Every organisation runs on documents: invoices, purchase orders, contracts, CVs, applications and claims. PaperTrail has captured, routed and stored them for South African businesses since 1998. In 2026 we went a step further. With Scout, PaperTrail’s AI engine, documents are no longer only stored. They are understood.
What is Scout?
Scout is the AI engine built into PaperTrail. It does two things. First, it lets people search their documents in plain language, the way they would ask a colleague. Second, when you choose to switch it on, it hands a document, its attachments and a plain-language instruction to a large language model, and writes the answer back where it is needed: as a note on the document, a value in an index field, or a decision that moves the workflow forward.
Scout is already live with PaperTrail customers. Here is what it does, and why it was built the way it was.
Scout Search: ask your archive anything
Most document searches fail because people don’t remember file names, reference numbers or which folder something was saved in. Scout Search lets anyone type questions such as “supplier contracts that auto-renew before December” or “invoices from Highveld still waiting for approval”, and refine the question as they go.
- It searches metadata and document contents together, including scanned pages.
- It is interactive, so you can narrow results with follow-up questions.
- It does not share your documents with an AI model.
AI that works inside your workflows
An AI chat window on the side of your screen doesn’t change a process. Scout works inside PaperTrail rules and workflow steps, on the document in front of it. That makes it useful in almost any process that runs on documents:
- Recruitment (live in PeopleHub): Scout reads every CV for a vacancy, summarises it, filters on the must-haves, scores each candidate against the job description and ranks them, best fit first.
- Contract risk analysis (live in Contract Management): every contract gets a risk score, with high-risk clauses such as uncapped liability, one-sided indemnities and auto-renewals explained in plain language.
- Accounts payable: invoices, purchase orders and delivery notes compared line by line, with differences flagged before payment.
- Leasing: application packs checked for missing documents, with income and commitments pulled into the application.
- Travel: supplier invoices in any layout matched to bookings, with overbilling and duplicates flagged.
- Approvals: a short, plain-language brief at the top of every document that needs a decision.
Because each instruction can itself be stored as a PaperTrail document, business users can refine what Scout is asked to do without touching code.
Your data, your rules
The first question every client asks about AI is where their documents go. With Scout the answer is simple: only where you decide.
- Search shares nothing. Scout Search works without sending your documents to any AI model.
- Sharing is your choice, per process. Documents are only shared with a language model for the processes where you switch it on.
- You choose the model and where it runs. Anthropic Claude, Google Gemini, Microsoft Azure OpenAI, AWS Bedrock in your own AWS account and region, or a local LLM running on your own servers with Ollama, where data never leaves your network.
- People stay in control. AI output is reviewed and approved in the PaperTrail screens your team already knows, and every step is recorded in the document’s audit trail.
No lock-in, and costs you control
The AI market changes every month. Scout talks to every supported platform through one interface, so switching provider is a one-line change: your instructions, workflows and documents stay exactly as they are. Every response reports the tokens it used, output limits can be set per call, and budgets and spend warnings keep AI costs where you planned them. A local LLM has no per-token fees at all.
Built for production
Scout sends the real document to the model, including PDFs, scans, Word and Excel files, and falls back to extracted text automatically when a model can’t read a format natively. A failed AI call never halts a rule or workflow: it comes back as a clear, inspectable result. API keys are held as PaperTrail system properties, never in scripts.
See Scout on your own documents
Scout is available to PaperTrail customers and is part of every custom solution we build. The fastest way to understand it is to see it working on documents from your own process.