AI Adoption is Like Handling Fire

September 5, 2025 Shammy Narayanan Applications & Technology, Artificial Intelligence, Elevating Architecture, leadership, Machine Learning, Software, Strategy & Planning 0

Standing at the GenAI frontier, I see business owners reacting in two very different ways  some rushing headlong into the wave, others waiting cautiously on the shore

By Shammy Narayanan

The first is the FOMO tribe  those driven by the fear of missing out. They rush into the AI race without truly understanding it beyond a handful of GPT prompts. You’ll find them attending every AI conference, spending heavily, and often relying on “YouTube and WhatsApp scholars” as decision-makers.

This makes them easy prey for marketing pitches. Vendors sell half-baked products under the glitter of “AI-enabled,” convincing them to pursue use cases that add little or no real business value. In their haste, they build costly AI teams, buy expensive tools, and overlook due diligence. The fallout? Burnt dollars, disillusioned legacy teams, and yet, a misplaced sense of belonging  the comforting illusion that they are “in the AI herd.”

The second group I call the Bay Watchers. They stand on the shore, waiting for the tide to calm before stepping in. This strategy worked during the early days of mobile adoption, and to some extent, with Cloud  where early adopters did face stability hiccups and a few even rolled back. But AI is a different beast. It’s not a ripple, it’s a tidal wave. Waiting too long is perilous. Business thrives on calculated risks, and sitting too long on the bench could leave you so far behind that catching up may become impossible.

AI adoption is like handling fire, grab it carelessly and you’ll get burned; wait too long and you’ll be left in the cold. Instead, consider this balanced path forward:

Steps for AI Enablement

  1. Shut off the noise
    • AI marketing today is aggressive, often overwhelming. If you enter with a preconceived mindset or in blind haste, judgment gets clouded.
    • Remember: no great decision was ever made by a terrified mind.
  2. Form a cross-functional AI team
    • Bring together not just technologists (preferably architects), product owners, security engineers, and data analysts, but also other critical stakeholders.
    • Include compliance and legal experts to handle privacy/IP concerns, risk and audit representatives to flag ethical and reputational risks, and operations specialists to spot real workflow opportunities.
    • Depending on maturity, you can also involve HR/L&D teams to drive awareness and adoption, and customer experience leaders to ensure AI enhances rather than disrupts user interactions.
  3. Secure your boundaries
    • Don’t let employees use frontier models inside your organization with unchecked access to sensitive code or data.
    • Draft clear AI usage policies  the dos and don’ts with AI tools and LLMs  and enforce them diligently.
    • Remember, even something as simple as an employee pasting source code or customer data into a public chatbot can inadvertently leak intellectual property. For example, in 2023, a leading semiconductor company discovered engineers had uploaded proprietary chip designs into ChatGPT. The AI didn’t “steal” it, but the data was now out of the company’s control.
  4. Formulate your AI strategy
    • Decide where AI should create impact:
      • Elevating end-user experience (e.g., conversational assistants in banking).
      • Enabling internal productivity (e.g., code copilots for developers).
      • Building new offerings (e.g., AI-driven health monitoring apps).
  • Prioritize use cases not just by business need, but also by data readiness. A brilliant idea still fails if your data foundation is weak. For instance, a retail company may want to build a personalized recommendation engine like Amazon. But if their customer data is scattered across multiple systems, poorly labeled, and inconsistent (that’s the case in most of the organization) the model will produce irrelevant suggestions and frustrate users. In contrast, the same company might achieve far more impact by starting with a simpler but data-ready use case  like AI-driven demand forecasting  where structured sales and inventory data is already reliable.
  1. Tool selection – Funnel, don’t hunt
    • Avoid the goose chase of trailing every tool. Instead, define funnelling criteria:
      • Users → Who will use it? (e.g., marketing teams may benefit from Jasper or Copy.ai, while engineering may lean towards GitHub Copilot).
      • Volume → How much scale do you need? (e.g., customer service chatbots like Ada vs. large-scale platforms like LivePerson).
      • Licensing → Global or localized rights? Does it fit your compliance needs?
      • Compatibility → Does it integrate seamlessly with your existing systems (e.g., Microsoft 365 Copilot in a Microsoft-heavy ecosystem)?

By this stage, you’ve only begun scratching the surface of AI enablement. But it’s a disciplined start neither reckless nor hesitant

So, what comes next on this journey? Stay tuned for the next edition of Navigating GenAI Frontier.

Shammy Narayanan is the Vice President of Platform, Data, and AI at Welldoc. Holding 11 cloud certifications, he combines deep technical expertise with a strong passion for artificial intelligence. With over two decades of experience, he focuses on helping organizations navigate the evolving AI landscape. He can be reached at shammy45@gmail.com.