5 Things to Consider When Deploying AI in Your Enterprise
0 likes133 views
AI is reshaping how organizations approach data, integration, and automation. But as powerful as it is, AI success in your workflows doesn’t come from simply plugging in a model and hitting “go”. Critical steps can shape your success, from choosing the right models to handling multi-modal data and scaling your approach. Join us as we break down five key considerations that will help you make smart, scalable decisions when bringing AI into your enterprise. In this webinar, we’ll explore: -The role of multimodal AI in integrating various data types like spatial, text, and imagery. -The difference between applying AI for data integration vs. chat-style interactions. -How to choose the right AI model for each task within your FME workflows. -.AI for cloud vs. on-premise deployment, including security, cost, and data sovereignty considerations -When batch processing beats manual prompting, and how to reuse prompts to scale effectively. Plus, we’ll challenge you to think beyond traditional limitations. AI can enable workflows and insights that were once impossible. Whether you're experimenting with generative tools or automating complex processes, this session will help you make confident choices to maximize your use of AI and get more from your data.
![Input
Task: Watch the attached video (total length is 70
seconds) very carefully and Identify the frames in
where traffic lights appear. When you see a traffic
light, determine its color (red or green) and the time
in seconds from the beginning of the video.
Recalculate minutes and seconds to seconds only.
Requirements: Output timestamps and light colors.
Each entry should have: "timestamp": The time in
seconds when a traffic light is detected. "light": The
color of the traffic light ("red" or "green"). Do not
guess times—only return actual detections. Ensure
all numbers are correct and consistent. Output
format (JSON) example: [ {"timestamp": 45,
"light": "red"}, {"timestamp": 78, "light": "green"},
{"timestamp": 101, "light": "red"} ]
Structured Output Visualization
Blogs about video in FME:
FME does computer vision
Spatially Enabled Video Editing with FME](https://image.slidesharecdn.com/jun25-5thingstoconsiderwhendeployingaiinyourenterprise-250625172613-34d60148/85/5-Things-to-Consider-When-Deploying-AI-in-Your-Enterprise-25-320.jpg)
5 Things to Consider When Deploying AI in Your Enterprise
- 2. Meet the Presenters Don Murray CEO Dmitri Bagh Scenario Creation Analyst, AR Product Strategist Chris Berger Customer Solutions Team Lead
- 3. Welcome to Livestorm. A few ways to engage with us during the webinar: Audio issues? Click this for 4 simple troubleshooting steps.
- 4. How to download slides 1. Hover over the slide deck in the webinar room 2. Click this button
- 5. Agenda 1 8 Resources & Next StepsIntroduction 2 Choosing the right AI model for each task 3 The role of multimodal AI in integrating various data types 4 AI for data integration vs. chat-style interactions 5 AI for cloud vs. on-premise deployment Security, cost, and data sovereignty considerations 6 Batch processing vs. manual prompts & how to re-use prompts to scale 7 Conclusion 9 Q&A Agenda
- 7. Confidently implement AI in your enterprise with today’s key considerations and shape scalable, successful deployments with FME.
- 8. ● Choose right AI models for each task ● Integrate AI with all data types ● Deploy AI securely, cost-effectively (cloud vs. on-prem) ● Scale single prompts to automated workflows ● Go beyond traditional limitations, multimodal AI AI is powerful… when applied wisely! In this webinar, you'll learn how to:
- 9. Common Frustrations With AI Deployment ● “I don’t know where to start with AI in my workflows.” ● “We’re overwhelmed by too many tools and models.” ● “Our data is too complex or too varied.” ● “Manual prompting isn’t scalable for real business use.” ● “We’re concerned about data privacy or cloud limitations.”
- 10. There is a better way with FME! FME makes AI practical by helping you integrate the right models with the right data: securely, scalably, and without code.
