A Smart Approach to Artificial Intelligence: Processes First, Tools Second

This presentation, delivered at the 2025 Tax and Finance Conference (DFK) in Portorož, explores the responsible adoption of Generative AI (GenAI) across public administration, finance, and policy. It argues that true digital transformation begins with rethinking processes, not with chasing tools. AI should serve people and organizations—not the other way around.

The session offers practical illustrations of where GenAI can already deliver real value:
• automating rule-based administrative work;
• summarizing and structuring documentation;
• generating well-formatted reports or responses;
• supporting decision-making through transparent, explainable outputs.

It also shows where AI should not be used— when context or accountability matters, or when the temptation arises to produce polished but meaningless “workslop” that merely looks intelligent. Using examples from finance, law, and management, it demonstrates both the power and the limits of automation.

Participants are reminded that GenAI tools make mistakes in new and unexpected ways. Understanding those failure modes—hallucinations, biased reasoning, fabricated citations—is now an essential professional skill. Several real cases are discussed, including the Slovenian Tax Administration’s award-winning chatbot, illustrating how responsible design and testing can minimize such errors.

Beyond technology, the talk reflects on organizational learning. If decision-makers require employees or applicants to use GenAI just to navigate complex forms, perhaps the forms—not the people—need redesign. In an AI-rich world, long documents no longer prove commitment; clear thinking and effective collaboration do.

The concluding message is simple yet profound:
“Processes first, tools second.”
Generative AI is the first family of IT systems that can reason, create, and err in human-like ways. That makes it powerful—but also risky. To use it well, we must understand both its capabilities and its illusions.

By combining insights from academic research, real projects, and more than 100 workshops across sectors—from management and education to public administration—this presentation invites finance and policy professionals to harness AI responsibly, transparently, and intelligently.

1. Portorož, 9. 10. 2025 Peter Trkman [email protected] Pametno k umetni inteligenci: najprej procesi, nato orodja

2. Klasično profesorsko hvaljenje: veliko člankov, citatov, raziskovalnih projektov Čez 100 delavnic o uporabi Generativne umetne inteligence za management, raziskovanje, finance, pravo, trženje, javni sektor in upravno akademijo, izobraževalce in celo otroke ter skavte Številni javno dostopni viri in gradiva: https://www.slideshare.net/PeterTrkman Kontakt: [email protected] ali https://www.linkedin.com/in/peter-trkman/ Vir: https://www.perplexity.ai/search/i-will-be-presenting-here-http-1NGgVXORSOSgncseLoZQFQ#0 O meni 2

3. 3 “ Živimo v času nenehnih sprememb ” Eksponentna rast sposobnosti AI in uporabe orodij . Česar danes še ni , bo jutri obstajalo Marsikatero današnje znanje jutri ne bo več potrebno Potrebna sprememba sposobnosti zaposlenih in podjetij Strah nas je kaj prinaša jutri ; kakšni šoki nas čakajo The one thing that does not change is that at any and every time it appears that there have been 'great changes‘. Within a Budding Grove, Marcel Proust, 1919 10 urno predavanje v 10 sekundah

4. 4 Uvodne opredelitve : umetna inteligenca Vir: Li et al , 2021 Vir: AT Kearney , 2023

