Missed the latest Charted webinar on using natural language AI directions when handling invoice automation edge cases? Below is a round-up of some of the topics discussed during the Q&A. If you’re interested in watching the recording, you can download the webinar here.
If you’re evaluating or implementing AP automation with AI-powered invoice processing, you’ve probably run into a common question: how much do you actually need to configure for the system to work the way you want?
The answer depends on how well you understand a key feature: AI directions. These are the rules and instructions that tell the AI what to extract from specific invoices, how to handle edge cases, and how to populate fields in NetSuite automatically. Getting them right is the difference between a touchless process and a manual one.
Here’s what you need to know.
What are AI directions (and what’s the difference between the two input fields)?
In Charted Clerk, there are two places where you can give the AI instructions: a general AI direction box and custom field extraction rules.
The general AI direction box is designed for broad, high-level context, similar to a system prompt if you’re familiar with how AI tools work. Think of it as background information you’d want the AI to “always remember” when processing invoices from a particular vendor or category.
In practice, though, most teams won’t need to use it often. The bulk of your configuration work will happen in custom field extraction rules, where you define what to look for on the invoice and what field to populate in NetSuite as a result.
You don’t need something in both fields for a rule to work. Custom field extraction alone handles the majority of real-world use cases.
Can AI directions be layered or prioritized?
Yes, multiple directions can apply to the same invoice at once. When they do, all of the relevant instructions are passed to the AI simultaneously, and the system uses that combined context to determine what to return.
There’s no built-in priority ranking between directions. Instead, triggering keywords are what control which rules apply: if the AI sees a specific piece of text on an invoice, that triggers the matching direction. This matters when you’re building out your rule set. If two directions apply to the same invoice and they contradict each other, the AI will try to reconcile them, but the outcome isn’t guaranteed. In those cases, it’s worth reviewing whether your keyword triggers are specific enough to prevent overlap.
Should AI directions be long and detailed, or short and focused?
Short and focused, almost always.
One of the most common mistakes in setting up AI directions is trying to build conditional logic directly into the rule. For example, writing something like “if you see this department code, populate department X; if you see this location, populate location Y”. Instead, this creates complexity that the AI has to parse through rather than clear instructions it can act on reliably.
A better approach: write one focused direction per use case, triggered by a specific keyword. If a single vendor has multiple extraction scenarios, use multiple directions scoped to that vendor, rather than combining them into one long rule.
If you’re not sure how to structure a direction for a specific scenario, your customer success manager can review examples with you before you go live.
What about populating fields that aren’t on the PO?
This is a more nuanced area. For PO-backed bills in NetSuite, there’s a defined set of fields that the system expects to come from the PO versus fields that AP automation solutions should populate. Fields like amortization dates, for instance, can sometimes appear on the PO itself, in which case the system would typically carry them over automatically.
If you have a field that isn’t being populated as expected in a PO-backed workflow, the right move is to loop in support with specifics. The answer will often depend on your NetSuite configuration and how your PO template is structured.
How do you give other users access to manage AI directions?
Access to the AI directions settings is controlled through NetSuite role-based permissions. An administrator manages this through the Charted settings panel, accessible via the gear icon in the upper right corner.
Under the Permissions tab, there’s a specific permission called “Manage AI directions” that can be granted at the role level or the individual employee level. Once granted, those users can create and edit directions without needing to involve an admin each time.
This is worth setting up early if you have a team member who will own the ongoing maintenance of your extraction rules.
The honest reality of working with AI extraction rules
One thing worth acknowledging: AI directions are not perfectly deterministic. The behavior of any given rule depends on the AI interpreting the context you’ve given it, and sometimes you’ll need to test and iterate before a direction works exactly as intended.
That’s not a limitation unique to any one tool. It’s a reflection of how AI-powered extraction works. The practical implication is to start simple, test with real invoices, and refine from there rather than trying to build a comprehensive rule set all at once before you’ve seen it in action.
Quick tips for getting more out of Invoice AI
These are smaller details that don’t always come up in initial setup conversations but tend to matter once you’re in production.
AI directions and AI rules are different things. Directions are natural language prompts that guide the AI during the scanning and extraction phase. Rules are structured if-then logic that applies after scanning is complete. For example, a rule can automatically assign a subsidiary or queue owner based on the email address an invoice arrived from, or based on text in the file name. If you’re trying to control routing or ownership rather than extraction behavior, rules are the right tool.
Use a period as your keyword to apply a direction universally. Keyword triggers control which invoices a direction applies to. If you want a direction to fire on every invoice regardless of vendor, enter a period (.) as the identifying keyword. Since a period appears on every invoice, the direction will always trigger.
Skip the word “extract” in your field directions. Invoice AI already knows it needs to extract data. Using the word “extract” in your direction adds noise without adding clarity. Write your directions the way you’d prompt any AI model: be specific, give an example of the expected output, and include location cues where helpful (“upper left-hand corner,” “first column”). The more concrete your instruction, the more consistent the result.
Touchless processing requires an exact PO quantity match. For PO-backed bills, Invoice AI will only auto-create the vendor bill and move it to auto-processing status when the invoice quantities exactly match the remaining billable quantity on the PO. If the invoice includes tax or shipping that wasn’t on the PO, or if there’s any quantity variance, the bill routes to the inbox for manual review instead. Knowing this upfront helps set realistic expectations around touchless rates and can inform how your team handles those vendors.
You can limit which pages get scanned. If your vendors frequently include supporting documentation after the first page or two, you can set a page end limit in the direction to stop scanning at a specific page. This prevents extra lines from appearing on the bill in NetSuite from content that was never meant to be extracted.
The general AI directions box has a useful translation application. If you work with international vendors who send invoices in other languages, the whole-invoice AI direction box can translate invoice content into your preferred language before it’s brought into NetSuite. It’s one of the more concrete use cases for that field beyond the advanced system-prompt scenarios described earlier.
Getting more from touchless invoice processing
The goal of AI-powered invoice automation isn’t just to eliminate manual data entry. It’s to create a system that processes invoices accurately, routes them correctly, and stays aligned with how your business actually operates in NetSuite, without requiring someone to intervene on every document.
AI directions are the mechanism that makes that possible. The teams that get the most value out of them are the ones who treat them as living configuration, something to refine as your vendor mix, chart of accounts, and business rules evolve over time.
Interested in learning more about AI use in invoice automation?
Register for our next webinar, The building blocks for truly touchless invoice processing in NetSuite to learn about how to build a truly efficient AP operation, with analysis on how to measure success. This webinar is the third in our webinar series, the Invoice AI Master Class Series for NetSuite. The first webinar is available to watch as a recording here.
