In today’s fast-paced business environment, efficiency and accuracy are paramount for any accounts payable department. The advent of artificial intelligence offers new possibilities automating repetitive tasks, minimizing errors, and streamlining approval processes. Rillion AI stands at the forefront of this transformation, delivering a powerful solution to automate account coding and generate intelligent workflow proposals - based on your organization’s unique invoice history.
This guide is designed to help you quickly set up and begin working with AI-powered coding and workflow, ensuring a smooth transition towards a more automated way of handling invoices. Whether your goal is to reduce manual intervention, increase transparency, or enhance compliance, this will empower your team to focus on higher-value activities while the AI handles the manual work.
Table of content
- Before you start
- Enable AI in Prime
- Additional configuration
- How to get started
- FAQ
Before you start
Make sure you are on the right subscription
Existing customers: AI is an add on module and can be added to all Packages.
Note! As of Feb 2026, AI is included in Professional and Premium packages.
If you are unsure if you have AI included in your Rillion subscription, please contact Rillion Support or your Customer Success Manager.
AI is not per-default activated
AI needs to be activated by Rillion before starting. Reach out to Rillion Support or your CSM.
Enable AI in Prime
Enable AI for the companies you want to start with
In Company settings → Invoice Log, you can enable AI for each company at varying automation levels. Several settings are available, but for activation, focus on the three below.
1. Enable AI matching for Invoice Posting
This setting turns AI on for the company. We recommend option C (Automatic), which is designed to maximize the automation and functionality of the AI feature.
- No - AI is not activated.
- Only Manually - AI support is enabled, but must be triggered for each invoice individually by clicking "Match" in the Invoice Log (done by the Accounts Payable team).
- [Recommended] Automatic - AI runs automatically for every invoice that is a candidate. The coding and flow proposal is set when invoices arrive in the Invoice Log.
Confidence level to set Transfer for AI Matched Invoices
When AI determines the coding and flow, each invoice receives a confidence level. This setting defines the minimum confidence required for an invoice to bypass AP review in the Invoice Log. We recommend starting with High, then later allowing Medium to bypass review.
- All
- Medium and High
- High
Enable AI matching for Flow predictions
To let AI generate approval flows for incoming invoices, select the checkbox Use AI for Flow Predictions.
Then follow the step-by-step guide "AI - Company settings to Activate AI in Rillion Prime" to get started.
Additional configuration
Disable AI for suppliers with stable coding
Some vendors should always use the same coding, or have invoices split across several lines at a fixed percentage. For these vendors, Reference Matching in Rillion Prime may work better than AI.
- Go to the Vendor register and find your vendor.
- Open the Account Posting section to find the AI settings.
- Untick 'Enable AI Matching for Invoice Posting' to deactivate AI for account coding.
- Untick 'Activate AI matching for the Invoice Flow' to deactivate AI for building the flow.
For details, follow the guide "AI - How to inactivate AI for a specific Vendor".
Disable AI for specific fields in the account coding
If certain dimensions are frequently incorrect, you can exclude them from AI prediction. Under Company settings → Invoice Log, use the field 'Do not use AI predictions on following fields' to specify which account posting dimensions AI should skip.
Note: Dimensions (called Objects in Prime) are specified by their database title, with multiple values separated by commas. For example, to exclude Object3 and Object4, enter: Object3, Object4
Tip: If you're unsure which Object types your dimensions belong to, go to Administration → Registers → Objects. The drop-down under "Object type no" shows the Object types in use, ordered Object1 to Object8.
Reference match to be used when AI generated posting lines exceed
This setting caps how many account posting lines AI may create. If AI generates more than the specified number, reference matching is used for that invoice instead.
- 1 = use AI only if not more than 1 line
- 2 = use AI only if not more than 2 lines
- 3 = use AI only if not more than 3 lines
- [Recommended] Always = always use AI lines, no limit
We recommend Always, so AI handles invoices regardless of line count. To be able to maximize the value of AI in use case of periodization and allocation.
How to get started
1. Inform your organization
Tell your team and stakeholders that you're enabling AI in Prime to reduce administrative burden through intelligent invoice routing and auto-coding. Make clear that approvers remain responsible for reviewing every AI proposal, as errors can occur.
What you need to inform approvers about
- The organization should be informed that AI functionality is active, including its purpose and expectations.
- AI output is shown by this icon. By clicking and/or hoovering this icon users (both in the invoice log and the flow) can see why AI picked the specific flow or coding, called AI explanation.
- The coding and/or flow proposal now includes a confidence level color (see previous section). If the approver changes the coding, their values replace the suggestion and the color indicator disappears.
- If the invoice coding suggestion is correct, the approver simply approves the invoice. If changes are needed, they follow the same steps as usual.
- If manually changed the AI output icon and explanation will disappear.
2. Decide your rollout scope
With your team, choose between a complete rollout or a phased start with a limited set of companies or vendors.
3. Start with your most common scenarios
Every company has different invoice types, coding structures, and approval hierarchies, and some invoices are far more complex than others. Beginning with the most common, standard scenarios gives you the strongest automation platform fastest, especially when AI is combined with base-level reference matching.
4. Start using AI
When using manual AI matching
If AI matching is set to manual for your company, you must specify for each invoice whether you want to apply AI matching. This is done on the invoices in the Invoice Log
- Open the Invoice in the Invoice Log
- Choose Type of Matching to AI matching
- Click Match.
- The AI model is now generating a flow proposal and coding - this usually takes under 1 minute. You can continue to work with other invoices during this time.
