Tech CXO Says Leaders are Over-automating Instead of Fixing Bad Product Design
Shankar Sahai explains why CX leaders should fix bad product design and messy data before using AI to automate customer journey problems.
Shankar Sahai
CXO at UpendNow.com
AI Research: AI cuts qualitative research time from weeks to minutes by exposing specific user friction in real time.
Human Judgment: Humans must interpret industry context, redesign journeys, and manage sensitive service recovery decisions.
Contextual Support: AI equips support staff with complete session context, enabling faster, more relevant help without trapping customers in bots.
Design First: Better product design often beats personalization when users need simpler workflows, clearer interfaces, and faster search.
Data Foundation: Successful AI pilots require cleaned, unified data; otherwise automation amplifies contradictions, noise, and poor decisions.
Shankar Sahai is CXO at UpendNow.com, an AI production orchestration platform that streamlines video, film, and ad creation. His focus is on preventing the organization from building expensive things that nobody wants. And he does that with AI.
We sat down with Shankar to learn what he's automating — and where design trumps automation.
Making Sure the Human Experience Matches Boardroom Promises
Hey everyone, I’m Shankar Sahai, and I’m the Chief Experience Officer at UpendNow.com.
Continue Reading for Free
Create a free account to finish this article, plus get ongoing access to timely insights and practical resources.
Have an account? Log In
If I had to boil down what I do every day into a single sentence, it’s this: I stop companies from building incredibly expensive things that nobody wants or knows how to use. I spend most of my time at the chaotic intersection of product development, customer research, and operational strategy.
At UpendNow, we operate deeply within the B2B space, helping organizations navigate massive digital transformations and fix broken customer journeys. When we talk about "scale," it’s less about counting a massive volume of support tickets — though we see plenty — and more about tackling massive operational complexity. We examine multi-channel ecosystems where customers might bounce between a web portal, a mobile app, direct support, and a Sales rep all in the same week, expecting the company to magically remember every interaction.
As CXO, I help leadership teams stop viewing CX as just a "support desk" cost center and start treating it like a core business engine. We do the heavy lifting — from deep user research to workflow automation and agile transformations — to ensure the human experience matches the big promises made in the boardroom.
How AI Can Be Embedded in Qualitative User Research
We built an internal AI pipeline that automatically ingests, categorizes, and sentiment-maps raw conversation data in real time…That removed the guesswork. Our time-to-insight went from three weeks to minutes. In fact, we recently caught a major UX flaw in a new portal within 48 hours of launch, patched it immediately, and avoided a massive spike in support tickets.
Our biggest shift wasn't installing a flashy customer-facing chatbot. Instead, we embedded AI deep into our qualitative user research and post-interaction analysis.
We had a mountain of unstructured data — thousands of hours of user interviews, support transcripts, and open-ended feedback. Manually sifting through it for actionable insights used to take weeks. By the time we mapped friction points, the data was stale.
We built an internal AI pipeline that automatically ingests, categorizes, and sentiment-maps raw conversation data in real time. Instead of just flagging "happy vs. sad," it identifies specific behavioral roadblocks — like where a user gets confused by UI terminology or drops out of an onboarding workflow.
As far as tooling, we use OpenAI’s Whisper API for audio and video transcription. For analysis and categorization, we use Anthropic’s Claude (3.5 Sonnet).
That removed the guesswork. We replaced boardroom hunches with hard data. We now have a live dashboard showing exactly why and where users are struggling. And it increased speed. Our time-to-insight went from three weeks to minutes. In fact, we recently caught a major UX flaw in a new portal within 48 hours of launch, patched it immediately, and avoided a massive spike in support tickets.
A High-touch, AI-supported CX Workflow
Our whole world at UpendNow.com revolves around Solt OS, our platform that helps filmmakers and ad agencies automate the pre-production nightmare — taking a messy script and turning it into storyboards, shot lists, and budgets. Because filmmaking is so visual and messy, a standard support ticket workflow doesn't work for us. When a director is stuck trying to generate a pre-visualization video and character consistency keeps breaking, they don't want a generic IT help desk.
We handle this seamlessly now. Here's the workflow.
The moment a creator flags an issue or hits a wall, our internal AI doesn't just read their text complaint; it sucks in the entire context of their creative session. It looks at the script they uploaded, the prompts they used, and the storyboards that failed.
From there, the AI performs the heavy analytical lifting. It diagnoses the exact creative friction point — like realizing the filmmaker used a non-standard scene header that confused the system's location tracking.
Then, instead of making the user wait or talk to a bot, the system hands everything over to one of our human team members. But it hands it over with a complete cheat sheet. The human operator opens the case and instantly sees which scene numbers are looping, what the formatting glitch is, and a pre-drafted fix.
The filmmaker gets a fast, deeply contextual answer from a human who speaks their cinematic language, and we keep their creative momentum from dying. To close the loop, the AI automatically logs that specific formatting glitch into our engineering backlog so we can patch the product UI and prevent it from happening to the next director.
And at no point does the customer get stuck talking to a bot or explaining complex creative jargon to a support rep.
How AI Informs CX Decisions While Humans Guide Strategy
We follow a strict rule of thumb: AI handles the "What" and the "Where," but humans own the "Why" and the "How."
Here are a couple of examples of what AI handles well:
- Issue prioritization and trend spotting: AI scans thousands of incoming support transcripts and user sessions in real time. It instantly flags anomalies — like a sudden 15% spike in users getting stuck on a payment page — long before a human manager notices the pattern. We use PostHog for this.
- Journey friction mapping: AI analyzes user behavior across our platforms and highlights exactly where drop-offs occur. It eliminates guesswork. We're currently shifting our tooling to FullStory (StoryAI and API).
