0:06 — Where AI fits, and where it doesn’t — The presenter opens by noting that not every task is a good fit for AI, and that tools attempting to replace decisions requiring nuance or human judgment have backfired. This motivates defining clear boundaries around how and when the agents operate. 0:26 — Boundaries for the AI agents — Three stated design rules: summarize past activity but do not rewrite it, suggest alternatives rather than force or automatically switch them, and flag potential violations rather than block without context or nuance. 0:49 — Three tools: shipped, previewed, planned — Drawing on relationships built with clients over nearly three decades, the presenter introduces the GL coding tool released last year and shown in the live demo, a sneak peek at the PO matching tool built from user feedback, and an approver selection tool on the roadmap. 1:20 — How GL code suggestions are generated — The GL coding tool draws on the history of items already processed in Pairsoft, so existing customers benefit from their own past coding. Suggestion fields are defined by the customer’s team working with Pairsoft’s designers, and can include who is doing the coding, invoice dollar amount, and description fields. 2:26 — Defaults versus exceptions — Where ERP defaults can be pulled in, Pairsoft prefers to do that. The tool is aimed at the exception cases — the roughly 20% of invoices not covered when defaults handle about 80%, such as unusually large amounts, a particular location, or a different description. 3:02 — Customer feedback since release — The presenter revisits a slide of user feedback from an earlier session, describing the response as overwhelmingly positive. The point singled out as most impactful is that saving even a few seconds on each invoice adds up, since users no longer have to stop and look up the correct code. 3:44 — Switching to the live demo — The presenter switches screens to the Pairsoft interface, where several invoices have already been pulled in. They are sitting at the first step after AP review, in a reviewer’s queue, and one is opened at random. 4:47 — An invoice with no default distribution — The opened invoice shows fields already extracted by OCR, but it has no default distribution. This is framed as the case where the AI tool helps — for example a new employee who does not know how this particular invoice should be coded. 5:10 — The suggestion panel in the interface — A small panel in the upper-right corner of the screen holds the feature. The presenter notes that existing Pairsoft customers can have it added by talking to their account manager. 5:34 — Applying a suggested code — Clicking into the panel shows recommendations based on how the invoice was coded before. Selecting one — office supplies in this case — fills in the account along with the amount and the description. 5:48 — “Well, that’s it?” — The presenter recounts showing the feature to a family member outside the AP field, who responded “well, that’s it?” The brevity is the point: three likely codes offered instead of searching through a large list of GL codes. 6:17 — Suggestions do not restrict further edits — Accepting a suggestion does not prevent adding additional codes, splitting the distribution further, or pulling account descriptions back from the ERP as the system always has. The tool only offers the codes most likely to be correct. 6:47 — Add-on licensing and next steps — The AI capability is not part of the base product — it is an optional add-on for the organization. Customers are directed to their account manager to discuss how it fits into their tooling and contract.

Home / Resources / Webinars

Demo: AI for General Ledger Coding and our Upcoming AI Finance Products

Key takeaways

Pairsoft’s stated design principle for its finance AI agents is bounded assistance: summarize past activity without rewriting it, suggest alternatives without automatically switching, and flag potential violations without blocking.
The GL coding tool, released last year, suggests general ledger codes for an invoice based on the customer’s own historical coding data already in Pairsoft.
Suggestions can be driven by fields the customer defines with Pairsoft’s designers — examples given include who is doing the coding, the invoice dollar amount, and description fields.
The tool is positioned for exception handling rather than replacing ERP defaults; the example given is a customer using defaults about 80% of the time and needing help with the remaining 20%.
In the live demo, an invoice with no default distribution is opened, a suggestion panel in the upper-right corner offers three likely codes, and selecting one fills in the account, amount, and description.
Accepting a suggestion does not lock the coding — users can still add codes, split distributions further, and pull account descriptions back from the ERP as usual.
The AI capability is an optional add-on rather than part of the base product, and existing customers arrange it through their account manager.
Also previewed: a PO matching tool built from user feedback, and an approver selection tool described as next on the roadmap.

