Skip to content
NextLog

AI-powered process intelligence

About 40% of your process is invisible

Process mining only sees what your systems record. The handoffs, clarifications, exceptions and approvals that actually drive your cycle times live in emails, notes, PDFs and reports — and never reach an event log. NextLog reads them and turns them into events.

The same process, as your event log records it and as it actually runs.

Why the gap matters

~40%
of process information sits in unstructured artifacts Emails, chat, free-text fields, PDFs, service reports, notes
2–4
weeks to a quantified proof of concept Using a representative sample of your historic communications
0
rip-and-replace required Event logs merge with your existing process mining platform

The problem

Process mining can only analyse what was recorded as an event

Conventional process mining excels where systems generate clean event logs. ERP order-to-cash, procure-to-pay, CRM case handling, workflow engines and industry platforms all produce structured records with timestamps, activities, case IDs, resources and attributes. From that data, mining tools reconstruct the as-is process, quantify throughput and waiting time, identify variants, and check conformance against the expected model.

The limitation is structural, not a matter of data quality. If a handoff, a decision, a clarification or an exception is handled through an email thread, a PDF form, a note field or a chat exchange, it typically never appears in the event log at all. The mined model can therefore look clean and linear while the real process is iterative, delay-prone and full of rework.

Those invisible steps are rarely incidental. They are where delays occur, while people wait for responses. They are where quality issues get documented, in notes and comments. They are where compliance evidence lives, in reports and approval trails. And they are where exceptions get resolved, through human coordination rather than system transactions. Optimising only the visible portion of a process risks solving the wrong problem — confidently.

Three trends make this urgent now

  • 01

    Process mining has hit diminishing returns

    Adoption is mainstream and the easy wins are taken. The problems that remain live outside system logs, which is precisely why they have survived.

  • 02

    LLMs are finally reliable enough

    Language models have reached the maturity needed for dependable text understanding at enterprise scale and cost — reading thousands of messages is no longer impractical.

  • 03

    Audit pressure keeps rising

    Regulators increasingly expect complete process documentation. Much of that evidence already exists, unstructured, in communications and documents.

Before and after

From a tidy fiction to the operational truth

On the left, the process as structured system logs describe it: a short, linear path. On the right, the same process once unstructured interactions are extracted, attributed to cases and classified — the version your teams actually work in.

Process diagram of a sales cycle as structured system logs record it: customer request, proposal, sales contract, sales order, delivery, billing, payment — a single linear chain of seven steps.
Traditional process mining: only system-recorded events
The same sales cycle after NextLog processing. The seven system-recorded steps remain, joined by eleven steps recovered from email — order confirmation, clarify conditions, answer from sales, clarify production costs, answer from production, invoice creation, send invoice to sales, approval guideline, request approval and approval — drawn with dashed borders and envelope icons.
With NextLog: system events plus recovered unstructured steps

The solution

NextLog turns unstructured communication into structured, case-linked events

NextLog connects to the text your organisation already produces — email, tickets, document repositories, free-text system fields — and extracts the process-relevant facts: who did what, when, and for which case. It assigns each interaction to the correct business case even when no explicit case ID exists, classifies it into a process activity, and generates event logs in the formats standard process mining platforms already ingest.

For executives

Decisions get made on the full operational truth: true cycle times including communication delays, transparent exception handling, and an evidence trail that stands up in an audit.

For technical decision makers

A deployable, modular architecture that fits your existing data landscape and exports to established process mining toolchains. No rip-and-replace, no proprietary lock-in on the analytics side.

  • Process discovery from unstructured sources
  • Automated event logs for process mining tools
  • End-to-end visibility across systems and human interaction
  • Confidence scoring and full auditability for every AI output

How it works

Four modules, one pipeline

Raw text enters at one end; a case-aligned, confidence-scored event log leaves at the other. Each stage is independently callable, so you can run the whole pipeline or just the part you need.

