Use Cases

Concrete solutions we built for South Tyrolean SMEs. Anonymised by sector and region — the pattern transfers to other businesses.

How to read these use cases

The examples are intentionally anonymised. What matters is not the individual client, but the pattern behind it: recurring work, scattered data, manual review and a process that can become much leaner through software plus AI.

›real process patterns from operational projects
›no one-size-fits-all — every project is adapted to data, systems and approvals
›a practical starting point for a first MVP with measurable value

Process

All Quoting (1)Collections (1)Documents (5)Field workflow (2)Forecasting (4)Communication (1)

Industry

All Crafts (1)Trade (5)Industry (3)Agriculture (2)Tourism (1)Energy (2)

Power generation and self-consumption · Italy

Automated energy declaration to Agenzia Dogane (ADM)

Situation: The annual Dichiarazione Energia Elettrica to Agenzia Dogane Monopoli is mandatory for producers, self-consumption plants and large consumers. Consumption data from multiple sources is manually translated into the Tracciato XML, checked by the tax advisor, signed and uploaded through the ADM portal — error-prone, time-consuming and a stress point every year.
Solution: We build a lean workflow that translates consumption and generation data from ERP or plant monitoring directly into ADM’s XML schema — including XSD validation, qualified signature and system-to-system submission through ADM web services. Errors become visible before upload, and an audit log documents every step.
Outcome: What used to take days runs in minutes — with consistent quality because the schema is validated before submission. The tax advisor gets a clean preview for approval, and the consumption-data path remains traceable for future audits.

B2B information services · South Tyrol

Sales outreach emails from your own data — in seconds instead of hours

Situation: Personalised sales emails are assembled recipient by recipient. Someone has to search four data sources for suitable examples, write a friendly message and check opt-out status — 15 to 30 minutes per email.
Solution: We connect the internal data sources with an AI module that automatically selects the most relevant examples for each recipient — by trade, region and industry. The result is a ready-to-review email draft in the recipient’s language. Nothing is sent without human approval.
Outcome: A sales email is created in seconds instead of half an hour. Recipients receive examples that fit their business. Opt-out and approval are documented.

Craft manufacturing company · South Tyrol

Quotes in minutes instead of hours

Situation: For every customer request, sales manually searches the order history for comparable jobs, estimates price and effort, asks colleagues and assembles text blocks. Different salespeople calculate the same job differently. Margins are not enforced systematically.
Solution: We connect ERP history with an AI module that finds comparable orders and suggests a price corridor — including effort estimate and suitable text blocks. Sales sees the proposal, reviews it, adjusts it and approves it.
Outcome: Quotes are created in minutes instead of hours. Similar jobs are calculated consistently across the sales team. Margins can be managed more deliberately because the data basis is visible.

Nursery and plant production · South Tyrol

Irrigation forecasting — every plant gets what it needs

Situation: Irrigation usually runs by schedule or visual inspection. During heatwaves the reaction comes too late; when it rains, watering often continues without need. Water and energy are wasted, while individual crops still suffer stress.
Solution: We connect soil-moisture and climate sensors with weather forecasts and plant master data. A model recommends the optimal water volume and irrigation time for each bed and crop type. Control can run as a recommendation to the gardener or directly through existing valves.
Outcome: Water and energy consumption drop measurably, plant growth becomes more consistent, and documentation for organic certification or subsidy applications is created automatically. The nursery keeps decision sovereignty — AI recommends, humans approve.

Bakery and pastry · South Tyrol

Bread forecasting — the right products on the shelf every day

Situation: Bread and pastry quantities are planned by gut feeling and the baker’s experience. On weekends, holidays or when the weather changes, estimates miss the mark — sometimes returns pile up, sometimes shelves are empty in the afternoon. Both cost margin and customer trust.
Solution: We connect POS history from recent years with weather, holiday and tourism-season data. A forecasting model suggests production volumes for each branch, day and item. The baker sees the recommendation, can adjust it and approves it — the model learns from every correction.
Outcome: Returns drop noticeably, shelves remain attractive throughout the day, and staff and raw materials are aligned with actual demand. The baker’s experience remains the decisive lever — now backed by data.

Tank truck cleaning · South Tyrol

From paper form to digital process

Situation: Field employees fill out cleaning forms on paper. The data is later typed into the ERP a second time — error-prone, time-consuming, and historical jobs end up in folders.
Solution: A tablet app guides staff through the workflow directly on site. Photos of freight documents are read automatically, licence plates are recognised, and everything flows straight into the ERP system. Paper becomes one digital process.
Outcome: No more media break. Several minutes are saved per job, transfer errors disappear, and every past job is searchable from now on.

Fruit cooperative · South Tyrol

Harvest forecasting and storage optimisation for fruit cooperatives

Situation: Two cooperating cooperatives process more than 100,000 tonnes of fruit per year. Variety-changeover times cost productive hours, one site is overfilled while others are underused, and packaging material is purchased separately by each cooperative.
Solution: We combine time-series forecasting for incoming harvest volumes with an optimisation model (MILP / CP) for storage and sorting allocation. A computer-vision pilot on one sorting line adds automatic quality classification. Integration runs through existing ERP and WMS interfaces — no data copy into a third-party environment.
Outcome: Losses from variety changeovers fall, storage utilisation evens out, and bundled packaging purchases become negotiable based on data. Spoilage and inventory become visible before they turn into problems — the pilot shows possible six-figure annual savings in the first season.

