From Data Pipelines to AI Agents with Angshuman Rudra

Angshuman Rudra is transforming how marketing teams work with data and AI. As a product leader at TapClicks with nearly two decades of experience spanning Yahoo, Adobe, and the MarTech space, Rudra has built AI-powered platforms that process over one million campaigns monthly for 5,000+ agencies and brands. His work focuses on AI agents, data engineering, and marketing infrastructure, helping organizations move from manual reporting to intelligent, automated operations.

Rudra explains how AI agents are changing marketing operations, why data strategy is the foundation of successful AI implementation, and what it really takes to scale AI in enterprise environments.

Can you explain exactly what “AI-powered marketing infrastructure” means and why is it becoming essential for modern marketing teams?

AI-powered marketing infrastructure refers to the integrated system of tools, technologies, and processes that apply AI (particularly machine learning, natural language processing (NLP), and generative AI) to optimize, automate, and personalize every aspect of the marketing lifecycle.

This infrastructure enables marketers to move beyond static automation and into a new era of real-time, data-driven, and scalable decision-making.

Key Components:

• Data Collection & Integration: Aggregates customer and campaign data from CRM, websites, social media, email, ads, chatbots, and more – creating a unified, AI-ready data foundation.
• ML & Predictive Analytics: Powers segmentation, forecasts customer behavior, and recommends actions across the funnel – from lead scoring, channel optimization to many use cases.
• NLP & Generative AI: Enhances content creation (copy, images, video), supports chatbots and assistants, and enables scalable social listening and sentiment analysis.
• AI-Enhanced Automation Platforms: Refines campaign targeting, ad bidding, email delivery, and posting times, continuously learning and adapting at a scale not possible earlier.
• AI-Driven Analytics & Reporting: Delivers instant (or more frequent than earlier), actionable insights and optimization recommendations – removing reliance on manual analysis.

Why It’s Becoming Essential:

• Scalability: Enables personalized experiences at scale—something humans alone can’t deliver.
• Speed to Insight: AI dramatically reduces time-to-insight and enables real-time campaign adjustments.
• Better ROI: Resources are allocated dynamically based on predictive signals, boosting conversion and efficiency.
• Customer Experience: From proactive chatbots to 1:1 personalization, AI ensures every touchpoint is relevant and timely.
• Strategic Shift: Frees marketers from manual execution so they can focus on orchestration, strategy, and brand building.

Leading marketing teams are already seeing the impact. According to BCG and Google’s Dec, 2024 survey, companies that embrace AI-powered infrastructure report 60% higher revenue growth and adapt to market changes 2x faster than their peers. This is going to get much more drastic and it’s essential for marketing teams to adapt AI faster.

You’ve built AI Agents that integrate with marketing dashboards to summarize performance and generate actionable insights. Can you walk us through how that works in practical terms?

At a practical level, our AI Agents sit on top of marketing dashboards and connect directly to filtered performance data – campaigns, channels, geos, budgets, and KPIs. AI-strategy really is Data-Strategy and one of the new terms of Context Engineering.

Here’s how they work:

1. Context Awareness: Each agent automatically inherits the dashboard context – filters, date ranges, metrics – so the insight is always relevant to the user’s current view. This is key!
2. Data Interpretation: Using AI-driven analytics and RAG (Retrieval-Augmented Generation), the agent scans trends, outliers, pacing, and anomalies across campaigns. We also store meta-data in a structured way and that helps us augment the model output.
3. Our platform and the output of the AI Agents are used by thousands of business owners who are looking for quick insights that are still actionable and not filled with jargons. Our insights are generated with that persona in mind and generates a human-readable summary. We have other AI Agents that cater to other personas.
4. Our agents aren’t just solving the efficiency problem. Many of our users have come back and said the insights have given them a different perspective and have made them smarter. Our agents are developed to improve as we refine the context we provide with each use.
5. Some of our agents go beyond reporting and we have developed operator agents. For example, a budget pacing agent might recommend reallocation across channels, while a creative agent might suggest copy or asset changes based on past success. That is the next level where our orchestration layer on top of our agents can not only glean insights but perform Operations based on those insights.