Overview

This session covers Pairsoft’s AI agents for accounts payable, centered on a live demonstration of the general ledger coding tool that the company released last year. The presenter opens with the framing that not every task suits AI: tools that try to replace decisions requiring nuance or human judgment have been seen to backfire. Pairsoft’s response has been to define explicit boundaries on how and when its agents operate — agents should summarize past activity but not rewrite it, suggest alternatives rather than force or automatically switch them, and flag potential violations rather than block without context.

The GL coding tool works from the history of items already processed inside Pairsoft, which the presenter notes makes it immediately useful to existing customers because it draws on their own past coding rather than a generic model. Which signals drive a suggestion is configurable: customers work with Pairsoft’s designers to define the fields that matter, with examples including the person doing the coding, the dollar amount of the invoice, and description fields. The tool then looks at how that customer coded similar invoices in the past and offers up likely codes.

The presenter is explicit that this is not meant to displace ERP defaults. Where defaults can be pulled in, Pairsoft would rather do that. The value case offered is the exception path — the scenario where defaults cover roughly 80% of invoices and a user has to stop and hunt for the right code on the remaining 20%, such as when an amount is larger than usual, when the invoice goes to a particular location, or when the description differs. Feedback since release is described as overwhelmingly positive, with the point that mattered most being that saving even a few seconds per invoice adds up across volume.

The demonstration uses the Pairsoft interface with several invoices already loaded and sitting at a first review step, assigned to a reviewer. One invoice is opened at random. OCR has already extracted a set of fields, but this particular invoice has no default distribution — the situation a new employee, or anyone unfamiliar with how that invoice should be coded, would face. A small suggestion panel appears in the upper-right corner of the screen; clicking into it surfaces recommendations based on how the customer coded similar invoices before. Selecting one — office supplies, in the example — populates the account along with the amount and description. The presenter emphasizes that the interaction is deliberately small, recounting that a family member outside the AP field reacted with “well, that’s it?”, and that being it is the point: three likely codes, no scrolling through a long GL list. Accepting a suggestion does not constrain what follows; users can add more codes, split the distribution further, and pull account descriptions from the ERP exactly as before.

The session also flags two things beyond the shipped GL coding tool: a PO matching tool built in response to user feedback, shown as a sneak peek, and an approver selection tool described as the next item on the roadmap. On commercial terms, the presenter states plainly that the AI capability is not part of the base product — it is an optional add-on, and existing customers should contact their account manager to discuss how it fits into their tooling and contract.

