PaperTrail AI engine

Meet Scout. AI at work on your documents.

Scout reads invoices, contracts, applications and claims, checks them against each other, pulls out what matters and moves the workflow forward, right inside PaperTrail. You choose the AI model and where your data goes, all the way down to a local LLM on your own servers.

See Scout on your documents

How Scout works

Live in customer workflows
Local LLM or cloud
No vendor lock-in

AI platforms, one interface

1 line

to switch AI provider

Local

LLM option: the AI runs on your own servers

Live

in production with PaperTrail customers today

The problem

Your documents are captured. Reading them still falls on people.

Your organisation runs on documents: invoices, purchase orders, contracts, applications, claims, reports. PaperTrail already captures, routes and stores them. But the hard part, reading them, checking them against each other, pulling out what matters and deciding what happens next, still falls on people.

Generative AI can do much of that reading. The obstacles are practical.

Which AI?

The market moves monthly. Committing to one vendor today risks lock-in tomorrow.

Where does the data go?

Some documents can go to a public cloud API. Some must stay in your domain. Some must never leave your building.

How does it fit in?

An AI chat window on the side doesn’t change a process. AI has to act inside the workflow, on the document in front of it.

Meet Scout

One instruction. Any document. The AI of your choice.

Scout is PaperTrail’s AI engine. Any PaperTrail rule or workflow step can hand a document to a large language model, together with its attachments and a plain-language instruction, and get the answer back where it’s needed: as a note on the document, a field value, or a decision that moves the workflow forward.

That’s the whole integration. No separate platform to buy, host or learn.

Reads the real document: PDFs, scans, Word and Excel files
Compares several labelled documents in one go
Carries the conversation forward for follow-up questions
Writes results back with a full audit trail

PaperTrail · Contracts · MSA-2026-114

Scout brief for approvers

Master services agreement, 36 months

Renewal: automatic, 90 days’ notice to cancel
Deviation: liability cap differs from your template (clause 14.2)
Missing: signed annexure B (service levels)

ApproveSend to legal

Scout Search

Ask your archive anything. Without sharing a single document.

Scout Search lets anyone find documents the way they think about them: “supplier contracts that auto-renew before December”, “invoices from Highveld still waiting for approval”. It searches metadata and the contents of every document together, interactively, and it works without sharing your documents with a language model.

Plain language, not search syntax. Ask the way you would ask a colleague, then refine the question as you go.
Metadata and contents together. Index fields, document text and scanned pages are searched in one go.
Respects permissions. People only ever see results they are allowed to open.
Nothing shared. Scout Search does not send your documents to any AI model.

Supplier contracts that auto-renew before December

MSA · Northwind FacilitiesAuto-renews 30 Nov · 90 days’ notice
SLA · Cloudline HostingAuto-renews 15 Nov · 60 days’ notice
Supply · Highveld PackagingAuto-renews 01 Dec · 30 days’ notice

Scout Search · searched inside PaperTrail, no documents shared with an AI model

Live with customers in 2026

Already at work in PaperTrail.

LiveScout SearchInteractive, plain-language search across metadata and document contents. No documents shared with an AI model.See how it works →
LiveCV screening and rankingIn PeopleHub, Scout reads every CV for a vacancy, summarises it, filters on must-haves, scores each candidate against the job description and ranks them.See how it works →
LiveContract risk analysisIn Contract Management, Scout scores every contract for risk and explains high-risk clauses in plain language.See how it works →

Private by design

Your data, your rules. Down to every document.

Scout runs public cloud, private cloud and on-premises models side by side, and you decide per process where each document goes. Send routine correspondence to a fast cloud model, keep regulated records inside your own AWS region, and give the most sensitive material to a local LLM running on your own hardware. Documents you don’t want AI to see are simply never sent.

Sharing is always your choice. Scout Search never needs it. Reading, checking and summarising documents only happens for the processes where you switch it on, with the model and location you choose.

Supplier correspondencePublic cloud AIClaude, Gemini or Azure OpenAI
Invoices & claimsYour private cloudAWS Bedrock in your own region
HR & legal filesYour own serversLocal LLM: data never leaves your network
Board papersNo AIScout is never called

Run a local LLM

Host the model yourself with Ollama. Every prompt and every page stays on your network, with no internet round-trip.

Keep it private

Use AWS Bedrock to keep AI traffic inside your own AWS account and region, under your own security controls.

Share only what you choose

Scout only sees what a rule hands it. You decide which processes, which documents and which attachments are sent, and to which model.

API keys held as PaperTrail system properties · Never hard-coded in scripts · Every AI step in the document’s audit trail · People approve what matters

Choose your AI

Pick the best model today. Change your mind tomorrow.

Scout works with the leading AI platforms through one consistent interface. Switching provider is a one-line change: your instructions, your workflows and your documents stay exactly as they are.