AI is not just summarizing dashboards but replacing static reporting with dynamic, decision-ready insights. And with the orchestration layer, we’re building toward automated marketing operations driven by AI-generated intelligence.

Your AI systems use RAG and ReAct methodologies. What’s the difference between these approaches? Why are they particularly effective for marketing operations?

At TapClicks, we use both RAG (Retrieval-Augmented Generation) and ReAct (Reasoning + Acting) to power our AI Agents – each playing a distinct role in how we deliver intelligence and automation to marketing teams.

Retrieval-Augmented Generation (RAG) is at the core of our Insight Agents. It retrieves relevant information from structured and unstructured sources – dashboard filters, performance data, client goals, meeting notes, order info, industry benchmarks – and combines that with LLMs to generate grounded, context-specific outputs. This ensures our agents deliver accurate, explainable insights, not hallucinations or generic summaries. RAG is particularly effective for:

• Performance summaries
• Anomaly detection
• Benchmark comparisons
• Client-ready narratives tailored to current dashboard views

ReAct: Reasoning + Acting

We’re now incorporating ReAct to go further – from insight to decision-making. ReAct enables our agents to reason through multi-step problems, evaluate conditional logic, and take context-aware actions. Think of it as “if-then” logic powered by AI and guided by data.

This aligns directly with our mission at TapClicks: we’ve always focused on automating tedious, repetitive tasks in AdOps and Marketing Operations. But with ReAct and generative AI, we can now scale that automation across far more complex workflows – where human-like reasoning is needed.

ReAct powers our Operator Agents, which can:

• Assess pacing or ROI conditions
• Recommend or simulate budget reallocations
• Trigger workflow updates or downstream actions via orchestration

These agents operate within our orchestration layer, which combines structured metadata, tool integrations, and business logic—allowing autonomous and semi-autonomous agents to reason, decide, and act in real-time.

Why This Matters for Marketing Operations

• In a world overwhelmed by data, our RAG-based Insight Agents surface fast, accurate, and actionable insights rooted in real context.
• Our ReAct-based Operator Agents bring intelligent workflows that can adapt, reason, and automate complex and repetitive tasks.

Together, these methodologies are transforming marketing operations – from dashboards and manual reporting to intelligent systems that think, explain, and act.

You’ve been leading AI initiatives at TapClicks. What role has AI played in scaling the business, and what have you learned from deploying AI at enterprise scale?

At TapClicks, AI has been central to how we’ve scaled both product capabilities and operational efficiency. We’ve used AI to transform manual, repetitive tasks across data aggregation, reporting, and campaign operations – empowering thousands of marketing teams to move faster and make smarter decisions.

But scaling AI in an enterprise context is not just a technical effort – it’s a strategic transformation. Success hinges on getting the 3 Ps right: People, Process, and Platform – alongside critical lessons we’ve learned from real-world deployment.

People: What We’re Seeing Is a Top-Down AI Adoption Pattern—But Trust and Enablement Still Matter

Across the industry, most AI adoption today is top-down. Leadership teams are driving strategy, selecting tools, and piloting initiatives. But this approach often lacks structured enablement for teams on the ground.

Within any organization, there’s a wide spectrum of readiness – from early adopters to skeptics. Change management and training must be continuous and I’m seeing success in those organizations where there is an intentional investment in the following:

Change Management is Critical Organizations that succeed are the ones that treat change management as a core system. This includes training, internal communication, and active involvement from functional teams.

Human-in-the-Loop Builds Trust AI shouldn’t replace judgment – it should enhance it. We’ve seen greater adoption where AI provides recommendations with context, while giving marketers the final say. This “co-intelligence” model increases trust and long-term effectiveness.