  1. Module 1

    Summarization

    Long documents, email threads and attachments are condensed to their event-relevant essentials: the actions, decisions, requests and outcomes. This keeps downstream stages accurate and affordable on large corpora.

  2. Module 2

    Case attribution

    The foundational stage — everything downstream depends on getting it right. Where messages already carry order numbers, ticket numbers, claim IDs or similar, NextLog ingests and assigns them directly. Where no identifier exists, it infers the case from the content itself.

  3. Module 3

    Process definition

    The Event Master is the authoritative catalogue of process activities. NextLog can draft it for you by analysing the summarized corpus for recurring patterns and implicitly described activities, and you refine it interactively.

  4. Module 4

    Classification

    Each message is compared against the Event Master, in the context of what happened before and after it in the case. Linguistic cues and implicit process signals are weighed, and every classification carries a confidence score.

Inside the product

The React interface is where analysts validate what the pipeline produced — every run, every classification, every confidence score, traceable back to the message it came from.

Case attribution
NextLog case attribution screen showing attribution statistics, an execution history table of completed runs, and a results table listing messages with subject, direction, sender, recipient, assigned case ID and a confidence percentage.
Case attribution Run statistics, full execution history, and every message with its inferred case ID and confidence score.
Process discovery
NextLog process discovery screen showing automatically suggested process activities derived from the analysed message corpus, ready to be edited into an Event Master.
Process discovery NextLog proposes an Event Master from your own corpus; you accept, rename, split or merge.
Message classification
NextLog message classification screen showing messages mapped to process activities with per-message confidence scores.
Message classification Each message mapped to an activity, in case context, with the confidence behind the decision.
Process visualization
NextLog process visualization screen showing a process mining graph built from mixed events, with activity nodes such as product inquiry, quotation request, order received, invoice sent, payment received and customer complaint, mean durations on the connecting edges, and a process metrics panel listing 69 cases, 16 activities and 23 connections.
Process visualization The merged event log rendered as a process graph, with case counts and mean durations on every edge.

Governance

The Event Master: your activity taxonomy, in plain language

Classification is only as good as the definitions behind it. The Event Master is a curated list of the events that matter in your process, each with a short description — and it is designed to be owned by the people who understand the process, not only by the people who can write code.

The result is a move away from ad-hoc classification toward a stable, governance-ready activity catalogue that stays aligned with how work is really executed.

User-editable
Analysts and process owners maintain it directly through the web interface.
Defined in natural language
Events are described in plain sentences. There is no schema to learn and no technical constraint on how an activity may be expressed.
Manual or AI-assisted
Start from scratch, or let NextLog propose a draft taxonomy from your own corpus and edit from there.
Refined in a fast loop
Adjust definitions, granularity and naming conventions by instruction — "split this into two activities", "treat these as one". NextLog interprets and updates the event set accordingly.

Platform

Modular, API-first, and deliberately boring where it counts

NextLog is built as Python processing modules with a REST layer over them and a React interface on top. Nothing about the design assumes it is the centre of your stack — it is meant to slot into one.

Python processing modules

The core. Extraction, attribution, classification and logging, each with clearly defined entry points for starting runs, retrieving results and querying audit artifacts. Callable from the command line or from a scheduler, automation framework or integration platform.

REST API

A built-in server exposing every core function through one unified API that mirrors the underlying modules. NextLog can therefore run as a headless processing engine or as a service inside a larger automation landscape.

React frontend

Operational access to summaries, classifications, case attributions and run histories — for exploratory validation, monitoring and debugging by analysts, operations teams and data stewards.

PostgreSQL persistence

Entry tables for incoming raw artifacts; processing results with confidence scores; audit and execution history capturing prompts, model responses and runtime metadata; and output tables and views in a consistent, process-mining-ready structure.

