Despite her 20-year career as a coder, published author, researcher, mentor, and AI and cyber security expert, Mrunal Gangrade is a firm believer that, when it comes to expertise, AI levels the playing field. "I may have 20 years of experience and another person has one year, but AI puts us at the same level. Whoever learns first will benefit," she explains.
There's no shortage of talk about artificial intelligence (AI). Whether it's the latest tools, the faster resolution times that can be achieved, or the latency of modern voice AI, the conversations dominate watercooler moments, strategy planning, and board meetings.
However, Gangrade says the conversation must now move beyond talk of efficiency and automation to capture the bigger picture: not can AI do this task, but should AI do this task and can we trust it to do so consistently?
In this interview with CX Network, Gangrade explains where the accountability and governance gaps occur and how to close them.
The innovation accountability gap
In her work as a researcher, Gangrade has published several papers on AI explainability and has written about financial transaction security using AI and Blockchain. In the corporate world she has worked for Tech Mahindra, Kforce, Larsen & Toubro Infotech, Citibank and JP Morgan Chase & Co., and has received several accolades, including a Women in Tech Global Award and the Best Research Paper Award at ICDPN.
This broad experience means she has seen AI being added to journeys where it doesn't deliver returns, and processes where it doesn't achieve its goals. "Most organizations have implemented AI faster than they have implemented governance," she says. "And today, I believe there is a gap between innovation and accountability."
Gangrade says that when it comes to responsible AI, there are five questions CX leaders should be asking:
- How do we ensure AI decisions remain aligned with the customer interest and the company values?
- Can we explain AI-driven outcomes in a way customers and regulators can understand?
- How do we detect bias in AI before it reaches customers?
- Who is accountable when an AI system makes a poor decision?
- What safeguard exists to prevent AI from optimizing for business metrics at the expense of customer interest?
"Closing the governance gap requires collaboration across customer experience, technology, security, then GRC," Gangrade says. "Responsible AI should not be treated as a technology initiative. It should be viewed as a business and trust initiative."
Identifying where AI fails
The five questions become even more important as AI becomes more deeply embedded in systems and, as a result, we start to learn first-hand where it excels and where its limitations currently lie.
For example, silent failures like intent drift, where AI slowly and subtly deviates from how it was instructed to behave, are now better understood, but still mostly invisible on CX dashboards. AI can also drift from outcomes and "start optimizing results which are not ethical," Gangrade says. "This is why we need questions around the guardrails that are in place for those scenarios," she explains.
When AI systems are achieving a goal autonomously, ethical and accurate conduct must be monitored, rather than assumed. "With self-learning systems, while the system may technically improve a performance metric or achieve goals of the project, it can also create unintended consequences that may sometimes damage the customer relationship. In these cases, it has drifted away from the goal in a much wider way," she continues.
To tackle this, Gangrade says automated processes can be added to systems to detect drift and performance degradation, and govern how AI behaves. As always, human oversight is essential.
However, some instances of drift are the result of depending on the wrong data for the task. Giving the example of how a hospital may draw on a data set, Gangrade says that if the patient population shifts to from an older to a younger demographic and the AI model is not retrained on the new data, drift will occur. "Governance, monitoring and human oversight
of the results is very important," she says.
Incentives can also become problematic. "If the AI is 'rewarded' based on achievements, for example, getting the performance metrics perfect, then it could act with bias," Gangrade explains.
Measuring the benefits of AI
Measuring the benefits of CX is an established challenge – measuring the benefits of AI brings a new level of complexity.
In 2026, CX Network's annual research into the state of CX found proving ROI is the number one obstacle to investment and that 52 percent of practitioners believe the pressure to prove ROI is increasing.
When it comes to measuring the benefits – and returns – on costly AI investments, many service leaders need to track how handle time, resolution rate and satisfaction compare pre- and post-deployment. Journey designers may want to track how orchestration moves the needle on sales, while others may be looking at cross-sell and upsell rates after introducing a conversational interface to help customers navigate a product catalog.
"To measure any AI you, we, the project manager or whoever needs to first identify the output. There should be a measurable output and when AI gets deployed there should be measurable statistics whether the AI has actually achieved its outcome," Gangrade says.
When AI is the infrastructure, the measurement mindset needs to shift.
"AI is now supporting the infrastructure in many organizations and that is a very measurable output. You can compare the time taken pre- and post-deployment, for example. AI can also be used to identify the cyber security vulnerabilities. That is also a very measurable part: how many vulnerabilities were you able to identify with the help of your AI."
Despite the governance challenges practitioners still need to address, Gangrade says CX is "more secure and more interactive than ever before". And while some challenges still need to be ironed out, AI has improved CX for the better.
"There are a lot of gaps in the guardrails and governance, but from the UI perspective or what the customer experience right now it is good – it just needs to be governed."
Quick links
- VIDEO: JP Morgan I Operationalizing AI in CX: Practical steps, risks and guardrails
- Using AI to build CX tools: What to know and how to get started
- Changing the thinking around data and AI governance