Millions of documents.
The days that matter.
itsmt searches the e-mails, scans and files of a case – even where text recognition misread characters –, analyses the hits over time and shows on which days something unusual happened. For avoidance claims (transactions the administrator can reverse, §§ 129 ff. of the German Insolvency Code), directors’ liability for payments made after insolvency (§ 15b InsO), and anything else that needs evidence from large data sets.
Unusual days
Nobody reads three million e-mails. You need to know where to start.
In a case, mailboxes, server extracts and scanned binders come together, often several terabytes. The search is for signs of payment difficulties, arrangements, collateral. itsmt turns this into a manageable selection – and gives the reasons for it.
From the storage medium to the list of days you need to look at
Import
E-mails with attachments, Office files, PDF, images and archives are unpacked. The folder structure is preserved.
.eml .msg .docx .xlsx .pdf .tif .jpg .zipText recognition
Scanned pages and images are made readable by OCR. The result is stored next to the original.
Scan.pdf → Scan_ocr.txtError-tolerant search
All terms are searched in a single pass, even in poorly recognised pages. A misread character does not hide a hit.
insolvency ≈ lnsolvencvAnalysis
Hits over time, the tone of the correspondence and statistically unusual days. Every document is one click away.
effect size Δ · p < 0.01Text recognition misreads. The search still finds it.
Try it: the letter on the right is what text recognition reads from a poor scan. Yellow marks exact hits, orange the error-tolerant ones. Short terms are searched exactly, longer ones even with one or two differing characters.
While the search runs, you can watch it at work
Six views in the browser show the current run. For demonstrations, names and search terms can be replaced and preview images pixelated – with a single key press.
Status
Progress with time remaining, throughput and server load.
Factory floor
Each thread is a tower: blue and tall while it works, flat and yellow while it waits for text recognition.
Hit mountains
For each search term a ridge grows along the time axis: what was written about when.
Three views of the same hits
What was written about when?
Each bar shows how many documents of a month contain a search term. A click on a segment opens exactly those documents.
- E-mails with all attachments, Word, Excel, PDF pages and images directly in the browser, hits highlighted
- Viewed documents are marked, duplicates detected
- Copy what matters to your own target folder with one key
When does the tone turn?
Every word of the correspondence is rated positive or negative. Green is the weekly value, red the moving average, blue the individual days.
- Selectable measure: per word, per letter, positive and negative separately
- Spread as a band, moving average over 3 to 31 units
- A click on a day shows the letters of that day
Which days stand out?
Each day is compared with hundreds of randomly shuffled data sets: same terms, same amounts, only distributed differently. What chance does not explain is marked red.
- Effect size (Glass’s Δ) and significance per day, separately for each principal component
- For each day, the terms that drive it
- Checklist of all unusual days, printable report including the method
Even without a search: browse through the data
The file browser shows every folder of a case exactly as it is on the storage medium. E-mails, PDF, Word, Excel and images open with the space bar; terms can be highlighted.
What matters goes into a target folder and from there to your own computer: individually or as a ZIP.
The data stay where they belong
Your own server
itsmt runs on a Linux server. Reading is done in the browser, encrypted via https. No service outside the server sees the data.
Login with roles
Every user logs in and sees only their own cases. Roles control who may watch, who may switch the display and who may manage everything.
Audit log
Every login, every opened file, every copy and every download is recorded with user and time.
Demonstrable without disclosure
Names, paths and search terms can be replaced in the live views, preview images pixelated. You can show how the work is done without showing what it is about.
Developed in-house
Search, analysis and web interface are written in C without third-party libraries. What the software does can be traced down to the last detail.
Traceable
Every statistical statement names its method. The report on the unusual days also states how many of them would be expected by chance alone.
AI starts where itsmt has done the groundwork
For more than ten years we have been using AI to analyse insolvency and company data – always without third-party cloud or AI services, compatible with § 203 of the German Criminal Code and the GDPR.
Own AI in itsmt
Runs exclusively in-house: detection of junk images that would otherwise burden text recognition and review, and speech recognition that makes audio and video files searchable. More functions to follow.
Sort first, then assess
itsmt fishes the potentially relevant documents out of millions. Only this selection is assessed by an AI or a human – ideally both together.
Assessing AI: with us or with you
Rented on our servers in-house, or on-premise on your own hardware. Either way, the data stay on protected premises, never in someone else’s cloud.
From the practice of insolvency administration
itsmt is part of the INVEP-Quantum portfolio used and offered by TOMAREC. It is developed by André Koppel Software GmbH in Berlin, which has been building INVEP, the software for insolvency administrators in Germany, for many years. The questions itsmt answers come from their daily work.
Request a demo
We demonstrate itsmt on a sample data set or, if you wish, on your own data on our server.
TOMAREC Business Intelligence GmbH
Goerzallee 311 · 14167 Berlin · Germany