Skip to guide

Case study guide

AI accounts payable and finance automation: company examples, step by step

Updated 7 Nov 2026

AI accounts payable automation reads invoices and supplier emails and sorts, codes and routes them. AI accounts receivable automation does the matching of payments and drafts customer messages. This page walks through four company examples, step by step.

Invoices, a calculator and a laptop on a desk
40% to 90%payment matching accuracyMicrosoft, reported Jun 2026
38%less invoice processing time in 3 monthsAppZen on WSH, undated
91%straight-through cash applicationHighRadius on Shiseido, undated

Quick answers

  • Microsoft: consolidated data first, then added AI to match payments (40% to 90% accuracy), draft replies and rank each person's work each morning.
  • Westbury Street Holdings: AI labels every email in the AP inbox, flags duplicates and keeps an audit trail. The vendor reports 38% less processing time in three months.
  • Billerud and Shiseido: results are published but the steps are thin.
  • The pattern: clean up the data and process first, automate the routine messages and matching, and keep people on exceptions.

What the examples have in common

  • Fix the process and data first. Microsoft's director advises knowing your process well and designing the new one, rather than putting an agent into the old one.
  • Automate the high-volume routine work: reading invoices, labelling email, matching payments, drafting standard replies.
  • Keep an audit trail. WSH says every step is recorded, whether a person or the automation did it.
  • Let people handle exceptions, disputes and judgement calls.
  • Start with the hardest cases or one team, then expand.

Microsoft Treasury: collections with an AI agent helping case managers

Microsoft's Global Collection team uses AI agents to help people who chase customer payments. Microsoft published this account on 4 June 2026.

A laptop with financial charts, a phone calculator and banknotes
Workflow diagram
Before
  1. Before1,000+ collectors
  2. BeforeSeveral systems to find invoices and past conversations
  3. BeforeNo single view of what a customer owes
After
  1. 1
    SystemData moved into SAP and Dynamics 365
  2. 2
    SystemIQ layer adds business context; templates and worklists standardised
  3. 3
    AI agentRoutine messages and statements fully automated
  4. 4
    AI agentPayments matched to invoices (40% to 90% accuracy)
  5. 5
    AI agentEmails routed, replies drafted, late payments and disputes predicted
  6. 6
    PersonEach morning the agent ranks work; case managers lead, with Copilot suggestions

Before

More than 1,000 collectors worldwide had to check several systems to see a customer's full invoices and what other teams had already said. Information was spread across systems and handoffs, with no single view of what a customer owed.

How the process was automated, step by step

  1. First, move customer, invoice and payment data into one place: an SAP and Microsoft Dynamics 365 environment.
  2. Add Microsoft's IQ intelligence layer on top, so the AI understands business context. Standardise templates and worklists so all regions work the same way.
  3. Automate routine messages and statements completely.
  4. Use AI to match payments to invoices (accuracy 40% to 90%), draft customer replies, and route emails to the right collections manager.
  5. Put Copilot assistance inside the case managers' daily work: suggestions, call summaries and drafted replies.
  6. Each morning the agent ranks each case manager's work by urgency and the customer's past behaviour. It also predicts late payments and likely disputes.

Tools and systems connected

  • SAP
  • Microsoft Dynamics 365
  • Microsoft IQ intelligence layer
  • Copilot

SAP and Microsoft Dynamics 365 are the single source of data.

Where people stay in the loop

  • Case managers stay in charge of the work. The agent prepares, drafts, prioritises and routes. Microsoft calls it a "human-led, AI agent-assisted" system.
  • Microsoft ran change management and role-based training on real daily scenarios, and included frontline users in early testing.

Reported results (company figures)

  • Payment matching accuracy from 40% to 90%; 98% of payments applied within 48 hours.
  • 40% less call preparation time; automatic cash application doubled; customer inquiries resolved 2.5 times faster; up to 60% less inquiry handling time.
  • Hundreds of thousands of hours unlocked each year (base not published). The source does not show how this was calculated.

Sources: Microsoft Inside Track, 4 June 2026

Westbury Street Holdings: invoices from inbox to payment

Westbury Street Holdings (WSH) is a UK food and hospitality company with more than 4,000 locations. AppZen published this case study (no date shown on the page).

A colourful paper tray holding stacked documents on an office desk
Workflow diagram
Before
  1. BeforeSite managers key invoices by hand
  2. BeforeShared inbox, no owner or tracking
  3. BeforeSeveral accounting portals
After (AppZen)
  1. 1
    SenderEmail arrives in the AP inbox
  2. 2
    AI agentLabelled: invoice, statement or payment inquiry
  3. 3
    AI agentInvoices classified, coded and routed
  4. 4
    AI agentDuplicates and suspicious senders flagged
  5. 5
    SystemOne workspace with audit trail (Dynamics GP)
  6. 6
    PersonFinance staff handle exceptions and supplier disputes

Before

Site managers keyed in invoices from small suppliers by hand, because e-invoicing was not practical for those suppliers. Invoices outside the standard food purchasing model needed manual coding. Staff logged in and out of several accounting portals. All supplier questions went to one shared inbox with no owner and no tracking.