Hospitality · South Tyrol

Hotel assistant — house rules, menu and region in one answer

Situation: Guest questions about house rules, breakfast times, wellness, events and the region arrive verbally at reception or in the email inbox. The same answers are rewritten every day, with varying quality depending on language and shift.
Solution: We index house policies, menus, wellness and regional content in a RAG knowledge base. The assistant answers guest questions directly in multiple languages, with sources — as a chatbot in the app or embedded on the website. Answers are curated by the hotel; every statement remains traceable.
Outcome: Reception is relieved of routine questions, and guests receive answers around the clock in their language. Content is maintained once and then played out consistently across multiple touchpoints — new house rules or menus only need to be added to the index once.

Wholesale · South Tyrol

Receivables management that sorts itself

Situation: Payment reminders are sent one by one. Nobody knows which customers chronically pay late and which are only late once. Liquidity suffers, and accounting loses time on routine emails.
Solution: We analyse historical payment data and group customers by payment behaviour. Reminders go out automatically — via WhatsApp or email, in the right language, at the right time. A human approves every escalation level.
Outcome: Accounting no longer handles standard cases manually. Early payment delays are detected sooner, liquidity improves, and difficult cases get the attention they need.

Food wholesale · EU

Warehouse logistics — AI-optimised forklift routes

Situation: In a 10,000 m² warehouse, forklifts currently drive around 1,400 km per month — often sequentially, often by gut feeling and often with empty trips. Poor picking paths, unbalanced slotting and bottlenecks at crossings tie up staff and maintenance; frozen and fresh zones add further constraints.
Solution: We combine WMS logs, forklift IoT telemetry and a layout model into route optimisation with task interleaving and genetic algorithms. Orders are bundled spatially, empty return trips are filled with put-away tasks, critical bottlenecks become visible through heatmaps and are reduced through revised slotting suggestions.
Outcome: Pilots show around 25% faster transport cycles and savings of 300–350 km per week — for one food wholesaler this meant two fewer forklifts in operation and a shift model streamlined from three shifts to two.

Renewable energy · Photovoltaics · EU

PV plant monitoring with anomaly detection

Situation: Inverter data is spread across different portals for each plant. Anomalies — power drops, defects, weak strings — are often only noticed during month-end reporting. Yield losses quietly accumulate in the background.
Solution: We build a central dashboard that consolidates inverter and sensor data. Anomaly detection flags unusual patterns per string and plant almost in real time, while forecasting models define expected yields based on weather and plant size. Alerts run via email or Teams.
Outcome: Issues are detected in hours instead of weeks. One interface replaces portal hopping, yield losses are quantified and weighed against maintenance costs — a basis for O&M-as-a-Service offerings or ISO 50001 reporting.

Print and packaging manufacturing · South Tyrol

1,500 die-cutting forms become a searchable 3D library

Situation: A historically grown collection of around 1,500 die-cutting forms exists partly as print data (PDF/vector) and partly as physical samples. Sales and production need minutes to hours to find the right form for a customer request — there is no central overview.
Solution: We parse cut and crease lines from print files, automatically generate a 3D model of each die-cutting form and store everything in a searchable asset database. Search works by dimensions, product type, material and similarity embeddings — for example “folding carton ~10×6×4 cm with tuck closure”. Links to ERP orders and calculation are included.
Outcome: Knowledge that used to be scattered across people’s heads becomes a searchable library with preview renderings. Sales can immediately compare customer ideas with existing inventory, production sees historical usage — and new die-cutting forms can be compared with similar existing forms before ordering.

Construction information services · Italy

Text recognition for resolution PDFs — less review work, better data quality

Situation: Thousands of Italian municipal resolutions per year are read by an older OCR/NER pipeline. With the existing field quality, too many incorrect master-data matches land in the database, date heuristics confuse format variants, and accents and layout breaks create manual review work.
Solution: We rebuild the pipeline on modern Document AI — LLM vision for OCR and information extraction, cross-checking through two models, and predecessor-project search through a vector database. Master data never leaves the customer infrastructure; only top-K candidates go to the LLM reranker. A confidence heatmap in the UI sends only fields below threshold to reviewers.
Outcome: Data quality on critical columns — contracting authority, resolution number, date and amount — improves significantly. False positives become rarer, manual review focuses on truly uncertain cases, and the database becomes more reliable as a calculation basis and search index.

Construction company · South Tyrol

Internal knowledge assistant across your own file server

Situation: The file server has grown over decades — several terabytes, mixed Office documents and CAD files (DWG, RVT, IFC). Employees search for minutes or hours until they find the right document. Knowledge is scattered across Outlook, SharePoint and deep folder structures.
Solution: We index the file server directly, without data copies — including CAD metadata. Active Directory permissions are passed through 1:1 on every query; the AI model runs hybrid (agent + data on-prem, model encrypted externally). Employees ask in natural language, and the assistant answers with source references.
Outcome: Nobody sees documents they would not be allowed to access on the file server. Search times drop noticeably, answers come with provenance, new sites are rolled out step by step — without copying data into an external cloud.

Book initial call ›