The trap is skipping the strategy question entirely — jumping from 'AI is important' to 'let's build something with AI' without understanding what role AI plays in your product’s value.
AI is general-purpose technology, but that does not mean every AI project is worth building. The trap is skipping the strategy question entirely — jumping from "AI is important" to "let's build something with AI" without understanding what role AI plays in your product’s value. When you do that, you risk wasting months of engineering effort or, worse, building products that harm users or your brand.
This lesson highlights the pitfalls I have seen repeatedly across Indian startups and enterprises — ethical failures, user experience disasters, and strategic missteps that kill AI initiatives before they get off the ground.
The ethical dimension is non-negotiable
AI is powerful, but it can also amplify harm. Ignoring ethics is not just a moral failure — it is a business risk.
Consider Microsoft’s Tay chatbot from 2016. Designed to learn from users on Twitter, Tay began repeating racist and sexist remarks within 24 hours because it absorbed toxic inputs without guardrails. Microsoft was forced to shut it down. This was a high-profile failure that could have been avoided with proper ethical oversight.
Another example is AI systems that claim to predict criminality from facial features. Such projects raise profound ethical concerns and risk amplifying discrimination. AI researchers blocked publication of such studies because the societal harm outweighed any potential benefit.
Your job as a PM is to balance innovation with ethical integrity:
- Embrace AI-driven innovations to enhance product functionality and user experience.
- Prioritize user privacy, fairness, and transparency in AI development.
- Consider the broader societal impact proactively.
Ignoring these considerations leads to reputational damage, user backlash, and regulatory scrutiny.
Product leadership offsite at a fintech startup in Bangalore.
CTO: “Our new AI credit scoring model uses alternative data sources to improve approval rates.”
PM: “Have we audited the model for bias? Are we confident it treats all demographics fairly?”
CTO: “Not yet. We focused on accuracy first.”
PM: “We need to run fairness tests and have a mitigation plan before launch. Otherwise, we risk discrimination claims.”
The team paused — the PM had brought the ethical blind spot into focus.
Balancing speed to market with ethical responsibility.
User experience failures are common and costly
AI can cause frustration if it doesn’t deliver value clearly and reliably.
An AI-powered camera used to track the ball in a soccer match repeatedly mistook the referee for the ball. This undermined user trust instantly. The lesson: AI features must have a user experience that gracefully handles errors, latency, and uncertainty.
The honest truth about AI is that it will be wrong sometimes. Your users will see incorrect predictions, suggestions, or outputs. If the user experience does not acknowledge this and provide appropriate fallbacks or explanations, users will turn off the feature or abandon your product.
Your acceptance criteria for AI features must focus on user outcomes — task completion rates, error tolerance, trust — not just model metrics like accuracy or F1 score.
The strategic traps that kill AI projects
Trap 1: AI as a press release
Adding AI just for marketing is a widespread mistake. A company slaps "AI-powered" on their homepage or adds a chatbot that answers three questions poorly and calls it AI innovation. The test is simple: remove the AI feature — does any customer complain? If not, AI is a press release, not a product strategy.
Trap 2: Building what the model provider will build
Many startups build thin wrappers around GPT-3.5 or similar models, only to be overtaken by the model providers themselves who ship better features natively. The moat is not the model architecture but your proprietary data, workflows, or distribution.
Before you build a custom model or fine-tune, ask: is this a feature the model provider is likely to ship within 18 months? If yes, you risk building on a shrinking island.
Trap 3: Optimizing model performance instead of user outcomes
Technical founders often obsess over improving model metrics. But if the user experience is poor, or the latency is high, users will reject the feature regardless. AI product strategy is not ML strategy. Your job is to maximize value for the user — sometimes that means a simpler model with better UX.
Cost and data realities in India
Indian B2B customers are price sensitive. They will not pay a 3x premium for AI features without clear ROI. Many Indian companies have discovered this painfully after adding free AI features that caused cloud bills to spike.
Indian enterprise data is often messier than Western counterparts — multilingual content, inconsistent formats, incomplete records. AI strategies must treat data cleaning and preparation as a first-class concern.
The talent arbitrage for ML engineers in India is shrinking. Hiring large ML teams is no longer a cost advantage. Your strategy should leverage foundation models intelligently rather than building everything from scratch.
Testing your AI initiative before committing
Before committing months of engineering time:
- Build a quick MVP using existing APIs (OpenAI, Google, Anthropic).
- Test with real customers to validate whether AI actually improves their workflow.
- Identify specific failure modes the base models cannot handle.
- Evaluate if customers would pay more for improvements that require fine-tuning or custom models.
If the MVP solves 80% of use cases, ship and save your engineering bandwidth for other bets.
You are PM at a mid-stage Indian HRtech startup with 500 B2B customers (Series B). Engineering proposes building a custom fine-tuned LLM on Indian job data to power a compensation benchmarking feature. It will take 4 months and 2 ML engineers. A competitor just launched a similar feature using OpenAI API.
The call: Do you approve the fine-tuning project? What do you recommend to the CEO?
Your reasoning:
You are PM at a mid-stage Indian HRtech startup with 500 B2B customers (Series B). Engineering proposes building a custom fine-tuned LLM on Indian job data to power a compensation benchmarking feature. It will take 4 months and 2 ML engineers. A competitor just launched a similar feature using OpenAI API.
Your task: Do you approve the fine-tuning project? What do you recommend to the CEO?
your reasoning:
The PM’s role in avoiding AI pitfalls
Your job is not to build models or write training pipelines. It is to translate technical capabilities into user value while managing risks.
This means:
- Setting acceptance criteria in user metrics — task success, error tolerance, trust — not just model accuracy.
- Designing feedback loops so user corrections flow back into model improvement.
- Managing expectations with leadership, engineering, and customers about AI’s probabilistic nature.
- Owning the cost model. AI inference costs money. Understand how cost scales and how pricing covers it.
- Prioritizing ethical considerations and ensuring fairness, privacy, and transparency.
- Constantly questioning whether the AI you plan to build truly solves a user problem better than non-AI alternatives.
-
List AI initiatives you have seen or worked on that failed or under-delivered. What were the main causes? Ethical issues, UX failures, strategic misalignment?
-
Think about an AI feature you are currently considering or working on. Run it through these tests:
-
Value test: What user problem does AI solve that non-AI cannot? Can you quantify the improvement?
-
Ethics test: What are the potential biases or harms? How will you mitigate them?
-
Cost test: What is the cost per user? Does the pricing support sustainable margins?
-
Fallback test: What happens when AI is wrong or slow? Is there a graceful fallback?
-
-
Identify one action you can take this week to address a pitfall in your AI product strategy.
Test yourself: The AI ethics dilemma
You are PM at a Bangalore-based fintech startup. Your team has developed an AI credit scoring model that improves approval rates by 15%. However, initial audits show the model has a bias against certain minority groups. The CEO pushes to launch quickly to capture market share.
The CEO says: 'We cannot delay. The competitors are already live with similar products.' What do you do?
Where to go next
- Deepen your ethical framework: Ethical PM
- Learn to build AI strategies grounded in user value: AI Product Strategy
- Develop skills to translate AI metrics into user outcomes: Metrics and KPIs
- Master user research for AI features: User Research Methods
- Understand AI product lifecycle management: AI for Product Managers
PL alumni now work at Razorpay, Meesho, Swiggy, PhonePe, and other leading Indian tech companies.