On the human side:
- Journey redesign and strategy: AI can tell me where people drop off, but it lacks the business context and creative empathy to redesign the experience. Fixing a broken B2B journey requires cross-functional politics, engineering compromises, and a deep understanding of human behavior. AI can't sit in a room and get three different departments to agree on a solution.
- Service recovery and escalation: If a system outage disrupts a high-value client's operations, making them furious, they do not want a perfectly generated AI apology. High-stakes recovery requires genuine human empathy, accountability, and the authority to bend the rules to make things right.
In other words, we don't automate human relationships. We use AI to handle analytical busywork so our team has time to focus on the high-value, deeply human work of building relationships and solving complex problems.
How AI Struggles with Industry-specific Context
Integrating AI into our qualitative data analysis and issue prioritization has transformed our operational pace.
Our team used to spend weeks manually tagging data to spot systemic UX flaws or onboarding bottlenecks across accounts. Now, AI flags these recurring friction patterns almost instantly. Catching these issues faster allows us to patch bugs and fix confusing workflows before they become a flood of customer complaints. This fundamentally shifted our product and research teams away from tedious spreadsheet data entry, allowing them to focus on designing solutions.
But it hasn't all been a smooth ride. Our biggest issue was analytical noise and a lack of true context.
Early on, we over-indexed on the AI's ability to tag sentiment and trends. We noticed a sharp spike in negative sentiment tags surrounding a newly released feature. The team almost scrambled to roll back the update. But a human deep-dive revealed that AI completely misinterpreted our power users' industry-specific jargon and technical venting. They weren't angry at the feature, they were just talking passionately about their own complex workflows.
Blindly trusting the automated dashboard without a human sanity check would have wasted weeks rewriting perfectly good code. AI finds patterns but struggles heavily with deep, industry-specific context.
Why Product Design Can Be More Important Than Fancy AI
Contextual hyper-personalization is another place where AI is underdelivering.
The promise was that AI could seamlessly look at a B2B customer's past behavior, current product usage, and support history to dynamically tailor their onboarding or dashboard experience in real time. We expected it to feel intuitive — like a digital concierge that anticipated exactly what a user needed next. Instead, it mostly felt intrusive or irrelevant.
In a complex B2B tech environment, user intent changes radically from hour to hour. A user might log in to do a quick, repetitive administrative task, but the AI — trying to be helpful based on broader account trends — clutters their view with recommendations for an advanced feature they do not care about right now. It added cognitive load rather than reducing it.
We realized that when navigating a platform, users do not want AI to guess what they want to do next; they want clean UI design, predictable workflows, and lightning-fast search functionality. We expected AI to solve a user experience problem that ultimately required better, simpler product design.
Why Data Quality is Critical Before Launching AI Pilots
I wish I had known how much our own messy internal data would screw things up early on.
When you listen to AI vendors, they make it sound like you just plug in this magical tool, and it instantly spits out brilliant, human-ready insights. But the reality is your AI is only as good as the data you feed it.
Before our first pilot, we had years of customer feedback, old user research, and support logs scattered across different silos. Some of it was outdated, some was flat-out contradictory, and a lot of it lacked the specific filmmaking context of what we do at UpendNow.
Because we didn’t clean up that data foundation first, the AI spent the first few weeks hallucinating bizarre trends and creating a ton of analytical noise that my team wasted time debunking.
If I could do it over, I would have spent way less time evaluating flashy AI features and way more time cleaning up our internal data architecture before turning the system on. You have to fix the foundation first, or you’re just automating chaos.
The Area That All CX Leaders Should Be Focusing On
The one area everyone needs to completely redesign right now is customer journey mapping…If you don’t redesign how you map and view the customer journey, you’ll use AI to optimize individual, isolated touchpoints while completely missing that the macro-experience is broken. We need to stop mapping the past and start orchestrating the present.
The one area everyone needs to completely redesign right now is customer journey mapping.
Historically, journey mapping was a static, boardroom exercise. People sat in a room with sticky notes, mapped an idealized path, put it in a PDF, and then nobody looked at it again for a year. It represents what we hope the customer is doing, not reality.
In an AI-augmented world, that approach is obsolete.
Leaders should be actively redesigning customer journey maps to be dynamic, real-time, and fed by live data. Because AI tools can now process unstructured data — like live user sessions, support calls, and drop-off points — in real time, your journey map shouldn't be a document. It should be a living, breathing software dashboard.
Instead of guessing where the friction is, your system should be actively telling you, "Hey, right now, filmmakers from boutique agencies are hitting a wall at step three of onboarding because of a UI lag." If you don't redesign how you map and view the customer journey, you'll use AI to optimize individual, isolated touchpoints while completely missing that the macro-experience is broken. We need to stop mapping the past and start orchestrating the present.
You Can't Automate Your Way Out of Bad Design
My advice is simple: Stop trying to automate your way out of bad design.
Right now, CX leaders face a massive temptation to throw AI at every single friction point in the customer journey because it feels fast and cheap. If your onboarding is confusing, throw an AI assistant at it. If your documentation is a mess, build an AI search bot. But all you're doing is putting a high-tech band-aid on a broken process. You are using AI to help customers navigate complexity that shouldn't exist in the first place.
Before you purchase another AI tool or launch another automated workflow, strip everything back. Look at your core product and your core service delivery. Fix the underlying UX flaws, clean up the messy data pipelines, and simplify the customer journey first.
Follow Along
You can follow Shankar Sahai's work on LinkedIn and X.
More expert interviews to come on The CX Lead!