Process: Rethink What to Automate – and Prioritize for Business Impact

Many organizations attempt to overlay AI onto legacy processes. But simply automating existing workflows and processes misses the point.

Strategic leaders go further: they encourage teams to reimagine what’s possible, not just optimize what exists. My suggestion is to do not use AI just to get you faster horses, but challenge existing processes.

Start Narrow, Show ROI, Then Scale That said, organizations still need to start with specific use cases – like pacing summaries, budget recommendations, or client reporting – and expand only once measurable value is proven. This is connected with the people and change management problem and the right balance is needed.

Platform: Technology Is Available – But Data Maturity Is the Differentiator

While the tooling has matured rapidly – LLMs, orchestration layers, co-pilots – the real barrier is data readiness. Many organizations still struggle with inconsistent schemas, missing metadata, or siloed data streams.

Data Quality Is the Bottleneck – and the Foundation Enterprises that get the most out of AI invest early in data governance, schema design, naming conventions, and validation pipelines. AI strategy depends on the Data Strategy.

Infrastructure Must Scale with Intelligence As AI scales, performance and cost become real constraints. Smart organizations are optimizing cloud workloads, using GPUs selectively, and designing for responsiveness and SLAs.

Ethical AI and Bias Aren’t Optional Especially in marketing, unchecked AI can over-optimize or reinforce bias. Leading organizations are now investing in audit trails, monitoring frameworks, and fairness reviews as part of their AI maturity roadmap. A scalable eval process that takes into account the above factors is critical for long term success.

AI is helping organizations scale – by replacing manual decisions with data-driven systems, and transforming reporting into recommendations. But long-term success depends on aligning the 3 Ps:

• People: AI adoption is top-down today, so leadership must support training, feedback, and trust-building for better adoption.
• Process: Don’t just automate legacy workflows. Use AI as an opportunity to reimagine your operations for the AI-native world.
• Platform: AI won’t differentiate your business unless you invest in data readiness and infrastructure. The winners are those who treat data strategy as a strategic asset.

Building AI products for marketing teams presents unique challenges—how do you create AI tools that work with diverse data sources and varying technical expertise across marketing organizations?

At TapClicks, we serve both the operations team and the end consumers of dashboards and insights. Operations teams need control and customization to support stakeholders at scale, while end users prefer simplicity and guided experiences.

We follow a dual approach. For non-technical users, we use a “convention over configuration” model – offering smart defaults and templates that reduce setup time. For power users, we enable deep configuration and flexibility where needed. We’ve been intentional from day one about solving for both ends of the spectrum of marketing personas.

If you’re starting from scratch, I recommend designing for the non-technical persona first. Demonstrating quick wins here drives broader adoption and builds internal momentum.

What’s the reality versus the marketing buzz when it comes to practical AI implementation in business?

While AI is often marketed as a plug-and-play or turnkey solution, the reality on the ground is more nuanced. Many organizations are still grappling with foundational issues: lack of a clearly defined end goal, fragmented data environments, and siloed teams each focused on their own metrics, tools, and processes.

This fragmentation creates a major gap between AI’s promise and its practical impact. Without shared objectives, unified data, and collaborative ownership, AI initiatives often remain stuck in isolated MVPs or fail to deliver sustained value. What we consistently see is that AI only drives transformational outcomes when it becomes a cross-functional effort – bringing together marketing, operations, IT, analytics, and executive leadership around common goals and integrated workflows.

The main problem is still organizational alignment and operational readiness required to put AI to work. Success depends on addressing data quality, change management, process reengineering, and cross-team governance. Until those foundations are in place, AI will underdeliver.

Your experience spans both traditional data engineering and modern AI systems. How has the evolution from big data to AI changed how companies should think about their data infrastructure?

