Practical Foundations for Organization-Wide AI Adoption

Practical Foundations for Organization-Wide AI Adoption

• Practical Foundations for Organization-Wide AI Adoption
• Ivan Lourenço Gomes
• Daweb Schools

NoteNotes

• Audience: The talk is aimed at business owners, managers, team leaders, and employees who want to push AI adoption forward in their workplace.

• Central argument: Buying individual AI subscriptions is not enough. Real organizational value comes from shared processes, reusable knowledge, testing, and team-wide adoption.

• Current problem in AI adoption:

  • Many companies report using AI, but most remain stuck in experimentation or pilot projects.
  • A common reason is that businesses have not redesigned workflows around AI.
  • In many cases, companies have not even implemented the basic foundations needed for useful AI adoption.

• Three layers of AI adoption:

  • AI starter pack: Basic shared AI workflows that every organization should implement first.
  • No-code agents: More advanced automated workflows using tools like Zapier, Make, n8n, Salesforce, or HubSpot.
  • Custom AI tools: Internal applications built with developer support to solve business-specific problems.

• Layer 1: The AI starter pack

  • The starter pack consists of:
    • A shared knowledge base to ground AI responses in accurate company information.
    • Reusable instructions so teams can standardize prompts, tone, formats, and procedures.
    • Systematic testing to identify and fix weak or incorrect AI outputs.
  • The speaker emphasizes that AI work should be collaborative, not isolated to individual employees.

• Hotel customer-service example:

  • A hotel customer-service team can use AI to answer guest emails faster.
  • Without a shared knowledge base, employees must manually add details such as parking prices, pool availability, or booking policies every time.
  • A shared knowledge base can include documents on booking policies, hotel information, room types, facilities, and location-specific details.
  • A reusable AI assistant, such as a Google Gemini Gem or equivalent ChatGPT tool, can combine those documents with predefined instructions for tone and structure.
  • The result is faster, more accurate, and more consistent customer communication.

• Importance of testing:

  • Teams should test AI assistants with real or simulated cases before deployment.
  • Errors should be logged in a simple worksheet.
  • Some failures indicate missing knowledge-base content; others require better instructions.
  • The process should be iterative: build the knowledge base, write instructions, test, refine, and repeat.

• Other starter-pack tools:

  • Meeting transcription and summaries, such as MeetGeek.
  • AI features in Google Workspace.
  • GitHub Copilot for developers.
  • Gamma for presentations.
  • NotebookLM for policy, handbook, or training-material use cases.
  • These tools are useful only when combined with shared knowledge, instructions, and testing.

• Layer 2: No-code AI agents

  • No-code agents are a next step once teams already use basic AI workflows effectively.

  • The distinction between Gems/custom assistants and agents is:

    • A Gem is reactive: it waits for the user to ask.
    • An agent is proactive: it can monitor triggers, watch folders, react to schedules, and initiate workflows.
    • A Gem mainly produces outputs in chat.
    • An agent can orchestrate actions across tools, such as creating documents, scheduling meetings, or triggering other agents.

• Consulting workflow example:

  • A new client schedules a meeting through a form.
  • The consultant researches the client, prepares a meeting brief, conducts the call, summarizes notes, and prepares next steps.
  • The speaker argues that the human conversation should remain human because it builds trust.
  • Other tasks—research, briefing, summarization, proposal preparation, and scheduling—can be delegated to AI agents.

• Customer Intelligence Team example:

  • A research agent investigates potential customers.
  • A briefing agent prepares meeting briefs from the research.
  • A follow-up agent summarizes meetings, drafts proposals, schedules calls, and defines next steps.
  • The speaker recommends splitting agents into small, specialized tasks rather than building one large agent for everything.

• Layer 3: Custom internal AI tools

  • Custom tools require more IT effort but can deliver high business value.
  • The speaker recommends building a basic internal web app infrastructure using Firebase, authentication, user accounts, permissions, and APIs.
  • Once the foundation exists, teams can quickly add AI-powered tools for document processing, translation, summarization, image recognition, audio/video processing, and data workflows.

• Custom-tool examples from a client project:

  • A multilingual content tool for a website operating in 24 languages.

    • It used the DeepL API and glossaries to improve consistency.
    • It reduced expensive and manually intensive translation workflows.

  • An invoice-processing tool that extracts product codes, quantities, and prices, then compares them with purchase orders.

    • It saved around 10 hours of work per week.

  • A knowledge-base CMS feeding a website chatbot.

    • It reduced load on human specialists by answering common customer questions.

  • An image archive that classified more than 10,000 product images using Gemini.

    • It allowed the team to search images using predefined taxonomies.

  • Google Chat bots for internal workflows.

• Main lesson from the custom-tool section:

  • Success came not from the technology alone, but from interviewing the people doing the work and building tools around their real pain points.
  • AI adoption should solve specific operational problems, not exist as “AI for the sake of AI.”

• AI effort-benefit curve:

  • AI can eventually make individuals and teams dramatically more productive.

  • However, the easy productivity gains come only after significant upfront effort:

    • interviewing teams,
    • gathering data,
    • creating knowledge bases,
    • writing instructions,
    • testing,
    • and building reusable infrastructure.

  • Many organizations get stuck before reaching the high-benefit stage.

• Final takeaway:

  • Start with simple, shared, testable AI workflows.
  • Move to no-code agents only after the basics are working.
  • Invest in custom internal tools when the organization has clear, repeated, high-value problems.
  • Sustainable AI adoption requires intention, discipline, iteration, and team-wide participation.