Case Study · Document Management
AI Purchase Order Automation Case Study: How Sanderson Turned Emailed POs into Reviewed, Finance-Ready Records
Sanderson PLC replaced manual purchase order entry with an automated process that extracts PO data from emailed attachments using AI Builder, presents it for finance review in a Power Apps canvas app, and exports approved records to its payroll and billing software.
PO data extracted automatically
Finance reviews, never retypes
Nothing submitted twice
Executive summary
Sanderson PLC's finance team received purchase orders as email attachments from a wide range of hiring sources, in inconsistent formats, across different mailboxes. Every PO was read by hand and rekeyed into the payroll and billing software, with incomplete documents chased before they could be processed. Inteknix built an automated process that captures POs from the shared mailbox, extracts the key fields using AI Builder regardless of document layout, and presents each record in a Power Apps canvas app for finance to review and approve. Approved records are exported in a single controlled action, with each entry flagged so nothing is processed twice.
At a glance
About the Client
Sanderson PLC is a recruitment group providing permanent and contract staffing services to client organisations across the UK, operating across multiple entities. Its finance team handles purchase order processing for contractor placements across the client base.
Inteknix is Sanderson's go-to team for business process improvement and automation, with a programme of work spanning process automation, AI, SharePoint migration and reporting.
Services used:
The challenge
Sanderson's finance team processed every purchase order by hand. PO details arrived as email attachments from a wide range of hiring sources, and each one had to be read, interpreted and rekeyed into the payroll and billing software. This caused:
Purchase orders spread across different mailboxes, split by operating entity
Attachments arriving in inconsistent formats and layouts, usually as PDFs but not always
Key fields — PO number, value, candidate name and the period covered — appearing in different places on every document
Incomplete documents, with candidate names sometimes missing and needing to be chased before processing could continue
Every PO manually rekeyed into the finance system, field by field
No central record of which purchase orders had been received, reviewed or entered
The approach
Inteknix started by mapping how purchase orders actually reached the finance team and what happened to them once they arrived. That surfaced the real obstacle: the documents themselves. Because POs originate from many different hiring sources, no two look alike, and a template-based or rule-based approach would break on the first unfamiliar layout.
That pointed to AI-based extraction rather than fixed parsing rules — a model that reads the document for meaning rather than position. It also confirmed that extraction alone was not enough. Purchase orders carry financial commitments, and documents arrive incomplete often enough that a fully unattended process would push errors straight into the finance system. The design therefore keeps a deliberate human checkpoint: automation does the reading and the data entry, while finance retains the judgement and the approval.
The final piece was pragmatism about sequencing. Rather than hold the benefit back until a direct integration with the payroll and billing software was built, Inteknix delivered a controlled export that lets finance move approved records across in a single step, with that integration following as a later phase.
The solution
Power Automate captures and moves the data, AI Builder reads the documents, SharePoint Online holds the records and files, and a Power Apps canvas app gives finance the place to review and release them. Together they cover the journey from emailed attachment to finance-ready record, through four core capabilities.
Automated capture from the shared mailbox
Purchase orders arriving in the shared mailbox are picked up automatically and the original attachment is filed into a SharePoint document library, so every PO is retained in one place and remains available alongside the record created from it.
AI-powered data extraction
AI Builder reads each attachment and extracts the fields finance needs, including the PO number, client, value and the period covered. Because the model interprets the document rather than relying on a fixed template, POs from different hiring sources are handled without configuring a new rule for each layout. The extracted values are written into a structured SharePoint record.
Review and approval in a canvas app
A Power Apps canvas app presents each extracted record to the finance team for checking against the source document. Anything missing or unclear — a candidate name absent from the original, for example — is visible at the point of review rather than discovered later, and records are marked as approved once confirmed.
Controlled export to the finance system
Approved records that have not previously been exported can be exported together in a single action, and each entry is flagged as exported once released. Finance imports the file into the payroll and billing software, with no risk of the same purchase order being submitted twice. Direct integration with the finance system is in progress as the next phase.
