Exploring Agentic AI with ROI for High Impact Industries (By: Shaheer Khawaja & Muhammad Hassan Raza) - Google I/O Extended 2025

Talk by Shaheer Khawaja (https://www.linkedin.com/in/shaheerkhawaja/) & Muhammad Hassan Raza (https://www.linkedin.com/in/hassanraza22/) at Google I/O Extended 2025 by GDG Lahore.

Transcript

1. Entropy & Co. Exploring Agentic AI with ROI for high impact industries .
2. Meet Your Speakers Shaheer Khawaja CEO | Co-Founder 5+ Years of Experience In Sales, Marketing, AI, and Business Consultancy. Hassan Raza CPO| Co-Founder 3+ Years hands on Experience in full stack software development and agentic AI.
3. Our Agenda 1. Knowing the problem you’re solving. 2. Prompting Techniques and Context Engineering. 3. Best Practices 4. Hand’s on with Vertex AI 5. Take Home Exercise
4. Knowing the Problem Patient Why the right problem matters Receptionist Nurse Doctor Add. Doctor
5. Creating Applications For Real Problems. 1. Win more customers: Automations can be created for multiple channels such as SEO, SMM, E-mail marketing, cold calling, e.t.c. 2. Keep more customers: Automations can be created for inbound activities such as customer support, chatbots e.t.c. 3. Transform the way you work: Optimize workflows within the organization such as HR processes, Project Management e.t.c & More
6. Brand Promise Product/Service For Whom? Value Proposition Where? How? Proposals Competition Closing Advocacy Lead Generation Engagement Marketing Hand off Escalation & Nurturing Opportunity Hand-over Acceptance of Sale Negotiation Advancement Closing Advocacy Buyer Focused Ideation. Seller Focused Sales Marketing
7. Brand Promise Product/ Service For Whom? Value Propositi on Where? How? Proposal s Competitio n Closing Advocac y What it looks like: Sales Marketing Seller Focused Funnel With A.I Driven Marketing. Branding • A.I Driven Competitor Research. • A.I Driven feedback and test scores. • Processes for data driven creativity. Design Implementation • A.I Driven A/B Testing. • A.I Driven feedback and test scores. • Capturing customer feedback and engagement. • Making relevant changes. • Introducing A.I Agents and training on available data. • Highlighting USP’s and building brand awareness amongst target audience. A.I Analysis A.I Driven Creatives A.I Escalation A.I Research A.I Outreach A.I Powered Closing
8. What it looks like: Seller Focused “Social Media Funnel” With A.I Driven Marketing for Company X. A.I Powered Research: Conducting research on market sizes, product penetration strategies, ideal customer personas and early competitor analysis. Insights: Research has concluded that the ideal customer is aged between 35-40, dominantly male, prefers to work with brands with strong presence , shows a strong affiliation to “creative & innovative” tone of voice and prefers the visual color blue. Visual Identity Creation and Implementation with A.I: Insights: Audiences have shown engagement with designs with human elements, minimalist designs and prompts for engagement. Winning Call to action includes: “ Comment your biggest hurdle to get a free consultation. ” A.I Agent Adam: “Adam” responds to comments with a unique message, understand the sentiment of each comment and escalates relevant buyers to book a free consultation. Insights: Adam’s timely response and follow ups have increased organic reach, retained following and set appointments with MQLs. Adam has also improved it’s data set to identify and escalate only the most relevant comments and messages. Actions: • Adam reads through all the comments and messages received. • Adam understands the context of each message, looks for keywords and filters each message based on priority and customer criteria. • Adam messages an offer, relevant details to the customers requests and prompts for a call. A.I Agent “Jake”: “Jake” takes a follow up call to understand customer depth in needs, creates transcription and sets a follow-up call. Insights: Jakes’s timely response and follow ups have increased customer interest. Jake’s transcription’s have created and matured a new lead, shared relevant transcriptions and set a call with a “human team member”. A.I Agent “John”: “John” takes all data shared by “Jake” and creates a proposal. Insights: John’s proposal details every need the customer had expressed, highlights areas where input is needed and shares it with the ”closer” saving time. ”Closer” takes the call and closes the deal with all the information gathered by the customer.
9. Some of our Numbers By Solving the right problem Providing Results such as: 2000+ Calls Made in Under 4 hours Human-Like Conversation Capabilities Real Time Lead Research Capabilities 99.9% Accuracy When Navigating Complex IVR’s At a Fraction of the Cost: 10X Cheaper Than Enterprise Development Solutions 5X Volume Compared to top performing agent. Fully Compliant With SOC type 1, type 2, HIPPA compliances and tcpa. Our SDR
