Speaker
Akshaya Murthy
Appearances over time
1 episodes
Episodes
1
Podcasts
Quotes & moments
Early Zendesk AI adopters cut email handle time by 92% using intelligent triage and routing powered by AI.
Zendesk AI Copilot customers have cut time spent responding to customer questions by 74%.
Hiring a single AI expert capable of building systems at scale costs at least $500K per year and is rising as the talent wars heat up.
A 2025 survey found 90% of companies say AI is central to customer loyalty, yet loyalty drops sharply when customers discover they are talking to a bot.
Zendesk's internal AI POC and platform implementations have delivered between 2x and 50x ROI.
OpenAI API pricing is approximately $2 per million tokens, while open-source alternatives like Llama can reduce this to fractions of a cent.
Zendesk serves around 100,000 brands including Airbnb and Shopify, handling billions of customer interactions per year.
Having clean, centralised data is a prerequisite for any effective AI strategy; messy data will break even the best AI model.
The majority of enterprise AI failures come from poor data, people issues, and unclear ROI — not from the AI algorithm itself.
AI has become the number one investment line item in company priorities, overtaking cybersecurity.
Integrating an off-the-shelf AI system like Zendesk can be done in one to two weeks, whereas building from scratch can take months to years.
AI model updates are happening on a weekly basis from commercial providers and on a daily basis on platforms like Hugging Face for open-source models.
Customer expectations have shifted from 'soon' to 'now'. They want instant resolution, full context of their history, and service that feels human — even if it's AI. Fail any one of those three and you've already lost them.
Most enterprise AI projects don't fail because the model is bad. They fail because the data is messy, the people aren't bought in, and nobody can prove the ROI. Fix those three things first.
AI washing is a buzzword salad with no specifics. If a vendor can't show you the KPIs, the defined problem it solves, and a measurable ROI, they're masquerading a rule-based engine as AI. The rice cooker with an 'AI Powered' sticker is not a joke — it's a category.
Time to market, total cost of ownership, talent availability, and security compliance are the four axes that should drive your build-vs-buy decision. Right now, on all four, integration usually wins.
Building AI from scratch is losing the argument in most enterprises. Integration wins on speed-to-market, TCO, and talent access — and you'd need months to years plus massive CapEx to match what you can get up and running in two weeks.
The architecture decisions that will make or break your AI product: clean data pipelines, a modular API-first model layer that lets you hot-swap LLMs, CI/CD for models themselves, and continuous monitoring with human review. These aren't nice-to-haves — they're the difference between a product and a liability.
The gap between commercial LLMs and open-source alternatives is closing fast. OpenAI charges ~$2 per million tokens; Llama gets you there for fractions of a cent. The endgame looks like internet pricing — you'll pay for a utility, not per megabyte.
Akshaya Murthy asked ChatGPT to make an action figure from his own photo. It named the figure 'Ankur Patel' — a stereotypical Indian-American name. Bias isn't theoretical. It's baked into the training data and shows up in production.
Don't launch an AI initiative because AI is the thing to do. Start with a problem statement, validate that AI is part of the answer, then work backwards to data, model selection, talent, and cost. Every AI project that skips this step is setting itself up to fail.
At hyperscale — billions of tokens per hour — the per-token cost math flips and self-hosted models win. Mission-critical verticals like defence, banking, and healthcare also need bespoke models because general-purpose LLMs simply weren't trained on the right data.
Responsible AI isn't a checkbox at the end of the build cycle. The NIST govern-map-measure-manage framework needs to run through every phase: pre-training, training, post-training, delivery, and application. Bias and privacy must be by design, not bolt-ons.
Zendesk's AI transformation produced three seismic product shifts: intelligent triage cuts resolution time to seconds, Copilots replace rule-based macros with live agent guidance, and workforce management becomes fully automated. In extreme cases, that's a 92% reduction in email handle time.
Analysis
What they talk about
- Technology 50%
- Business 42%
- Education 8%