5. Vse, kar je digitalno There is an app for that ! Opomba: poziv pripravljen z https://console.anthropic.com/ Kaj GenAI že zna? A professional, policy-oriented business infographic–style illustration showing how Generative AI (GenAI) responsibly supports every stage of work in public administration, finance, and project management. The image should feel credible, institutional, and forward-looking , suitable for a keynote slide at the Davčno-finančna konferenca (DFK). Overall layout & composition: Horizontal panoramic layout divided into five interconnected segments , representing key domains of modern public and financial management: Data & Policy Analysis Strategic Planning & Budgeting Process Optimization & Compliance Digital Transformation & Transparency Reporting & Communication Each segment connects via subtle curved lines of light converging toward a central “Responsible AI Core” — a glowing sphere symbolizing transparent, explainable AI. The background is a light silver-to-white gradient with minimal geometric or mesh patterns, reflecting professionalism and clarity. Visual representation of GenAI tools: The central AI Core should look like a luminous globe made of interwoven data filaments (blue, teal, and soft gold tones) symbolizing trust, knowledge, and collaboration. Each segment integrates AI subtly: holographic dashboards, verified data overlays, or transparent assistants suggesting improvements — all controlled and audited by human experts. Segment details: Data & Policy Analysis: Analysts reviewing economic or financial data on clear holographic panels, AI summarizing trends with verifiable citations. Strategic Planning & Budgeting: A team aligning priorities with digital maps and budget projections, AI assisting in scenario simulations and funding optimization. Process Optimization & Compliance: Automated workflows improving accuracy and audit trails; AI flags anomalies or compliance risks. Digital Transformation & Transparency: Modern offices or municipal environments with integrated dashboards ensuring accountability, supported by AI governance tools. Reporting & Communication: Professionals presenting reports to decision-makers; AI converting structured data into clear, fact-checked summaries and visuals. Design style & color palette: Professional institutional tone — clean vector or semi-3D style. Primary colors: deep blue (DFK identity), silver-gray, and soft teal . Accent color: warm gold to symbolize efficiency and success. Avoid playful or futuristic neon; maintain trust, clarity, and balance. Cohesion & message clarity: All segments are connected through glowing data streams converging into the “Responsible AI Core,” symbolizing one transparent, reliable AI system supporting all activities. Include small subtle icons (data shield, document, chart, balance scales) indicating trust, evidence, and accountability. Mood & tone: Trustworthy, rational, and visionary , not hype-driven. Ideal for finance professionals, auditors, policymakers, and digital leaders . Emphasize the message: “AI serves people and processes — not the other way around.” 5

6. Nekaj online prezentacij: Konec svetovanja kot ga poznamo https://gamma.app/docs/Premagovanje-izzivov-vodenja-v-javnem-sektorju-v-Sloveniji-j6ci8kk08jkb9en https://gamma.app/docs/Qlector-Consulting-Report-Navigating-the-Future-of-Production-Man-seo169dwyq3az55 Moja pretekla konzultantska poročila Strategija za Petrol McKinsey poročilo za Kraljevino Dansko Predlogi za Qlector Pa začnimo 6

7. 7 Obstaja veliko specializiranih orodij Marsikaj vam bodo prodajali tudi različni ponudniki Povsem dobro pa deluje tudi Projekt v ChatGPT Prijava na razpis RZK No, lahko pa dokumentacijo poslušate tudi v avtu na poti iz Portoroža: Podcast z NotebookLM Obvladovanje dokumentacije

8. 8 Kje nam lahko pomaga?

9. 9 „Običajna“ avtomatizacija delovnih procesov, tudi kadar gre za zahtevne procese: npr. „izračunaj točke na podlagi jasne formule iz podatkov v Excelu in v drugih bazah, pripravi odločbe, pošlji vsakemu vlagatelju odločbo na njegov mail“. Priprava „serijskih“ dokumentov/odločb, ko so vsi podatki že pripravljeni. Aktivnosti, ki se jih da enostavno opisati s poslovnimi pravili, npr. Odobri kredit do 3000€ vsem, ki so zaposleni za nedoločen čas. Izplačaj denar na bankomatu v višini max (stanje na računu + 300 €;700€), če je vnesen PIN=PIN v bazi. Katerekoli „rule- based “ aktivnost, npr. „Za ta seznam krajev povej, v kateri občini so“ Kje ne rabite (Gen)AI?

