M²I
M²I is a non-profit AI lab built to test innovative concepts for data analytics and AI, like conservation based risk and internal audit assiatants.
M²I also provides a suite of forensic data analytics tools which have been used for both criminal investigations and data subject access request (DSAR) related employment disputes.
Finding the needle in a haystack
Data Forensic Services
Whether for business analysis, crime detection and prevention, compliance and conduct investigations or other matters, model-based data analytics finds things the naked eye never could. In particular Subject Access Requests (SARs) often miss important data. Not always is that deliberate, but knowing whether something is missing means peace of mind.
Herfel Limited performs all kinds of data forensic services, but we are particularly proud of our data integrity models, which are built to detect omissions, redated documents or simply lies.
Our proprietary models
Lachman Test
Backdated documents are a staple of financial fraud, employment disputes, and contract litigation. The trick with the computer time settings is well known, but one can change the timestamps, not what happened between them. Version updates of Word or Excel, the specific build of every embedded plugin at time of save, and critically, the state of the spell-check and autocorrect dictionary in use.
Language-model dictionaries are updated on predictable release cycles our database knows exaclty when which word was added.
Ommission Perturbation
Most employees filing an SAR worry about data being ommitted. We found a way to identify were data could have been withheld or deleted. How? By ommitting data to an AI.
The Ommission Pertubation models is trained on the provided data in thousands of iterations. Every interation deliberately leaves out certain data, until the model is so good in spotting these, that it knows where the gaps in the full data set are.
Semantic Drift Chronometry
We can instinctively hear how someone is feeling from the way they talk - and write. Seismic chronometry can detect sudden shifts in syntax which is different from someone just being moody - a strong indicator for deception.
Moreover, analyses show that certain seemingly unrelated words are used more frequently after a tragedy like a terror attack or the death of a beloved monarch. That's carbon dating with subconscious language choices.
SEISMIC v3
Investigators know what to look for. SEISMIC finds what nobody thought to look for. It runs unsupervised correlation passes across heterogeneous datasets, whether transaction logs, HR records, building access data, email metadata, expense claims or trading position, and surfaces statistically improbable co-occurrences that no human analyst would have hypothesised.
The model doesn't need a theory. It generates them.
*all our models are probabilistic and can contain false positives and false negatives. Therefore we never say yes or no, but give a percentage probability. For clients who further want to investigate any gaps we offer to provide our models to the investigators
AI that actually adds value