- Reopen the invoice to view any AI suggestions and their confidence levels
Follow the step-by-step guide "AI - Manual AI Matching in Rillion Prime" for more information.
When using automatic AI matching
When Automatic matching is configured in company settings, the invoice log displays all invoices processed by AI. Both the list and detailed views present the overall confidence level. Additionally, for each dimension and role suggested by the AI, the corresponding confidence level is clearly indicated.
Confidence levels
- Green = High confidence
- Yellow = Medium confidence
- Red = Low confidence
Changes can be made to both the coding and flow proposal.
Any changes made, both in the Invoice Log and during approval, will be reported back to the AI that will keep learning.
To change the account coding follow the step-by-step guide "AI - Working with an Invoice Matched by AI: Adjusting the AI-suggested Account Coding"
And to change the flow proposal follow the step-by-step guide "AI - Working with an Invoice Matched by AI: Adjusting the AI-suggested Flow"
5. Follow up on performance
In Analytics there is a report called “Invoice Summary” where you have some great KPIs to analyze the performance of the AP process. The report provides details on what matching type is used and how they perform. Use the report to find vendors and invoices that provide poor matching outcomes with the help of the AI feature and investigate potential changes that could be made to improve the matching.
See the guide “Getting Started with Analytics - AP automation insights” for more great reports to use to help improve your organizations Invoice Automation
Frequently asked questions
What is the new AI matching, and how is it different from the previous model?
The new AI matching is Rillion's internal AI model (currently InferenceVersion 3.3.2 as of April 10 2026). It's an LLM-based prediction service that handles coding and workflow suggestions, replacing the older machine-learning model (a classic ML model trained on customer history)
How does the AI weigh invoice content vs historical data?
Can I activate AI by day 1?
New customers without existing historical data can leverage AI coding & workflow from day 1. However, the AI model will learn and improve over time. You should be up and running in 2-3 months time.
AI prefers historical data before it can suggest reliable proposals, so activate it from the start but in manual matching mode.
If you're new to Rillion, set up basic reference matching first to generate proposals while AI gathers data. Once you have enough volume, typically within 2 to 3 months, switch to automatic AI matching.
Can we train the AI model?
No, models are no longer trained. The new way of teaching is AI is by instructions. This is currently done by your CSM/SDM but we are developing an instruction/configuration layer in Prime.
How does AI generate posting lines?
When the AI suggests account coding, it actually starts from your invoice history. It looks through past invoices you've already coded and finds the ones most similar to the new invoice, based on things like the supplier, the line descriptions, and the amounts.
Once it finds the closest match, it follows that pattern and adapts it to the new invoice. For example, if similar invoices in the past were split across two accounts (say, materials and services), the AI will apply the same split here, using the amounts on the current invoice.
The invoice helps it choose the right precedent. If your invoice carries clear signals like a project code, cost center, or department, the AI prefers past invoices that share those same signals, since an invoice from the same project or department is a better guide than one that just looks similar. And if a code on the invoice exactly matches one of your set-up object codes, the AI uses that directly.
The AI only uses accounts you've used before or ones on your approved list. It won't invent new accounts. It also keeps things simple, using as few lines as the invoice and your history support, and it follows your past pattern for how VAT code is handled.
NOTE! VAT lines are not generated from AI, but from Prime side based on the VAT code on the posting line.
How does AI generate an approval flow?
1. By default the AI checks if there is a current reference match based on your set-up reference matching in Prime.
2. When the AI suggests an approval flow, it looks at two things:
- what's on the invoice itself
- how similar invoices have been handled in the past
The AI checks the invoice first. It looks for clear signals like a reference, project or object code, cost center, or department. If it finds one of these and it points to a specific approver, the AI uses that to build the flow. This matters because the same supplier can send invoices that belong to different sites, projects, or departments, so what's actually on the invoice is often the most reliable guide.
3. If the invoice doesn't contain a clear signal, the AI falls back on history. It looks at how invoices like this one (same supplier, similar amount, similar content) were routed before, and follows that pattern.
In short: the invoice leads, history backs it up. Clear information on the invoice takes priority, and past behaviour fills in the gaps when the invoice image alone isn't enough.
Which roles are included in AI flow generation?
By default, the AI will only suggest roles that have been added into a flow through the following:
OriginOfRole (can be seen by mouse over in the Flow for an historical Invoice):
- Role manually added in flow (3)
- Role manually added through report (4)
- Flow copied from historical invoice (13)
- Manually added flow proposal in flow (14)
- Manually added flow proposal through reports (15)
- Flow proposal manually added in invoice log (16)
- Flow proposal from Vendor (18)
- Flow proposal from company (19)
- Flow proposal reference 1 (20)
- Flow proposal reference 2 (21)
- Flow proposal reference match (22)
NOTE! This means that if the flows you have today are built based on, for example ‘Rule based Flow auxiliary’ or ‘Dynamic flows’, those will not be used by AI to generate approval flows.
AI will also not add the Manager roles into the flow, the manager roles will still be added to the final flow but with help of normal flow proposal functionality, this means that the Manger role will not be visible in the Invoice log, it will be added at transfer from the log if needed.
Can the AI predict accruals / periodization?
Yes, but customers may correct AI's technically-correct period to match their period-close date (a day or two before month-end). That isn't an AI error.
Where is the AI hosted, and where is customer data stored?
Both the model and the predictions follow the customer's data region: • US customers: hosted and stored in the US • EMEA customers: hosted and stored in the EU