The outcome
Sanderson's finance team now works from a queue of pre-populated, reviewed records rather than a mailbox full of attachments to interpret.
The reading and the typing are handled automatically, leaving the team to do the part that genuinely needs a person: checking the extraction against the document and deciding whether it is right. Incomplete purchase orders surface at review rather than partway through data entry, so chasing a missing candidate name happens at a predictable point in the process instead of interrupting it. Every PO now has a single record and a stored copy of the original, giving finance a complete view of what has been received, what has been approved and what has been released to the finance system. Export tracking removes the risk of double entry.
Key results
"Sanderson and Inteknix have been developing solutions all of 2025. Inteknix team takes the time to truly understand how our core teams operate, then automated the right processes in a way that fits our day-to-day reality. For our Finance team, one of the busiest in the business, we've seen around a 20% reduction in admin time. The solutions are always supported by clear user documentation and robust technical solution documentation. Our Compliance and Contract teams have seen equally as impressive efficiency improvements"
"We now have reliable analytics that give us the visibility to make faster, better decisions. Inteknix are responsive, pragmatic, and consistently deliver. They are our go-to problem solvers across Microsoft 365, SharePoint, Power Platform and AI."
Mark Alford, Director of Technology, Sanderson PLC
Conclusion
Document-heavy finance processes resist conventional automation because the documents refuse to behave. Templates and parsing rules work until a new supplier sends a different layout, which is why so much purchase order processing is still done by reading and retyping. AI extraction changes that calculation — it handles variety without needing to be told about it in advance. What it does not change is where responsibility sits: financial data still needs a person to confirm it before it lands in the finance system. Sanderson's process now automates the reading and the typing while keeping the checking, which is the division of labour that makes the automation trustworthy enough to rely on.
Selected Work
Explore our recent work
FAQs
Frequently asked questions
Yes. AI-based document extraction interprets the content of a document rather than reading fixed positions, so it handles purchase orders from different suppliers and hiring sources without a separate template for each layout. This matters most where documents arrive from many external parties and no standard format can be enforced.
AI Builder is the AI capability within the Microsoft Power Platform. For document processing it identifies and extracts specified fields — such as a PO number, client, value or date range — from PDFs and images, and passes them into a structured destination like a SharePoint list. Because it is part of the Power Platform, the extracted data flows straight into Power Automate and Power Apps without separate integration work.
For purchase orders and other financial commitments, yes. Source documents arrive incomplete often enough that a fully unattended process would push errors directly into the finance system. A review step keeps the automation valuable — it removes the reading and the typing — while leaving confirmation with the person accountable for the data.
By surfacing the gap at the review stage rather than at the point of entry. When a field such as a candidate name is absent from the original document, the record shows it as missing so it can be chased as a known exception, instead of the omission being discovered partway through manual processing.
Yes, and it is often the faster route to value. Approved records can be exported in a controlled batch and imported into the finance system, delivering most of the time saving while a direct integration is built. Tracking which records have already been exported is what makes this reliable rather than risky.
By recording export status against each entry. Once a record has been released to the finance system it is flagged, and subsequent exports include only records that have not previously been sent. Duplicate submission becomes impossible by design rather than something the team has to watch for.
The measurable saving comes from removing document reading and manual data entry, which is where most of the time goes. Beyond time, automation creates a complete record of what has been received, reviewed and released, which manual entry into a finance system does not produce on its own. It also removes transcription risk from financial data, since values are checked against the source rather than typed from it.
A clear picture of where purchase orders arrive and in what forms, an agreed list of the fields that must be captured, and a decision about who approves the extracted data before it is released. Where documents come from many external sources, expect variety to be the norm and choose an extraction approach that tolerates it rather than one that assumes consistency.
Looking to remove manual data entry from your finance process?
Purchase orders, invoices and supplier documents arriving as email attachments are a common source of avoidable admin, and the variety of formats is usually what blocks automation. If your team is reading and rekeying documents into a finance system, explore how AI-powered extraction with a human approval step could take the manual work out while keeping the control where it belongs.
Practical advice, clear next steps, no obligation
