Implementing AI: Real-world Insights from Industry Experts

Artificial Intelligence (AI) is no longer a futuristic promise. It’s an operational reality reshaping businesses across industries. But behind every successful AI implementation are challenges, learnings, and strategic pivots that rarely get the spotlight.

We recently hosted an insightful Office Hours session where top experts shared their experiences navigating the complex, fast-moving AI landscape, along with cost-effective solutions to maintain competitive agility.

Here are some top insights our AI experts covered:

1. Staying ahead in a rapidly evolving AI environment

“How do you ensure that your AI projects stay current amidst rapidly evolving tools and models?”

The tech world moves fast, especially AI. OCR tools today costing $15 per 100,000 tokens might become obsolete or overpriced within months. Debashree Chatterjee highlights a pragmatic approach:

• API-first Design & Abstraction Layers:
  Keep business logic decoupled from specific AI services. Changes in underlying models become mere configuration updates, not re-engineering nightmares.
  
• AI Agent Frameworks:
  These offer flexibility, modularity, and easy tool integration. Regularly tracking over ten KPIs, including accuracy, latency, cost per interaction, and hallucination rate—helps decide if a switch is necessary.

AI technology evolves rapidly. What is cutting-edge today might become outdated or overpriced within months. A common challenge companies face is balancing innovation with stability. According to Debashree Chatterjee, Head of AI Automation at Astreya, this challenge is navigated by designing systems flexibly. Debashree provides the framework for this:

• Prioritize client-approved tools and the immediate business use case.
  
• Adopt an API-first design, separating business logic from underlying AI services for easy integration of new tools.
  
• Implement AI agent frameworks for better flexibility and modularity.
  
• Track KPIs beyond accuracy (task completion rate, latency, cost per interaction, hallucination rate) to determine when to update or replace existing tools.
  

2. Navigating unexpected competitor launches

“How do you adapt if a competitor unexpectedly launches a similar AI product?”

Vignesh shared a firsthand experience: Just as his team was about to launch a unique AI solution, a partner introduced a similar feature. Instead of panicking, they quickly assessed the competitor’s product and identified key differentiators.

• Pivot Your Messaging: Emphasize your unique value clearly and swiftly.
  
• Speed Matters: Expedite launch timelines strategically without sacrificing quality.
  

Vignesh’s team quickly assessed the competitor’s offering and realized their solution still had significant advantages. They pivoted by accelerating their launch timeline and shifting their marketing messages to highlight distinct strengths. As Vignesh succinctly put it, “A competitor’s launch isn’t fatal—it forces you to sharpen your value proposition and build a stronger, more resilient business.“

3. Tackling organizational readiness for AI

“What challenges do businesses face deploying AI?”

One of the biggest hurdles in AI implementation is not technical but organizational—often businesses aren’t truly ready for AI, despite their enthusiasm. Karthik highlighted this common mismatch between expectations and reality. His solution? Begin every AI journey with a comprehensive “reality check.”

One recurring pain point is the gap between organizational expectations and their readiness. Karthik emphasizes a critical step:

• Conduct a “Reality Check”: Assess current capabilities honestly. Does your organization have quality data? Is the ecosystem integrated?
  
• Invest in Data Engineering: High-quality, voluminous data tailored for specific AI use cases is crucial.

A formal assessment helps establish a baseline of the current state, laying out clear potential outcomes and benefits. A critical success factor he underlined is the availability of high-quality, comprehensive data. Without robust data, even the best AI systems falter. Investing in dedicated data engineering practices and frameworks is essential to building a foundation strong enough to support advanced AI solutions.

Vignesh adds the critical challenge of governance. GenAI capabilities evolve exponentially, while risk management and security often lag behind. He recommends a “Security by Design” approach—integrating security from the outset, offering secure, monitored environments rather than restricting AI usage entirely.

Check out our AI Agents on Service Now.

4. Increasing adoption of AI-assisted services

“What is the growth potential for AI-assisted services today and in the near future?”