- 11. The only All-Data, Any-AI Platform. FME Form FME Flow All Data workflows are built here. Brings life to FME Form workflows FME Flow Hosted SaaS version of FME Flow fme.safe.com/platform FME Enterprise Integration Platform Safe & FME FME Realize Experience your data where it matters most. In context
- 12. With 500+ supported data types in FME. Unrivalled Data Support GIS CAD Database XML Raster 3D BIM Web Point Cloud Cloud Big Data IOT Graph BI Indoor Mapping AR/VR Generative AI Cloud Native Tabular
- 13. Consideration #1 The role of multimodal AI
- 14. Poll: Which is your top priority when choosing how to deploy AI? (select top 1-2)
- 15. Richer Automation ● Pre/post processing requirements ● Multi-Agent driven workflows ● Enhanced reporting/notifications Expands Use Cases ● Extract more value ● Unlock new insights Power of Multimodal Platforms Key Considerations
- 16. OpenAI Capabilities ● Web Search, File Search, Vision Google Gemini Capabilities ● Document, Vision, Audio, Video Understanding Similar capabilities with Ollama, Bedrock, etc. Power of Multimodal Platforms Multimodal Agentic APIs
- 17. Path to File ● Upload or URL Tooling ● Google Search Grounding/Retrieval ● URL Context Advanced ● Max Retries, & Timeout Parameters GoogleGeminiConnector Key Transformer Updates
- 18. Action ● Chat & Text Completions ● Vision ○ File input (supported models) Advanced ● Max Retries, & Timeout Parameters OllamaConnector Key Transformer Updates
- 19. Built on the Responses API ● Responses vs Chat Action & File Inputs ● Text, image, file, web search, reasoning Advanced ● Max Retries, & Timeout Parameters OpenAIConnector (New) Key Transformer Updates
- 20. Getting Started with AI in FME: ● Classifying Unstructured PDF Files ● Extracting Insights from Unstructured Documents Coming soon… ● Prompt Engineering ● Web Searching ● Embeddings ● FME and AI Course (targeting Fall 25) New Learning Resources Where to Start
- 21. Poll: What FME and AI use cases or concepts are you most curious about?
- 22. Slide Title What is this demo going to achieve? Goal Block Key Integrating with Multimodal Models Result What’s in the way? Why don’t we have the goal already? How to overcome roadblock to get goal? Goal was achieved. How is life better now?
- 23. Slide Title Video Analysis Scenarios ● Description ● Sound-to-text extraction ● Object detection with timestamps ● Answering questions about videos Audio Analysis Scenarios ● Transcription of sound recording ● Sound classification ● Emotion or tone detection in voice Possible (and tested!) multimodal scenarios Image & Document Analysis Scenarios ● Object detection and localization ● PDF table extraction to HTML ● Physical size estimation from images ● Automatic descriptions, tags, and titles ● Answering questions about visuals
- 24. Demo
- 25. Input Task: Watch the attached video (total length is 70 seconds) very carefully and Identify the frames in where traffic lights appear. When you see a traffic light, determine its color (red or green) and the time in seconds from the beginning of the video. Recalculate minutes and seconds to seconds only. Requirements: Output timestamps and light colors. Each entry should have: "timestamp": The time in seconds when a traffic light is detected. "light": The color of the traffic light ("red" or "green"). Do not guess times—only return actual detections. Ensure all numbers are correct and consistent. Output format (JSON) example: [ {"timestamp": 45, "light": "red"}, {"timestamp": 78, "light": "green"}, {"timestamp": 101, "light": "red"} ] Structured Output Visualization Blogs about video in FME: FME does computer vision Spatially Enabled Video Editing with FME
- 26. Demo
- 27. Input Structured Output Visualization
- 28. More Examples
- 29. Input Output PDF to HTML Table Extraction
- 30. Photo catalog
