2025 Retrospective: 6 Automation Projects with Actionable Checklists

The year 2025 marked a transition in custom software development projects and OPTI's qualification discussions. The main theme was resolving critical operational bottlenecks. While in past years many potential clients were interested in "What can we theoretically do with technology X?" or "How do we also implement others' solution Y?", in 2025 they sought consultancy and development of the type "How do I solve this inefficiency I have found internally?".

Here are 6 real project typologies, anonymized and grouped into three themes. For AI enthusiasts, we link the 6 OPTI guides for hybrid architecture for implementing artificial intelligence in business-to-business workflows. We are also adding short necessary checklists for successful implementation to each project.

Theme 1. Unifying Fragmented Data and "Single Source of Truth"

Discussions in 2025 showed immense frustration regarding data fragmentation. Companies no longer want just a contact book (e.g., CRM like HubSpot), but a marketing, invoicing, and delivery data engine united in a single platform.

(1) Logistics and "Door-to-Door" Transport Quotes

A player in the international logistics field needed a system to unify transport prices (air, naval, road), automatically calculate margins, and generate complex offers with one click from HubSpot CRM.

Excel could no longer handle the complexity of calculating routes, prices, and profitability, and scaling the transport business to dozens of countries is impossible without a unique data coordination system.

✅ Quick checklist for implementation:

• Standardizing data: Define price variables clearly (km, weight, container type) before writing code or programming low-code tools (e.g., HubSpot Workflows).
• Integrations: Ensure the system is connected to external suppliers for updated rates, via API or data import.
• Profitability gate: Implement a rule that disallows sending the offer if the margin is under X%.
• Continuous optimization: Integrating artificial intelligence allows for easier predictive determination of demand and profitability.

➤ Discover Guide 5: Continuous Optimization & Reporting with AI

Theme 2. Automating Manual Input

Another leitmotif of 2025 was the elimination of human errors from repetitive and complex processes, with the additional goal of enabling business scaling.

(2) Sports Community and Automated Rankings

A sports organization with thousands of children registered in dozens of localities managed rankings, scores, and match history manually. The challenge was creating a system that takes results, automatically calculates scores (including attendance bonuses), and generates hierarchies in real-time, eliminating suspicions of fraud or error. Additionally, it offers personalized recommendations based on community, game history, and the possibility to purchase game accessories.

✅ Quick checklist for implementation:

• User roles: Strictly define who can input data (local organizer) and who can only view it (parent/athlete).
• Scalability: The database must support thousands of simultaneous queries on tournament days.
• Audit trail: Keep a history of every score modification for total transparency.
• Smart recommendations: Integrate AI to increase upsell possibilities.

➤ Discover Guide 1: Personalized Recommendations, Upsell, and Sales Rules with AI

(3) Complex Bidding in Construction Work: From Word Tables to Product Catalog

An entrepreneur with thousands of very complex historical offers in Excel, each worked on for hours or even days, wanted to transform this "data graveyard" into a resource used for populating new offers. Since projects are custom, every offer had been the result of consulting dozens of external catalogs and analyzing alternatives.

The solution: a system that standardizes chaotic supplier names into a unique nomenclature referencing the initial materials. Thus, a new offer is made by "checking" boxes, not writing from scratch, drastically reducing bidding time. Furthermore, it prepares the ground for automatic generation of offer drafts via AI.

✅ Quick checklist for implementation:

• Data cleaning: Clean historical duplicates (e.g., "cement bag" vs. "bag cement") before import.
• Master data: Create a "parent product" structure to unite variations from suppliers.
• Versioning: Ensure old prices remain saved in historical offers but do not appear in new ones.
• AI Integration: GenAI can create drafts of new offers based on existing models.

➤ Discover Guide 4: Hybrid Search with AI: Applications & Chatbots for Legacy Software

(4) Professional Training: Generating Navigation Scenarios

Perhaps the most complex idea of business logic: simulator and training (e.g., LOFT - Line Oriented Flight Training), compliant with European and global regulations. The system must combine weather variables, routes, types of failures and physical layouts, ensuring the scenario is realistic, compliant, and useful for high-quality instruction. Care for personal and confidential data used is mandatory, as is software output accuracy. Final decision and validation remain the responsibility of the instructors.

✅ Quick checklist for implementation:

• Constraints engine: Define "hard" rules (what is not allowed by law) vs. "soft" rules (what is meteorologically improbable) in a logic engine, not in AI.
• UX for experts: The interface must be fast (responsive), allowing specialists to modify parameters on the fly.
• Standardized output: Automatic generation of the official training scenario in PDF to be archived for compliance.
• AI Integration: Within the limits of the constraints engine, GenAI can be extremely creative with training scenario variants.

➤ Discover Guide 6: Security, Audit, and Standards

Theme 3. AI Beyond the Hype: Measuring Return on Investment (ROI)

In 2025, clients didn't ask for "something with AI", but rather AI tools that read, understand, and respond correctly.

(5) Financial Sentiment Analysis: From Days and Hours to a Few Minutes

A team of financial analysts requested a system to daily "read" dozens to hundreds of PDF reports and "watch" stock market analysis video clips, to automatically extract a sentiment score ("Bullish" vs. "Bearish") and key points. Automation reduced manual research time from hours to minutes. Read a case study on the same topic.

✅ Quick checklist for implementation:

• Multimodal ingestion: Ensure the pipeline can process text (PDF) and audio/video for transcription.
• Prompt engineering: Define strict criteria for what "positive" or "negative" means in a financial context. One approach is based on examples per grade tranche (0-10, 10-20, etc.) or based on adding/subtracting points starting from the average (50).
• Traceability: The AI must provide a link to the source (page/minute) from where it extracted the conclusion.

➤ Discover Guide 3: Multimodal Input: Photo, Image, Audio, and WhatsApp

(6) Green Energy Assistant with Zero Hallucinations

A project for the international market, requiring a virtual assistant capable of answering questions about green energy legislation. The critical challenge: the AI is not allowed to invent laws or regulations. The solution is a RAG (Retrieval-Augmented Generation) architecture that answers strictly based on uploaded official documents (laws, regulations, interpretation texts) and offers very high factual accuracy (in industry studies, well-configured RAG architectures can reach approx. 94% accuracy).

✅ Quick checklist for implementation:

• Manually verified Knowledge Base: Feed the system only with official and carefully verified documents.
• Grounding: Configure the model to answer "I don't know" if the information does not exist in the database.
• Continuous update: Implement an automatic re-indexing flow when laws or regulations change.

➤ Discover Guide 2: Integrating Exact Data with AI (RAG, Price, Stock)

In Place of a Conclusion

The diversity of these projects shows that organizations are looking for permanent solutions to optimize operations. If 2025 was the year companies defined what they want (clean data, specific automation, controlled AI), we hope 2026 will be the year of scaling custom developed solutions, so that the investments made bear fruit. Happy New Year!

➤ Don't leave scaling for 2027. Schedule a discovery session with us