Articles grouped by business problem, industry, and system type.
AI for duplicate detection suggests a related case from another channel before an employee repeats work already done. It has to tell the same problem described in other words from a similar-sounding ticket about a different event. In a scenario of 3,000 tickets a month, linking channels and the order number cuts errors from 8% to 4%, and fine-tuning to 2%, which together saves 45 hours of corrections, or €1,571 a month.
An estimator receives a list of works described differently from the company's catalogue and has to find the matching items. AI suggests entries with the description, unit and scope next to the source row, and the estimator approves the match. When the codes are shared, an import is enough. A matching app costs from €3,000 net at Syntalith and pays back in 1.5 months at 2,000 rows a month at 2 minutes each.
AI can point out a gap in a report while the technician is still writing it, such as a missing test print result. First try a separate question in the form, because a field filled with „done” still does not describe the result. In a scenario of 1,500 visit reports a month, the form question cuts errors from 15% to 7%, and fine-tuning to 3%, saving 30 hours of corrections, or €1,047 a month.
AI will read a handwritten form when ready OCR supports handwriting in that language and an employee checks the numbers beside a crop of the page. Customising the model makes sense when the same error keeps coming back on the company's forms. At Syntalith an application for reading and correction costs from €3,000 net. At 1,500 forms a month, it pays back in 1.8 months.
AI points to the passages in an IT ticket about the impact on work and the urgency, for example a blocked order-handling team, and to the gaps to ask about. The duty agent sets the priority by company rules. In a scenario with 1,500 tickets a month, a form with impact and urgency fields cuts errors from 20% to 8%, and fine-tuning to 4%, which together save 24 hours of corrections, €838 a month.
An AI model can split a PDF with several documents into separate index entries, each with the document type and page range. First check a ready classifier and the features of your current document system, and consider customisation for recurring mistakes. In a scenario of 400 PDF bundles a month, a ready classifier cuts errors from 30% to 12%, and fine-tuning to 5%, which together saves 17 hours of corrections, or €582 a month.
A shared report of machine messages needs the device family beside every code, because E17 concerns connectivity in one family and a sensor in another. A lookup table reads known codes, and a model helps with descriptions that have no readable code. In a scenario of 2,000 messages a month, a lookup table with the device family cuts errors from 12% to 5%, and fine-tuning to 2%, which together saves 20 hours of corrections, or €698 a month.
AI helps you find a person with specific project experience, for example running requirements workshops, by finding the passages in the documentation where that work is described under different names. The result is a starting point for a conversation about experience, with a link to the source.
AI recognises from a technician's note the type of work done, for example an adjustment or a module replacement, and quotes the sentence. An ordinary rule then picks the list of required attachments, so the office asks only for the material needed. In a scenario with 800 visit records a month, a field for the type of work done cuts errors from 15% to 6%, and fine-tuning to 3%, which together save 24 hours of corrections, €838 a month.
An AI model narrows down the matching of products from a supplier catalogue to your own item master when names and codes differ. First check search by codes, names and packaging levels, and consider your own adaptation when the model confuses products that differ in packaging or variant.
AI for supplier quotes keeps the full delivery term: six weeks and the event they run from, for example drawing approval. Without that part the table looks finished and planning goes wrong. In a scenario with 600 quotes a month, a shared form for suppliers cuts errors from 15% to 6%, and fine-tuning to 3%, which together save 12 hours of corrections, €419 a month.
When the team keeps shortening and rearranging AI replies, first name the edits that recur and add them to style guidance with examples. Fine-tuning helps when the model returns to the same habits despite that guidance. In a scenario of 2,000 customer replies a month, style guidance with examples cuts errors from 30% to 12%, and fine-tuning to 5%, which together save 16.7 hours of corrections, or €582 a month.
A construction firm usually does not need to train a model further for AI to understand abbreviations in site notes, such as „TBC”. A glossary of abbreviations and the project context work first, and fine-tuning makes sense when the same mistakes come back despite them. At 600 notes a month, a fall in wrong categories from 12% to 2% saves 6 hours of corrections, or €209.
Comparing contract versions marks the changes, and an AI model can give them categories so the coordinator knows whom to pass each difference to. The frequency of a report is a substantive change, and a new paragraph break is formatting. In a scenario of 600 contract changes a month, agreed category definitions cut errors from 20% to 8%, and fine-tuning to 4%, which together save 8 hours of corrections, or €279 a month.
With insurance forms the most common error is a value read correctly but put in the wrong field, such as the head office entered as the warehouse location. First check a reading template and a ready-made extractor with examples. Adapt the model when assignment errors return on legible forms. At 2,000 forms a month, an extractor cuts errors from 15% to 7% and fine-tuning to 3%, saving 24 hours, or €838 a month.
When technicians describe similar failures in different words, first check shared event codes and a ready model with a described list of groups. Fine-tuning makes sense when plant abbreviations still confuse the model. In a scenario of 1,500 work orders a month, a ready model with a list of groups cuts errors from 15% to 6%, and fine-tuning to 3%, which together save 12 hours of corrections, or €419 a month.
With notes written in Polish and English, AI has to keep the state of a case: „ordered” and „ETA to be confirmed” do not mean „in transit”. In a scenario of 800 service notes a month, a glossary and instructions cut errors from 12% to 6%, and fine-tuning to 2%, which together save 6.7 hours of corrections, or €233 a month.
AI splits an RFP question into parts when one sentence needs several authors, for example where backups are stored and who approves deleting them. The company's responsibility list assigns the team, and the model recognises the topic. In a scenario of 400 RFP questions a month, a topic list with a description of responsibilities cuts errors from 15% to 7%, and fine-tuning to 3%, which together save 12 hours of corrections, or €419 a month.
A service call summary has to tell what was done from what was only agreed. First fix the format: separate sections for completed actions and for next steps. Fine-tuning makes sense when the model still confuses a plan with its completion. In a scenario of 2,000 service calls a month, a format with separate sections cuts errors from 12% to 6%, and fine-tuning to 3%, which together save 15 hours of corrections, or €524 a month.
AI can group quality complaints that describe the same problem in different words. Start by agreeing what the categories mean and by trying a ready model with their description. Consider fine-tuning when specific mistakes keep coming back on complaints the model has not seen.
Private models
We select the hardware and model, deploy it and measure quality on your examples. You can start with a computer you already own.
Times are shown in your own time zone. We work with clients across time zones.
See private modelsOpenAI makes ChatGPT and Anthropic makes Claude. We belong to the partnership programs of both companies, so we work with their tools every day and learn from the creators themselves.
Claude Partner NetworkParticipation in programs does not mean that OpenAI or Anthropic endorses our services.