- 31. ● Rich Automation ● Enhances Use Cases ● Can be leveraged to supply data to fine tuned models ○ Continuous or batch based training Multimodal Automations Going Beyond Text/Chat Completions
- 32. Consideration #2 Data integration vs. chat-style interactions
- 33. Feature / Concern Chat Completions Text Generation in Data Integration Purpose Assist users interactively Automate and scale transformations or generation User Human System or automation tool Input format Dialog history and natural questions Structured data, fields, and prompt templates Output format Human-readable text Machine-readable formats (JSON, CSV, strings) Context handling Multi-turn conversation One-shot or batch with clearly scoped prompts Traceability Weak (convo logs) Strong (prompt stored, versioned, auditable) Repeatability Low; each session is unique High; embedded into repeatable data flows Use case maturity Early stage in enterprise AI workflows Increasingly mature in ETL/ELT and orchestration
- 34. Examples
- 35. Chat Completions FME Workbench AI Assist (2025.1+) ● Authoring tool to get help the guide users in the authoring process ● What Transformer to use next ● Add annotations and describes changes as the workspace is edited ● *OpenAIChatGPTConnector Real World Examples
- 36. AI Assist Use AI to accelerate workflow authoring. Assists with creating: ● Python ○ Actions: Generate, Refine, Explain, Add Comments ● SQL ○ Describe the requirement in plain English. Optional: Send database schema to AI. ● Regular Expressions ○ Describe a regular expression in plain English Real World Examples
- 37. AI Assist FME 25.1+
- 38. Chat Completions in FME OpenAIChatGPTConnector
- 39. Text Generation AI “Connectors” used in a workspace to access models ● Generate lookup tables to perform schema mapping process ● Expand acronyms (i.e. st = street) ● Standardize Inputs (columns, values, etc.) ● Fuzzy string matching, etc. Real World Examples
- 40. Chaining AI Services Multi-step text generation using multiple models: ● General → specialized (reasoning, etc.) ● Lower cost → higher cost ● On Prem → cloud Orchestration Considerations ● Parallel processing ● Processing location (Remote Engine Services) Text Generation AI “Connectors” used in a workspace to access models ● Generate lookup tables to perform schema mapping process ● Expand acronyms (i.e. st = street) ● Standardize Inputs (columns, values, etc.) ● Fuzzy string matching Real World Examples
- 41. Chaining AI Services 1 Routing Data Between Workspaces
- 42. Chaining AI Services 2 Looping Automations
- 43. Custom transformers with ChatGPT Blender integrations ● USDZReader ● glTFReader ● 3DModelRenderer ● 3DModelVideoRenderer
- 44. Custom transformers with ChatGPT Open3D integrations. Potential transformers ● PointCloudNormalEstimator ● PointCloudClusterer ● PointCloudNoiseRemover
- 45. Purpose & User ● Human or System Output & Structure ● Structured outputs or flexibility Execution Style ● Conversational or repeatable AI Powered Integration
- 46. Consideration #3 Choosing the right AI model
- 47. ● HuggingFace.io has millions of models to choose from ● Azure, Amazon, Google, Ollama, OpenAI all have Model Marketplaces ● Filter by Task, Operation, cost Best of Breed per Task Choosing the right AI model
- 48. Choosing the right AI model
- 49. All-Data, Any-AI Integration: innovations with FME and Google Webinar
- 51. AI model accuracy / cost getomni.ai/ocr-benchmark Gemini 1.5 Flash ● 100,000 record cards ~ £7 Gemini 2.0 Flash ● 100,000 record cards ~ £19 Accuracy and affordability - at scale
- 52. ● Cost & Performance ○ Model Strengths, weaknesses, accuracy ● Customizability (fine tuning, etc.) ● Task-specific selection ○ LLM or SLM ● Deployment, security, sovereignty discussed next Key Considerations Choosing the right AI model
- 53. Consideration #4 Cloud vs. on-premise deployment Plus security, cost, data sovereignty
- 54. Poll: Which AI service(s) are you currently using or exploring in your workflows?