AI-assisted services have transitioned from experimental to critical, mainstream tools, greatly enhancing user experiences. These solutions are increasingly becoming standard, handling tasks previously deemed too complex for traditional automation.

According to Rakesh, AI-assisted services have reached a crucial tipping point—moving from concepts to mainstream deployments. A significant trend is the shift toward hyper-personalized user experiences and sophisticated AI agents automating complex tasks.

Hyper-personalization is the trend to watch. AI agents routinely handle intricate requests and deliver highly personalized user interactions, transforming customer and employee experiences alike. Industry giants like Gartner predict the AI services market will hit $443 billion by 2027—a testament to AI’s transformative impact.

Want to see the AI agent demo in action?
Watch it on demand here: Watch now with passcode: %=$4vSYj

5. Aligning top management with AI strategy

“How do you engage stakeholders who lack deep AI knowledge?”

It’s common for top-level stakeholders to lack a deep understanding of AI’s nuances, often viewing it either skeptically or through exaggerated hype.

Debashree advises shifting the conversation from technology specifics to business outcomes. Present top management with tangible use cases, clear business objectives, and relatable success stories to bridge the knowledge gap.

Start conversations by clearly defining the objectives AI can achieve, supported by tangible success stories and relevant business cases. Highlighting proven results and realistic expectations ensures buy-in from senior management and aligns them strategically with AI initiatives.

6. Balancing AI automation with human oversight

“Should all AI-driven decisions require human oversight?”

Debashree advocates a balanced approach, advising that critical or complex tasks should retain a “human-in-the-loop” model to mitigate risks. However, procedural or low-risk tasks with reliable feedback loops can run effectively without direct human intervention. The key lies in assessing business-criticality and risk acceptance—ensuring a balance between innovation and safety.

Not every AI-driven task requires a human override, explains Debashree. Critical or complex tasks, however, should retain human oversight to mitigate risks, while procedural tasks can often safely run independently with robust feedback mechanisms.

7. Practical use cases: Enhancing enterprise workflows with ServiceNow

ServiceNow is pioneering AI-driven enterprise workflows. Vignesh highlighted their team’s achievement—developing and launching practical AI applications like the Attachment Summarizer — an AI agent that automatically reads and summarizes the contents of all attachments in a ticket. It then presents a concise summary of the issue and provides:

• Guided resolution steps
  
• Recommended knowledge articles
  
• Links to relevant resolution agents

This AI-powered agent automatically summarizes ticket attachments, guides resolution, and links to relevant resources, dramatically reducing triage and resolution times. It’s a plug-and-play solution that demonstrates AI’s practical value.

Check out our AI Summarizer Agent

Structured and unstructured Data: Navigating complexity

AI systems thrive on structured data, but handling semi-structured or unstructured data requires special consideration. Karthik explained that their platform currently processes structured and semi-structured data effectively, often using data flattening techniques. Unstructured data remains more challenging, necessitating dedicated, advanced solutions planned for future implementation. Understanding data types and their implications is crucial, as it significantly affects how AI agents perform downstream tasks.

Understanding AI solutions

Debashree highlighted one of the most pressing challenges when communicating AI capabilities to clients and stakeholders—managing expectations versus reality. AI hype often leads to unrealistic expectations, causing friction when actual capabilities don’t match exaggerated claims.

She recommends transparent conversations that explain what’s realistically possible within the client’s operational context. Instead of overselling technological capabilities, focus on foundational readiness, such as operational maturity, data availability, and ecosystem integration. Clear, honest conversations are key to successful client engagements.

Looking ahead: The future of AI implementation

AI’s potential is tremendous, but realizing this potential requires careful strategy, flexibility, and organizational preparedness. Leaders must remain adaptable to rapidly changing technology landscapes, foster strong governance frameworks, and maintain clear, realistic communication channels throughout their AI journey.

By embracing these insights shared by our experts, organizations can navigate AI implementation effectively, harnessing its power not just to innovate but to build resilient, competitive businesses for the future.

Ready to harness the power of AI? Book a call for a free AI assessment.