, and quantity
Authorizations	Customer acceptances and approvals	That they correspond to the milestone and an authorized person
ERP	Invoices, line items, credit memos, and advance payments	Actual coverage and status of each transaction

Define shared identifiers: customer, contract, project, order, and line item. If they are missing, AI can suggest relationships using descriptions and dates, but it must flag them as uncertain. A similar description does not prove that two records correspond to the same service.

For example, “monthly maintenance” may appear for several locations belonging to the same customer. Without a location and period, the system could associate an invoice with the wrong deliverable.

An enterprise AI integration can connect these sources while keeping the ERP as the system of record and attaching supporting documentation to each proposed billing item.

3. Cross-check the evidence and separate exceptions

The process should extract contractual terms, locate deliverables and their authorizations, and query ERP line items. Then, it applies explicit rules to classify each case:

Situation	Proposed treatment
Accepted, billable deliverable with no associated invoice	Potential confirmed unbilled item
Deliverable missing contractually required acceptance	Awaiting authorization, not ready for billing
Partial billing	Review only the eligible remaining balance for the milestone
Recorded advance payment	Verify its application against the contract and supporting documentation
Canceled service	Exclude it if cancellation eliminates the obligation to pay
Future billing date	Schedule a review when the condition is met
Issued but unpaid invoice	Route to the collections process

Advance payments require special care: they must not be deducted indiscriminately from any deliverable. You need to verify which contract they belong to, how they are applied, and which documents already exist. Likewise, a credit memo may change the balance without implying that an invoice should be issued again.

The calculation must compare amounts on the same basis: currency, quantities, rates, discounts, and taxes. Withholdings and special terms require explicit handling. Avoid letting the language model perform these calculations freely; they should be executed using verifiable rules.

Hypothetical example: a company completes three maintenance visits, but the contract allows billing only after monthly acceptance. If that acceptance is missing, work has been delivered, but there is not yet an item ready for billing. If it has already been accepted and one invoice covers two visits, the review focuses on the third, taking applicable advance payments into account.

4. Route each finding to a review queue

The useful outcome is not a message saying “there is lost revenue.” It is a queue where each case has supporting evidence, an owner, and a next action.

At a minimum, include:

• Identified customer, contract, project, and deliverable.
• Potential unbilled amount, currency, and calculation breakdown.
• Linked contractual clause, delivery evidence, and authorization.
• Invoices, credit memos, and advance payments considered.
• Reason for the alert, missing information, and data retrieval date.
• Owner, review status, and documented decision.

Operations confirms completion; the commercial or contract owner clarifies scope and changes; Finance validates eligibility and the balance. An authorized person decides whether billing should proceed.

For the pilot, use read-only access and restrict documents by role and entity. Log which data was accessed and who resolved each exception. Before issuing an invoice, query the ERP again: another user may have generated it during the review.

AI process automation can assign tasks and track progress without independently authorizing charges, modifying contracts, or contacting the customer.

5. Design a focused pilot with owners and metrics

Start with one service line, one entity, and a closed period. Choose contracts with comparable terms and identifiable deliverables. Avoid combining complex projects, multiple currencies, and hard-to-document exceptions from the outset.

Owners and execution

1. Finance leadership: defines what qualifies as a billable outstanding item and approves exclusion rules.
2. Operations and contract administration: gather evidence and validate deliverables, acceptances, and amendments.
3. IT or the ERP owner: enables limited access, maps fields, and verifies the integrity of queries.
4. Integration team: configures extraction, cross-checking, calculations, and traceability.
5. Billing: reviews approved cases and confirms their resolution in the system.

Before activating alerts, build a manually reviewed sample that includes known unbilled items and correctly billed cases. Use it to check both what the system gets right and what it misses.

Metrics for deciding whether to scale

• Precision: confirmed unbilled items as a proportion of reviewed alerts.
• Coverage: known unbilled items the system detected within the validated sample.
• Review time: effort per case compared with the manual process.
• Resolution: closed cases and the age of those still open.
• Financial outcome: amounts identified, approved, billed, and collected, reported separately.

Define success criteria before starting. The amount flagged by AI does not equal recovered revenue; nor does every additional invoice immediately turn into cash.

6. From an alert to a verifiable unbilled item

Identifying unbilled revenue with AI creates value when it helps explain what remains to be billed, why billing is appropriate, and what evidence supports it. AI in ERP should help investigate and prioritize, not replace financial judgment.

Slashweb can help you evaluate available sources and design an integration around your current ERP. Contact Slashweb to define a pilot with clear scope, owners, and controls before expanding the solution.