- 55. Cloud gives you access ● Fast access to cutting-edge models ● Lower setup cost ● Easier scaling (bursty requests) ● Data residency ○ ISO/IEC 27001, GDPR, SOC 2, or local data laws Cloud vs On Prem Cloud vs. On-Prem Deployments
- 56. On Prem gives you control ● Higher control (governance) ● Better for regulated industries ● Avoids data egress issues ○ Lower latency ○ Ideal for predictable, long running tasks ● Potentially higher start up costs Cloud vs On Prem Cloud vs. On-Prem Deployments
- 57. Hybrid gives you both ● Cloud for general request, scalability, & cutting edge models ● On prem for control over sensitive data or private cloud environments. ● Hybrid setups help meet sovereignty laws ○ Remote Engine Services Cloud vs On Prem Cloud vs. On-Prem Deployments
- 58. Cloud ● Google Gemini, AWS Bedrock, Azure Foundry, OpenAI , Anthropic, Mistral, etc. On Prem ● Ollama, Docker, Langchain, etc. Cloud vs. On-Prem Deployments Supported AI Services in FME Form Flow
- 59. ● Cloud agility for innovation ● On-prem control for governance ● Interoperability and orchestration for resilience ● Cloud to Cloud is also an option Hybrid isn’t a compromise, it’s a strategy that adapts AI to your enterprise Key Considerations Cloud vs. On-Prem Deployments
- 60. Consideration #5 Batch vs. manual prompting & reuse prompts to scale
- 62. ● Strings ● Attributes ● Parameters ○ User ○ FME ○ Deployment Concatenating Prompts Batch vs. manual prompting & reusable prompts
- 63. Key for batch processing and standardizing outputs. ● Highly reusable prompts ● Easier downstream consumption and validation Structured Outputs Batch vs. manual prompting & reusable prompts
- 64. Can be used to store prompts, JSON structures (Structured Outputs), API Keys, and more for easy use throughout your organization Deployment Parameter Store Batch vs. manual prompting & reusable prompts
- 65. 7 Conclusion
- 66. Summary ● Choose fit-for-purpose models per task ○ No one-size-fits-all AI ● Understand difference: AI for integration vs. chat-style tasks. ● Multimodal AI to work with spatial, text, imagery, and more in one workflow. ● Plan your deployment strategy, cloud vs. on-premise impacts cost, security. ● Scale with automation: batch prompts + reusable workflows enable enterprise-wide impact.
- 67. 30+ 30K+ 128 140+ 25K+ years of solving data challenges FME Community members countries with FME customers organizations worldwide global partners with FME services 200K+ users worldwide 200K+ users worldwide
- 68. All Data. Any AI. All Data Velocities Batch (ETL, Reverse ETL, ...) Event ( BPA, RPA, ...) Stream All Data Locations Any Cloud On-premises Hybrid Edge Containers Embedded Mixed All Data Types Unstructured Structured Spatial APIs Web Apps … Any AI Technology OpenAI Amazon Bedrock Google Gemini Ollama Deepseek Composite
- 69. 8 Resources
- 70. Past Webinars ● All-Data, Any-AI Integration: FME & Amazon Bedrock in the Real-World ● AI Agents Made Simple: Unleash the Power of All Your Data with Any AI ● All-Data Any-AI Integration: Innovations with FME and Google
- 71. Get our Ebook Spatial Data for the Enterprise fme.ly/gzc Guided learning experiences at your fingertips academy.safe.com FME Academy Resources Check out how-to’s & demos in the knowledge base support.safe.com Knowledge Base Webinars Upcoming & on-demand webinars safe.com/webinars
- 72. 9 Next Steps
- 73. We’d love to help you get started. Get in touch with us at [email protected] Experience the FME Accelerator Contact Us A world where data is not just a commodity but a catalyst for real change. fme.safe.com/accelerator Next Steps
- 74. ClaimYour Community Badge & Dive into the new Community! ● Get community badges for watching webinars ● community.safe.com ● Today’s code: 481U4M Join the Community today! Next Steps
- 75. 10 Q&A
- 76. ThankYou Recap of Next Steps 1 Follow us on LinkedIn! 2 Contact us 3 Experience the FME Accelerator Please fill out our webinar survey