Provider-agnostic AI layer

Any OpenAI-compatible provider, local or remote, can be used interchangeably — and combined within a single run for stage-specific model selection, fallback strategies and provider failover. LLM dependencies stay isolated from application logic.

Template-driven prompting

Prompts are templates supporting domain-context injection, validation constraints, few-shot examples and structured output guards. Every run is captured end to end, including parameters, model selection and error states.

Integration

A deliberately narrow interface: PostgreSQL in, event log out

NextLog does not reach into your email server, your document repository or your ticketing tool. Instead it reads exclusively from dedicated PostgreSQL entry tables that you populate with the text to be processed and its metadata — source system, timestamps, participants, any available case hints.

Separating source-system connectivity from AI processing is a deliberate choice. It makes ingestion predictable and auditable, keeps NextLog out of scope for your mail and document platform security reviews, and means the pipeline behaves identically no matter where content originated.

On the way out

Results are written back to PostgreSQL as consolidated tables and views in a process-mining-ready event-log structure, exportable as CSV or connected directly over standard database connectivity. Clean separation between ingestion, AI processing and downstream analytics.

PostgreSQL
entry tables
NextLog Event log
views & CSV

Deployment options

On-premises

Full data control for strict residency requirements.

Cloud

On the major providers, for operational flexibility.

Hybrid

Processing stays local while model calls use cloud services with your own keys.

Docker-based deployment is recommended for consistency and portability, with separate containers for backend, database and frontend, orchestrated by Docker Compose or Kubernetes depending on scale and governance needs.

Security and compliance

Built to survive an audit, not just a demo

Your credentials, your exposure
Access to AI providers uses customer-owned credentials, so you keep full control over what data leaves your estate and which external APIs are used.
Encrypted in transit and at rest
All internal communication uses TLS; the PostgreSQL backend supports encryption at rest.
Every step is auditable
Input artifacts, generated prompts, model outputs, execution metadata and error traces are all logged — enabling behaviour monitoring, traceability and regulatory evidence.
Confidence-scored, not black-box
Classifications carry confidence scores and remain reviewable, which is what makes controlled use in regulated environments possible.
Scales without drama
Concurrency across multiple workers, background execution and batch processing for large datasets. Incremental ingestion means only new artifacts are processed.
Standard interfaces only
Consolidated event logs export as PostgreSQL tables, CSV files or pipeline-friendly views — no proprietary format between you and your analytics.

Use cases

Wherever the process runs on text

NextLog is broadly applicable, and most valuable wherever unstructured data drives process execution. Six areas where that is consistently true:

Health care and life sciences

Clinical workflows span disconnected systems and lean heavily on human coordination. NextLog extracts and links steps such as insurance verification, patient preparation, quality checks, specialist queries and critical alerts — then merges them with EHR, RIS and PACS logs for a complete order-to-report view. For medical device service lifecycles, technician reports become a structured history across installation, preventive maintenance, repair, parts procurement, validation and closeout, enabling accurate MTTR that includes parts delays.

Illustrative example: in one medical imaging workflow, 8 of 13 real process steps (about 62%) existed only in emails and clinical notes.

Logistics and postal services

Scan events are plentiful, but the decisive steps happen between the scans — driver notes, depot handover comments, customer emails and SMS, call-centre summaries, customs documentation, partner communications. NextLog connects these to shipment or case IDs, exposing the real flow behind failed delivery attempts, address corrections, missed cutoffs, customs holds, damage claims and rerouting. Merged with TMS, WMS and track-and-trace data, it quantifies hidden dwell time and pinpoints exactly which handoff breaks down.

Financial services

In lending and underwriting, much of the work happens outside core systems: document requests, clarification exchanges, exception discussions. NextLog attributes these to application IDs and classifies them into activities such as document request, income verification, credit review and exception approval — making collection delays measurable and exception loops visible. In claims, adjuster notes and third-party correspondence reconstruct the investigation and negotiation steps that system logs omit entirely.