Over the past 15+ years, we’ve witnessed a fundamental shift: from collecting data at scale during the big data era to continuously extracting intelligence from data in the AI-driven era. This evolution has changed not just the tools, but the mindset companies need to adopt around data infrastructure. And in my opinion data strategy that can power the right AI solutions has and will be the key differentiator in the next few years.

In the Big Data era (2005–2015) The focus was on storage and scale – building Hadoop clusters, managing ETL jobs, and structuring enterprise data warehouses. Infrastructure was centralized, batch-oriented, and primarily built for descriptive analytics. The priority was to capture everything, even if you didn’t know how it would be used.

In the Modern AI era (2020 onward) Now, infrastructure must be built for speed, adaptability, and intelligence. AI workloads require:

• Real-time or low-latency data pipelines
• Access to unstructured and semi-structured data
• Versioned, high-quality datasets for training and inference
• Flexible compute that supports both analytics and machine learning

One of the most important factor for AI to succeed is Context.. For AI systems and agents to reason effectively, they must access data that is clean, governed, well-modeled, and embedded with business meaning. This is where metadata, data contracts, and unified governance frameworks play a foundational role.

Modern AI Systems AI is both an opportunity and a forcing function. It requires companies to:

• Rethink their data stack (e.g., lakehouse architecture, streaming-first pipelines, open table formats like Iceberg)
• Automate aggressively (via DataOps, observability, lineage tracking)
• Embed intelligence and feedback loops into pipelines (e.g., real-time predictions, self-healing workflows, autonomous agents)

The bottom line? AI has brought data infrastructure and data strategy as a strategic differentiator. Companies must now architect for adaptability, observability, and intelligence for a successful AI strategy and to compete.

You’re developing courses on product management with AI products. What are the biggest mistakes you see companies making when building AI-powered products?

1. No Clear Purpose or Deliverable

Teams often try to build AI agents without clearly defining what value the agent is going to provide – for whom, and why. Without a scoped task and measurable output, projects drift or become demos/POCs that never ship.

There has to be real value for the user and the business. Like any product or feature, AI is just a tool. If it doesn’t solve a real problem, it’s not worth building.

2. Unstructured Inputs and Outputs

AI is not magic – garbage in, garbage out. Many teams skip structured input/output design and rely on free-form prompts and responses, which makes the system brittle and hard to scale.

Your AI feature must be grounded in real data and context. The more intentional you are with context engineering – how you pass schema, and business logic – the more reliable your results will be.

3. No Guardrails or Instruction Tuning

Treating AI as a black box leads to unpredictable behavior. Without clear system roles, prompt structure, and instruction tuning, outputs vary too widely across users or scenarios.

You need to design your agents with consistency and safety in mind from the start, especially when they touch production data or customer-facing workflows.

4. Ignoring Reasoning, Memory, and Orchestration

Real agents often require multi-step reasoning, memory, and external tool use. Teams that treat AI as a one-shot question-answer system miss the opportunity to automate meaningful workflows.

Production AI products need more than LLM calls. They need routing logic, memory stores, function calling, and integration into your existing backend and data layers.

5. Poor UX and Lack of Role-Based Design

Even great AI won’t get used if the UX doesn’t match the user’s needs. Many teams build one-size-fits-all interfaces that fail both novice and power users.

While chat-based prompts are popular today, that doesn’t mean they’re the right interface for every use case.

Build UX features that are useful for your users – whether that’s guided flows, inline actions, or form-based assistants.

6. No Scalable Evaluation Strategy

Too many teams rely on ad hoc testing or gut checks. Without a way to measure quality and consistency across varied inputs, it’s hard to know if your AI is reliable.

Build scalable evaluation pipelines that test across datasets, edge cases, and personas. Define success metrics, create golden test sets, and use judge LLMs or scoring functions to monitor drift and performance over time.

Successful AI products aren’t built with prompts alone. They’re built with structured data, system-level thinking, thoughtful UX, and constant evaluation. That’s the mindset we teach – and the one more teams need to adopt.

