Evaluating AP Automation for Your Workday Environment
If your AP team uses Workday and is evaluating automation software, this guide covers what Workday handles natively, where manual work may remain, and what to consider when comparing additional AP tools.
Where the manual work actually sits
Workday's native OCR can capture invoice header fields, but invoice coding is a separate task. Fields such as company, cost center, spend category, and other worktags often cannot be determined from the invoice alone.
For example, the same vendor may need to be coded to different cost centers depending on which team placed the order.
.png)
Approval routing: the question most Workday buyers underweight
Approval routing is worth discussing early in an evaluation because organizations handle it differently.
When routing is maintained using individual approver names, changes such as promotions, transfers, or ownership of a new cost center can require updates to the approval structure. Most Workday environments maintain these mappings manually, including through spreadsheets. This creates ongoing administrative work and increases the chance that an invoice is routed to the wrong person, sent to someone who is no longer responsible for that area, or delayed while the approval structure is corrected.
Vic.ai can instead assign approvers using role, department, team, and entity information synchronized from Workday. Approval flows can also be created from an existing approval matrix, imported or exported through CSV, or maintained through an API. Approvers can review invoices from mobile devices as well.
In one enterprise deployment, 9,194 employees and their roles are synchronized from Workday and used across 1,622 role-based approval flows with approximately 530 active approvers.
Invoices below $5,000 can be configured for automatic approval, while invoices above that amount are routed to an appropriate current approver. In that environment, approximately 63% of recent invoices moved from receipt through posting without manual intervention.
An important decision to make early is where approvals should take place.
Some organizations want approvals to occur in the AP automation platform while Workday remains the system of record. Others want approvals to continue entirely within Workday.
If keeping approvals in Workday is a requirement, that should be discussed at the beginning of an evaluation because it affects the scope and design of the integration.
PO matching against Workday's math
Workday calculates purchase-order values using quantity multiplied by unit price rather than relying only on the line amount.
Because of this, an invoice and PO can agree on total value while still creating a line-level exception.
Common examples include:
- Invoices priced by the case against a PO written per unit
- Freight as an item line on one document, an expense line on another
- Bundled components on the PO, itemized on the invoice
Each has to be reconciled to Workday's calculation before the invoice can be posted correctly.
Vic.ai compares invoice and PO line information, including invoice descriptions, quantities, prices, and other available data. It can also identify a corresponding PO in some cases where the PO number is not listed on the invoice.
In the enterprise deployment referenced above, approximately 97% of invoices are PO-based. Of those invoices, 87.9% result in a clean match. Invoices that do not meet the matching requirements are routed for review rather than posted automatically.
What the integration involves
- Access: a Workday ISU account, scoped to the domains the integration touches. Expect your security team to want that scope documented.
- Auth: OAuth 2.0.
- Custom reporting: where Workday doesn't expose a field cleanly, RaaS reports fill the gap. Plan on a Workday SME on your side for a small amount of configuration work.
- SSO: SAML 2.0, with OpenID Connect as an alternative.
- HR data: synced for approval routing.
The smart (low-risk) way to roll out AI
Choosing to adopt AI doesn’t have to be an all-or-nothing proposition from day one. In fact, we recommend a measured, steady approach. As an example, auto-pilot (our term for invoice processing that does not touch human hands) does not need to be enabled for every invoice at once.
A team can begin with system-generated coding recommendations, compare those recommendations against actual invoice processing, and then enable automated processing for areas where the results meet its requirements.
Automation can be configured by vendor, invoice type, or entity, with confidence thresholds set by the organization.
Invoices that require review remain visible in review queues, along with confidence scores and records of coding changes and approvals.
In the enterprise deployment referenced above, the percentage of new invoices processed automatically increased from 2% to 47% during the rollout.
Over the same period, field-level accuracy increased from 95.4% to 98.5%.
The level of automation was increased gradually based on actual processing results rather than enabled across the entire invoice population at the beginning.
Questions to ask any vendor
- Does coding come from templates and rules, or from learning our history? Ask what happens when a vendor changes their invoice layout.
- Who maintains approval routing after go-live, and what happens when someone changes roles?
- How does the tool handle Workday's quantity times unit price calculation on PO matching?
- What's required from our Workday SME, and for how long?
- Can autonomy be enabled incrementally, by vendor or entity, with thresholds we control?
- What does the audit record look like on an invoice the AI processed without a human?
Where Vic.ai fits
Strongest fit: 1,000+ invoices a month, meaningful non-PO or PO-heavy volume, multiple entities, and a worktag structure you actually report on.
Vic.ai's AI was trained on more than one billion invoices and keeps learning from your vendors, entities, and corrections. In the enterprise operation referenced above, that came to roughly 289,000 invoices processed in six months at 97.7% field-level accuracy, with about 99% of posting executed by the system rather than a person, and net cash exposure across the hands-off cohort of 0.05%.
.svg)
