IMPROBABLY.ART / SELECTED INTERVENTIONS
Selected Interventions.
Not every automation deserves a case study.
Some simply deserve to stop consuming Friday afternoons.
Deployed systems appear first. Typical intervention patterns are labelled accordingly.
CASE 001 / CLAIMS ADMINISTRATION / DEPLOYED
An email arrived.
Someone used to have to turn it into a claim.
Retailer claims for BSH home appliances arrived by email as largely unstructured information.
Each message contained roughly fifteen relevant data points — claim references, product information, serial numbers, purchase dates, fault descriptions and customer details.
A person then had to:
OPEN → READ → IDENTIFY → COPY → CHECK → REFORMAT → ENTER → SUBMITThe destination was Repairsys, the manufacturer’s claims system.
Nothing about the process was particularly difficult.
It merely required a human being to perform the same careful transcription repeatedly.
The intervention
The workflow was rebuilt as an automated administrative process.
Incoming claim emails are now read automatically.
The system extracts the required fields from the unstructured message, distinguishes between similar identifiers, normalises dates, cleans problematic values and supplements the fixed information required for registration.
The resulting data is validated, structured and submitted directly into Repairsys.
The process becomes:
EMAIL → EXTRACT → CLEAN → VALIDATE → STRUCTURE → REPAIRSYSHuman involvement is reserved for cases that cannot be processed reliably.
The result
At approximately 100 claims per month, the former process would consume roughly 12–20 hours of administrative handling.
The automation does not make claims processing more intellectually sophisticated.
It makes most of the transcription disappear.
That was the objective.
OUTLOOK → AI EXTRACTION → VALIDATION → DATA NORMALISATION → API → REPAIRSYSCASE 002 / DOCUMENT PROCESSING / TYPICAL INTERVENTION
The employee had become a routing protocol.
Incoming documents were manually opened, renamed, classified, registered and sent to the appropriate department.
The intervention automated the standard path:
RECEIVE → EXTRACT → VALIDATE → CLASSIFY → RECORD → ROUTEOnly incomplete or ambiguous documents required human review.
The employee remained.
The routing career did not.
CASE 003 / CRM ADMINISTRATION / TYPICAL INTERVENTION
Sales had developed a second profession: maintaining the CRM.
After each interaction, information had to be entered manually, fields updated and follow-up tasks created.
The intervention connected incoming information directly with the CRM workflow.
Standard records were updated automatically.
Missing or contradictory information was flagged.
Sales returned to the minor matter of selling.
CASE 004 / RECURRING REPORTING / TYPICAL INTERVENTION
The organisation already possessed the information.
Unfortunately, every Friday someone had to locate it.
Data from several sources was collected manually and reconstructed into a recurring report.
The intervention automated collection, validation and report assembly.
The information did not become more intelligent.
It merely stopped requiring archaeology.
CASE 005 / EMAIL TRIAGE / TYPICAL INTERVENTION
A shared inbox received routine requests, documents and administrative questions.
Each message had to be opened, interpreted and forwarded.
The intervention classified incoming messages, extracted relevant information and routed standard categories automatically.
Uncertain cases remained visible for review.