With almost 20 years in the tech space and nearly a decade at TapClicks, you’ve witnessed multiple technology cycles. How do you evaluate which AI trends have staying power versus which are temporary hype?

With every new wave in Data and AI or any other technology, it’s tempting to chase the latest trend. But over the years, I’ve found three consistent signals that help me decide if a trend has staying power:

1. Are organizations redesigning their workflows or systems around it? If a trend is reshaping how work gets done – not just improving an existing process – it’s likely foundational. Look for changes in how teams operate, how platforms are architected, and where budget is being committed.
2. Is an ecosystem forming around reliability, governance, and scale? True adoption needs more than demos. It requires tooling for evaluation, monitoring, compliance, observability, and safety. If vendors and open-source tools are building serious infrastructure around a trend, it usually means the enterprise is paying attention. I will say some of these are pushed by larger companies or VC investment and you may have to be careful about the future of a trend once the artificial investment dies down. Modern Data Stack is one such example that has died in the last two years because of lack of investments.
3. Does it drive real productivity or capability gains? Is it solving a pain point or unlocking something meaningfully new? Technologies that become permanent usually shift cost structures, enable new business models, or compress time-to-value in significant ways. I think the move to cloud infrastructure (separation of storage and compute and the business model) enabled an entire new business model.

LLMs and Agentic AI: Transformative, But Not Equal

Some trends meet all three criteria. Others meet one or two, and still prove valuable, but only  in the right context.

LLMs, in my view, are one of the most important technology shifts in decades. Like the internet, mobile or cloud, they are horizontal in impact – touching everything from writing and coding to research and analysis. The supporting ecosystem is robust, and organizations are already redesigning internal workflows, customer interfaces, and even org structures around them.

Agentic AI, on the other hand, is a powerful concept – but context-dependent. The idea of autonomous or coordinated agents shines when you need multi-step reasoning, chaining of decisions, or collaboration across tools. But it’s not always the right answer. Many use cases don’t benefit from agent coordination and are better served by lightweight LLM APIs or deterministic automation.

At TapClicks, we’ve been thoughtful about this distinction. We’ve been building a marketing operations platform for years – solving problems around data harmonization, automation, and decisioning. So in our case the Agentic AI trend makes sense- as it’s a continuation of what we already enable. Our agents operate in contexts we understand deeply, and that’s what makes them valuable.

The key is not whether a trend is “hot.” The key is whether it is the right fit for you, your users and their use cases, and your business goals.

Looking ahead, how do you see AI transforming the MarTech landscape? Will future marketing operations rely more heavily on autonomous AI systems, and what does that mean for marketing professionals?

AI agents are profoundly transforming marketing operations, notably through enhanced insights and significant automation. For insights, we’re already observing substantial value as AI agents leverage their extensive knowledge to provide a deeper understanding of customer interactions. They excel at tasks like summarizing messages and content, personalizing content to individual preferences, and enriching customer profiles with more granular detail and predictive understanding. This capability extends to distilling unstructured data, such as call recordings or chatbot transcripts, into structured, actionable insights that traditional methods might miss.

For automation, we’ve begun to see AI agents capable of automating specific workflow segments, particularly tasks like summarizing messages and personalizing content, as these no longer inherently require human intervention. Organizations are increasingly achieving this, with greater success evident in those already possessing a robust data infrastructure for AI models to readily access. At the enterprise level, ensuring underlying infrastructure supports security, reliability, and data governance is critical for AI agents to achieve optimal outcomes.

We are already seeing this transformation in SMBs, where greater agility often allows for quicker adoption. However, it’s more difficult to effect this change in mid-market and enterprises due to their more complex workflows, heightened focus on data privacy, data governance, and regulations. For martech vendors, this means more pushback, as organizations will increasingly opt to build their own custom software, leveraging the rising ease of AI-powered “hypertail” solutions over traditional “buy” options.