Chapters

0:06 — Where AI fits, and where it doesn’t — The presenter opens by noting that not every task is a good fit for AI, and that tools attempting to replace decisions requiring nuance or human judgment have backfired. This motivates defining clear boundaries around how and when the agents operate.
0:26 — Boundaries for the AI agents — Three stated design rules: summarize past activity but do not rewrite it, suggest alternatives rather than force or automatically switch them, and flag potential violations rather than block without context or nuance.
0:49 — Three tools: shipped, previewed, planned — Drawing on relationships built with clients over nearly three decades, the presenter introduces the GL coding tool released last year and shown in the live demo, a sneak peek at the PO matching tool built from user feedback, and an approver selection tool on the roadmap.
1:20 — How GL code suggestions are generated — The GL coding tool draws on the history of items already processed in Pairsoft, so existing customers benefit from their own past coding. Suggestion fields are defined by the customer’s team working with Pairsoft’s designers, and can include who is doing the coding, invoice dollar amount, and description fields.
2:26 — Defaults versus exceptions — Where ERP defaults can be pulled in, Pairsoft prefers to do that. The tool is aimed at the exception cases — the roughly 20% of invoices not covered when defaults handle about 80%, such as unusually large amounts, a particular location, or a different description.
3:02 — Customer feedback since release — The presenter revisits a slide of user feedback from an earlier session, describing the response as overwhelmingly positive. The point singled out as most impactful is that saving even a few seconds on each invoice adds up, since users no longer have to stop and look up the correct code.
3:44 — Switching to the live demo — The presenter switches screens to the Pairsoft interface, where several invoices have already been pulled in. They are sitting at the first step after AP review, in a reviewer’s queue, and one is opened at random.
4:47 — An invoice with no default distribution — The opened invoice shows fields already extracted by OCR, but it has no default distribution. This is framed as the case where the AI tool helps — for example a new employee who does not know how this particular invoice should be coded.
5:10 — The suggestion panel in the interface — A small panel in the upper-right corner of the screen holds the feature. The presenter notes that existing Pairsoft customers can have it added by talking to their account manager.
5:34 — Applying a suggested code — Clicking into the panel shows recommendations based on how the invoice was coded before. Selecting one — office supplies in this case — fills in the account along with the amount and the description.
5:48 — “Well, that’s it?” — The presenter recounts showing the feature to a family member outside the AP field, who responded “well, that’s it?” The brevity is the point: three likely codes offered instead of searching through a large list of GL codes.
6:17 — Suggestions do not restrict further edits — Accepting a suggestion does not prevent adding additional codes, splitting the distribution further, or pulling account descriptions back from the ERP as the system always has. The tool only offers the codes most likely to be correct.
6:47 — Add-on licensing and next steps — The AI capability is not part of the base product — it is an optional add-on for the organization. Customers are directed to their account manager to discuss how it fits into their tooling and contract.

Questions this webinar answers

What does the Pairsoft GL coding AI tool actually do?

It suggests general ledger codes for an invoice being processed in Pairsoft. When a user opens an invoice that has no default distribution, a panel in the upper-right corner of the screen offers the three most likely codes based on how similar invoices were coded in the past. Selecting one populates the account along with the amount and description.

Where does the tool get its suggestions from?

From the history of items already processed inside Pairsoft for that customer. It looks at how the customer and their team coded that invoice or similar invoices previously, rather than applying a generic external model. Existing Pairsoft customers therefore have usable history from day one.

Which fields influence the suggestions?

The fields are defined by the customer’s team working with Pairsoft’s designers. Examples given include who is doing the coding, the dollar amount of the particular invoice, and description fields being used.

Does this replace ERP defaults?

No. Where a customer can use defaults in their ERP to pull codes in, Pairsoft would rather do that. The AI suggestions are aimed at the exception cases — the scenario described is a customer whose defaults cover about 80% of invoices, leaving roughly 20% where the correct code has to be looked up manually, such as when an amount is larger than usual, when the invoice goes to a specific location, or when the description differs.

Can a user override or add to what the AI suggests?

Yes. Accepting a suggested code does not stop a user from adding additional codes, splitting the distribution further, or pulling other account descriptions from the ERP the same way the system always has. The tool only offers up the codes most likely to be correct; it does not lock the coding.

What limits does Pairsoft put on how its AI agents behave?

Three boundaries were stated: the agents should summarize past activity but not rewrite it, suggest alternatives rather than force or automatically switch them, and flag potential violations rather than block without context or nuance. The reasoning given is that tools attempting to replace decisions requiring nuance or human judgment have been observed to backfire.

Is the AI capability included in the base Pairsoft product?

No. It is an optional add-on rather than part of the base product. Existing customers who want it should contact their account manager to discuss how it fits into their tooling and their contract.

What other AI tools are planned beyond GL coding?

A PO matching tool, built based on feedback from users of the GL coding tool, was shown as a sneak peek. An approver selection tool was described as next on the roadmap.

What benefit have existing users reported?

Feedback since release was described as overwhelmingly positive. The point singled out as most impactful was that even saving a few seconds on each invoice adds up over volume, because users no longer have to stop and search for the correct code themselves.

Table of Contents

Get A Demo:

Simple solutions. Powerful results. Seamlessly integrated.

Get a PairSoft Demo