When a better or cheaper model appears, you adopt it. No re-implementation, no migration project.

CloudAnthropic ClaudeDeep reasoning over long, complex documents
CloudGoogle GeminiFast, cost-effective processing at volume
CloudMicrosoft Azure OpenAIOrganisations standardised on Microsoft and Azure (Copilot)
Private cloudAWS BedrockKeeping AI traffic inside your own AWS account and region
On-premisesOllamaFully on-premises, self-hosted models. Data never leaves your network

Cost control

AI spend you set. Not spend you discover.

AI is billed by the token, and costs can creep up quietly. Scout puts you in charge: limit tokens, set budgets per process and get warned before spend gets out of hand.

Limit tokens per callCap how much each AI call can read and write, so one long document can’t run up the bill.
Budgets per processGive each process its own token budget, and decide what happens when it runs out.
Early spend warningsGet alerted as spend approaches a limit, before the invoice from your AI provider arrives.
Usage on every callEvery response reports its input and output tokens, so you see exactly where the spend goes.
Right model, right priceUse a fast, low-cost model for routine work and keep premium models for the hard cases.
Local means no meterA local LLM on your own servers has no per-token fees at all.

AI budget · this monthExample

Invoice matching1.24M of 2M tokens
Approver summaries0.38M of 1M tokens
Contract review0.92M of 1M tokens
HR files (local LLM)No per-token fees

Spend warning: Contract review92% of this month’s token budget used. Finance and IT have been notified.

Max output per call2,000 tokens

What Scout does

Works in any business process that runs on documents.

Finance, legal, HR, procurement, claims, compliance or a process unique to you: if a person reads it to decide what happens next, Scout can help.

Three-way matching

Compare invoices, purchase orders and delivery notes and flag discrepancies before payment.

Contract review

Summarise obligations, renewal dates and non-standard clauses, and highlight deviations from your templates.

Intelligent classification

Read an incoming document and decide its type, priority and where it should be routed.

Data extraction

Pull names, amounts, dates and reference numbers into index fields and eliminate manual capture.

Compliance checks

Verify that an application or claim contains everything your policy requires, and list what’s missing.

Summaries for approvers

Give every approver a short, plain-language brief at the top of the document so decisions are faster.

Correspondence drafting

Draft responses, queries or rejection letters based on the document and its history.

Your own process

Because the instruction can be stored as a PaperTrail document, business users refine what Scout is asked to do without touching code.

Why organisations use Scout

From a place where documents are stored to a place where they are understood.

Scout takes the reading, checking and capturing off your people, and leaves them the judgement calls. It is already running in production in PaperTrail customer workflows.

Faster decisionsApprovers start from a clear brief instead of reading every page.
Less manual captureAmounts, dates and references land in index fields automatically.
Fewer costly errorsMismatches and missing documents are caught before money moves.
Predictable AI costsToken limits, budgets and spend warnings keep AI spend where you planned it.
Workflows never stallA failed AI call returns a clear result. It never halts a rule or workflow step.
Future-proofNew and better models plug in without changing your processes.

How Scout works

Built into PaperTrail. Not bolted on.

Scout is a lightweight engine that runs inside PaperTrail and plugs into the rules and workflows your team already uses.

01
A document arrives
It enters a PaperTrail workflow as usual, from scanning, email or an eForm.
02
A rule calls Scout
The rule or workflow step sends the document, its attachments and a plain-language instruction.
03
The answer is written back
As a note on the document, an index value, or a routing decision.
04
People stay in control
Your team reviews and approves AI output in the familiar PaperTrail screens, with every step in the audit trail.

Built for documents

Not just text.

Scout sends the real document to the model: PDFs, scanned images, Word and Excel files. Where a provider can read a format natively, Scout sends it natively; where it can’t, Scout falls back to extracted text automatically.

Attach as many documents as the task needs, each with a meaningful label such as “Tax Invoice”, “Purchase Order” or “Signed Contract”, so the model knows what it is looking at. Ask follow-ups and build a review over several steps; the full context carries forward.

Reliable by design

Made for production.

Never breaks the workflow. A failed AI call comes back as a clear, inspectable result.
Cost visibility. Input and output token usage on every response.
Predictable behaviour. Output limits, timeouts and temperature set per call, with sensible defaults.
Secure credentials. API keys stored as PaperTrail system properties.

Scout at a glance

The short version.

Already working with your PaperTrail system? Scout can be added to the rules and workflows you run today.

Book a Scout demo

What	AI document intelligence built into PaperTrail
Providers	Anthropic Claude, Google Gemini, Azure OpenAI, AWS Bedrock, Ollama (on-premises)
Inputs	Instructions, PDFs, images, Office documents, conversation history
Outputs	Model response, token usage, completion status
Deployment	Cloud, private cloud (your AWS region) or fully on-premises
Lock-in	None. Switch providers with a one-line change