10. Idealen za workslop AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task. Q uickly produce polished output: well-formatted slides, long, structured reports, seemingly articulate summaries by non-experts, and usable code.  Vir: Niederhoffer , K., Kellerman , G. R., Lee, A., Liebscher , A., Rapuano , K., & Hancock , J. T. (2025, September 9). AI- generated ‘ workslop ’ is destroying productivity . Harvard Business Review . Kaj tudi zna? 10

11. 11 Kot odločevalci : Če je vaš tekst / sporočilo / slika vsebinsko neključen : razmislite , ali se ga da ukiniti ( neko polje v obrazcu ) ali zamenjati s klikom ( npr . potrdilo udeležbe na sestanku , zahvala za odgovor ) . Če morajo prijavitelji , zaposleni , stranke uporabljati GenAI, da izpolnjujejo vaše obrazce , a bi lahko spremenili obrazce ? Zakaj še vedno zapravljamo mesece s pisanjem 60-stranskih projektnih predlogov , nato pa pol leta čakamo na odločitev ? V dobi GenAI dolga besedila („ljubezenska pisma“) ne dokazujejo več zavezanosti ali kakovosti . Petrov predlog za spremembo razpisov Zato ključno sporočilo za vas

12. 12 GenAI orodja so prva orodja v zgodovini razvoja informacijske tehnologije, ki se motijo na izjemno nenavadne načine. In tudi zato morate znati še več, da to prepoznate Kdaj pa ne uporabljati: „ Ko ne poznate načinov , kako UI lahko odpove “

13. 13 Halucinacija: Bolj kot ga mučite, bolj odpove Vir: Perplexity Lahko preberete celoten tekst o Petru. Vse gradi na njegovih člankih, vendar je interpretacija izmišljena Nadaljnji primeri: https://www.linkedin.com/posts/peter-trkman_yes-genai-tools-are-great-they-can-win-activity-7352991067366948864-LJiS

14. Halucinacije Tako kot študenti , ki se soočajo z zahtevnimi izpitnimi vprašanji , tudi veliki jezikovni modeli včasih ugibajo , kadar so negotovi , in namesto da bi priznali negotovost , podajo verjetne , a napačne izjave … .. jezikovni modeli halucinirajo , ker postopki usposabljanja in ocenjevanja nagrajujejo ugibanje namesto priznavanja negotovosti Vir: Kalai , Adam Tauman , et al . " Why Language Models Hallucinate ."  arXiv preprint arXiv:2509.04664  (2025). 14

15. 15 Nagrajeni chatbot v Sloveniji: Furs Naj izboljšava v javni upravi Premik naprej 2025 Dejansko odličen, praktično nikoli se ne zmoti … …razen kadar se Primer dobre prakse

16. Kaj to pomeni za vas pri davkih/financah? 16

17. 17 Machines will be capable, within twenty years, of doing any work a man can do. Herbert Simon,1965 (Simon je 1978 dobil Nobelovo nagrado za ekonomijo . V letu 2001 je umrl ) Vendar pa ne pozabimo – marsikaj se spreminja GenAI lahko razbije tradicionalno poslovno šolo na koščke – loči vse, kar jo je prej držala skupaj: vsebino, skupnost in diplomo. Če bodo šole ostale pri klasičnih predavanjih, jim bodo tehnološka podjetja pobrala tako študente kot intelektualni prestiž. Morajo korenito prenoviti programe, digitalizirati učenje in raziskovanje ter se povezati z gospodarstvom. Sicer jim grozi, da bodo postale zgolj drage izdajateljice diplom brez prave vsebine in vpliva. Vir: Ives , Jarvenpaa , 1996 (opomba: besedo „internet“ v originalnem članku sem zamenjal z GenAI )

18. 18 Vendar pa ne pozabimo – še več stvari pa ostaja enakih

19. 19 Varnost, zasebnost, pravni vidiki: za kdaj drugič https://gamma.app/docs/Varnost-pri-uporabi-orodij-generativne-umetne-inteligence-b55fllmfe7lpoe7 Vendar pa ne pozabimo

20. 20 Skupaj zmoremo več Za konec

21. 21 Claude Canva ChatGPT Plus Google AI Studio Google NotebookLM Perplexity Gamma Leonardo ZKP Samorazkritje : v prezentaciji uporabljena orodja