Professional services

Consulting, legal and other knowledge work is managed through email, meeting notes and status reports. NextLog classifies these into activities such as client request, analysis, review cycle, approval and delivery, so actual project flow can be compared against the planned methodology. That quantifies review-cycle bottlenecks, measures client response times against SLAs, and surfaces early warning signals on at-risk engagements.

Manufacturing

Quality management and non-conformance processes are documented in inspection notes, corrective action reports and supplier correspondence. NextLog turns that evidence into analysable event logs supporting faster resolution, recurring root-cause identification and supplier performance insight. For equipment maintenance, it extracts troubleshooting steps, parts delays, workarounds and validation activities — real downtime analysis rather than the open/close timestamps of a work order.

Public sector

Case management relies on correspondence and documents moving between agencies and external stakeholders. NextLog reconstructs the real case lifecycle from communications, memos and forms, supporting transparency, faster turnaround and demonstrable compliance with procedural requirements in citizen services and regulatory workflows.

Outcomes

What changes once the hidden steps are visible

Full process visibility
Discover bottlenecks in handoffs, measure true cycle time, identify the variants that actually matter, and understand root causes documented in communication rather than in transactions.
Faster cycle times
Communication-driven delays become measurable, and therefore optimisable: standardise information requests, cut clarification loops, improve handoff protocols, automate follow-ups.
Simplified audit and compliance
Evidence scattered across emails, notes and reports is continuously structured into case-linked trails, reducing audit preparation effort and improving confidence in SOP adherence.
Rapid ROI
Historic unstructured data can be analysed immediately, so results arrive in weeks rather than months. Early findings give a business case concrete numbers.
Better customer experience
The delays and rework customers actually feel are usually the invisible ones. Fixing them improves satisfaction in a way that optimising visible steps cannot.
Continuous improvement
Recurring patterns feed back into refined activity definitions, so the taxonomy — and the insight — keeps improving as execution changes.

Getting there

A phased path, with value at the end of each phase

Successful implementations follow the same shape: prove it small, expand deliberately, then industrialise. Early value arrives without compromising enterprise readiness.

  1. 2–4 weeks

    Proof of concept

    Select a representative process and a manageable sample of communications. Validate extraction quality and quantify the improvement opportunity in real numbers.

  2. 4–8 weeks

    Pilot

    Expand scope, integrate production sources, refine activity definitions, and connect the outputs to your target process mining platform.

  3. 8–12 weeks

    Production rollout

    Scale processing, establish governance for taxonomies and prompts, deploy monitoring, and train the broader user group.

Who is behind NextLog

Two firms, one product

NextLog is built jointly by an AI solutions house and a management consultancy — deliberately, because closing this gap is as much a process governance problem as it is a machine learning one.

People at Work Systems AG

AI solutions and technology

Founded in 2004, with legal entities in Germany and Switzerland. Trusted advisory across customer relationship management, loyalty management, business intelligence and artificial intelligence — from consulting through to delivered solutions. Research cooperation on AI with Philipps University Marburg.

Unterföhring, Germany paw-systems.com ↗

Draxinger & Lentz

Management consulting and process governance

Founded in 1998, advising clients across business, public institutions and the wider economy as a trusted advisor. A holistic approach to corporate management, accounting, controlling and systems, grounded in systems-oriented management theory — bringing clarity to complexity and establishing effective control in organisations.

Frankfurt am Main, Germany draxingerlentz.de ↗

Next steps

Want to make the missing 40% usable?

The most effective way to evaluate fit is a guided demo followed by a focused proof of concept on a representative sample of your own unstructured process data. Within weeks you can expect a merged, process-mining-ready event log and a quantified set of improvement opportunities grounded in your real execution data.

Or reach us directly

Request a demo

Your enquiry goes directly to both partner firms. No newsletter, no tracking.

By submitting this form you agree that your details may be used to respond to your enquiry. See our privacy notice for how we handle your data. Privacy