10. Prompting Techniques and When to Use Them Chain of Thought (CoT) Prompting Problem: [Complex question] Let's think step by step: 1. First, we need to consider... 2. Then we calculate... 3. Finally, we conclude... Where it works best? Arithmetic GPT 3.5 and Earlier Models Chain of Density (CoD) Prompting Generates progressively denser summaries without increasing length, starting entity-sparse and becoming increasingly entity-dense Where it works best? Summarization’s and Analysis Tree of Thought (ToT) Prompting Creates tree structure where nodes represent partial solutions (thoughts) Explores multiple branches simultaneously rather than single linear path Uses evaluation prompts to assess progress toward solutions Where it works best? Creating Writing Thought Generation Evaluations Table of Thought (Tabular CoT) Prompting Structures reasoning process in tabular format for improved clarity and parallel processing of multiple factors Where it works best? Decision Making Tasks Evaluations Multi Variable Prompting
11. Examples: CoT Let's think step by step: 1. I started with 10 apples from the market 2. I gave away 2 apples to the neighbor: 10 - 2 = 8 apples 3. I gave away 2 apples to the repairman: 8 - 2 = 6 apples 4. I bought 5 more apples: 6 + 5 = 11 apples 5 . I ate 1 apple: 11 - 1 = 10 apples Therefore, I have 10 apples left.
12. Examples: CoD Article: [Your article content here] You will generate increasingly concise, entity-dense summaries of the above Article. Repeat the following 2 steps 5 times: Step 1: Identify 1-3 informative Entities from the Article which are missing from the previously generated summary and are the most relevant. Step 2: Write a new, denser summary of identical length which covers every entity and detail from the previous summary plus the Missing Entities. A Missing Entity is: - Relevant: to the main story - Specific: descriptive yet concise (5 words or fewer) - Novel: not in the previous summary - Faithful: present in the Article - Anywhere: located anywhere in the Article Guidelines: - The first summary should be long (4-5 sentences, ~80 words) yet highly non-specific, containing little information beyond the entities marked as missing - Use overly verbose language and fillers (e.g., "this article discusses") to reach ~80 words - Make every word count: rewrite the previous summary to improve flow and make space for additional entities - Never drop entities from the previous summary. If space cannot be made, add fewer new entities Remember, use the exact same number of words for each summary. Answer in JSON. The JSON should be a list (length 5) of dictionaries whose keys are "Missing_Entities" and "Denser_Summary".
13. Examples: ToT <branches> Explore multiple creative directions for the given request: **Branch A - Realistic Approach:** - Consider photorealistic style, lighting, and composition - Analyze what camera settings, angles, and environmental factors would apply - Evaluate feasibility and visual impact **Branch B - Artistic/Stylized Approach:** - Explore artistic movements, painting styles, or illustration techniques - Consider color palettes, brushwork, and aesthetic philosophies - Assess emotional resonance and creative interpretation **Branch C - Conceptual/Abstract Approach:** - Think about symbolic representations and metaphorical elements - Consider surreal, minimalist, or experimental visual language - Evaluate conceptual depth and meaning **Convergence Decision:** Compare all branches and select the most compelling approach, or identify ways to blend elements from multiple branches. </branches>
14. Examples: Tabular CoT Analyze the pros and cons of implementing remote work policies in a tech company. | step | reasoning | pros | cons | verification |
15. When and Where to Use Each Technique Structure Best Use Case Complexity Chain of Thought Linear sequence Logical reasoning, math problems Low Chain of Density Iterative refinement Summarization with controlled density Medium Tree of Thought Branching exploration Creative tasks, complex problem-solving High Table of Thought Tabular organization Multi-variable analysis, systematic evaluation Medium
16. Applying Context Engineering to a Real Business
17. Scan Me Github Repo Google Colab
18. Solving Problem’s for a Marketing Agency Inconsistent Quality Platform Specific Requirements Model Agnostic Frameworks Hyper Personalization Needs
19. Exploring Vertex AI’s Prompt Optimizer • Automatic Prompt Enhancement • Feed it your best work • It learns what makes it great • Generates optimal prompt formulas • How It Works • Input: Your top 50-100 performing pieces • Process: AI analyzes patterns of success • Output: Master prompts that replicate quality • Key Differentiator • Not another AI tool • It's an AI tool optimizer • Makes your existing AI work better • Platform Agnostic • Optimize for any model (Gemini, GPT, Claude) • Future-proof your processes • Switch models without starting over
23. Entropy & Co. Contact Us: Shaheer Khawaja +923491489359 shaheer@entropyand.co Got Questions? Let’s Connect