How the process was automated, step by step

  1. Every incoming email in the accounts payable inbox is read and labelled as an invoice, a statement document or a payment inquiry.
  2. Invoices are classified, coded and routed by the software instead of by site staff.
  3. Suspicious sender domains and urgent-sounding emails are flagged early. Duplicate invoices are flagged so they are not paid twice.
  4. Invoices, supplier emails and supporting documents sit in one workspace, with an audit trail showing each step and whether a person or the automation did it.
  5. WSH started with its most complex invoice types to prove the tool before removing manual review.

Tools and systems connected

  • AppZen Autonomous AP
  • AP Inbox Service Center
  • Microsoft Dynamics GP

Microsoft Dynamics GP is the accounting system.

Where people stay in the loop

  • Finance staff handle exceptions and supplier disputes.
  • Site staff no longer process invoices, but sites keep accountability for their costs.
  • Next planned: AI agents for VAT, invoice data extraction and supplier replies.

Reported results (vendor figures)

  • Over three months: 38% less invoice processing time; automatic processing rate up 2.4 times; average invoice review 2.3 minutes.
  • First four months: 3,400 emails labelled; 215 given a predicted reply; £774,000 of duplicate invoices flagged before payment; an estimated 89 hours of manual work recovered.

The vendor lists what the tool does, not the order of steps inside it.

Sources: AppZen case study (undated)

Billerud: reading PDF invoices with AI

Billerud is a paper and packaging company. Basware published this case study (no date shown on the page).

Stacked paper rolls seen end on in a warehouse
Workflow diagram
What the source says (order of steps not published)
  1. BeforeOld OCR tool: low data quality, much manual work
  2. 1
    AI agentBasware Smart PDF reads PDF invoices
  3. 2
    SystemTurned into e-invoices
  4. 3
    PersonMinimal validation; who does it is not stated

Steps not published. Only the before and after are published.

Before

The accounts payable team spent hours each day on manual invoice handling. An older OCR tool gave low data quality and needed a lot of manual work.

Tools and systems connected

  • Basware Smart PDF

Basware Smart PDF replaced the old OCR tool. Basware says it uses deep learning, text extraction and AI vision to read PDF invoices and turn them into e-invoices.

Where people stay in the loop

  • The page says the tool needs minimal validation and that the team now spends time on higher-value work. It does not say who checks the rest.

Reported results (vendor figures)

  • Over 90% of PDF invoices validated automatically.
  • 66% lower cost of handling PDF invoices and 25% lower total monthly costs, with payback within a few months.

Only the tool and the results are published, not the steps.

Sources: Basware case study (undated)

Shiseido: matching incoming payments to invoices

Shiseido's accounts receivable team uses HighRadius AI agents for cash application, the job of matching incoming payments to open invoices. HighRadius published the story (no date shown on the page).

A hand holding a card to a payment terminal against a coloured background
Workflow diagram
Four HighRadius agents named (order not published)
  1. 1
    AI agentEmail Remittance Capture
  2. 2
    AI agentCheck Remittance Capture
  3. 3
    AI agentPayment Invoice Matching
  4. 4
    AI agentPayment Exception Handler
  5. 5
    PersonAnalysts review the exceptions the agent flags (total payments not published)

Steps not published. The page names the agents but not how they connect.

Before

Not published.

Tools and systems connected

  • Email Remittance Capture
  • Check Remittance Capture
  • Payment Invoice Matching
  • Payment Exception Handler

Four named HighRadius agents.

Where people stay in the loop

  • Analysts handle the exceptions. The page lists "17K+ exceptions handled by analysts monthly with Payment Exception Handler agent". It does not give total payments, so that number cannot be turned into a share of all payments. The page also reports 91% straight-through cash application.

Reported results (vendor figures)

  • 91% straight-through cash application and a 71% improvement in cash application productivity.
  • 95% of remittances captured automatically from email, 83% from checks, and about 106,000 invoices posted automatically each year.

Sources: HighRadius customer page (undated)

What to watch out for

  • Most results come from the software vendors themselves, and none shows how the figures were measured.
  • Some vendor pages carry no date. Check before relying on a figure.
  • Data quality comes first. Microsoft says building the AI was not the hard part; getting data and workflows in shape was.
  • Duplicate and fraud checks matter. WSH reports duplicate invoices flagged and suspicious emails flagged, which is a control worth asking any vendor about.
  • Plan the change for people. Microsoft ran role-based training and included frontline users from the start.

Common questions

How do companies automate accounts payable with AI?
In the examples here, software reads and labels incoming invoices and emails, codes and routes them, flags duplicates, and keeps an audit trail. Staff handle exceptions and supplier disputes.
What is AI invoice processing?
It means software extracts the data from an invoice, such as a PDF or an email, and turns it into a record for the accounting system. Billerud reports over 90% of its PDF invoices validated automatically with Basware Smart PDF.
What is AI cash application?
It is the matching of incoming payments to open invoices. Shiseido reports 91% straight-through cash application with HighRadius agents, and Microsoft reports payment matching accuracy rising from 40% to 90%.
Does AI replace the finance team?
In these examples it takes over routine work and the team moves to exceptions, disputes and supplier relationships. Microsoft describes its system as human-led and AI agent-assisted.
Which AI tools did these companies use?
Microsoft used SAP, Dynamics 365, its IQ layer and Copilot. WSH used AppZen. Billerud used Basware Smart PDF. Shiseido used HighRadius agents.

Sources

Related on Ai Siftly

Browse money and cost workflows and case studies. More guides: AI customer support automation examples and AI recruiting automation examples.