My First Customer Was Me

Authored by: Aarohi Kulkarni

I’m the founder of DeskClone AI, a platform where businesses build AI support agents for their customer support. For the first few months, I was losing visitors in my sleep. Literally. I’m in India; most of my website visitors are not. Their questions arrived while I slept. By the time I replied in the morning, the person was long gone.

I had tried the obvious fixes years earlier, in a previous venture. A live-chat widget looks like an answer, and for the first few days it worked: it pinged at midnight, I got up, and I talked to the visitor with full enthusiasm. The enthusiasm wore off fast. The pings started going unchecked, and eventually my phone’s sleep mode silenced them for good. Live chat didn’t solve my time-zone problem. It only worked as long as I was willing to wake up, and nobody stays willing for long. Scripted chatbots had the opposite flaw: they could only answer the questions I had predicted, worded the way I had predicted them. Real visitors never ask the predicted questions.

The fix was sitting in my own product, and I’d never used it on myself. So I finally did. The agent imported my website – all 55 pages – and my docs, and started answering visitor questions from them around the clock. When the site changes, it re-checks on its own. When a visitor shows real interest, it collects their details instead of letting the conversation end there.

I wanted one thing: questions getting answered while I slept, so I’d stop losing those visitors. That part simply worked. Conversations that used to die overnight now continue, and some of them end with contact details in my CRM instead of an abandoned tab. My lead count went up without my traffic going up.

What I didn’t plan for was the second effect. Being my own customer means I hit every rough edge before a paying user does. When the agent configuration confused me, I felt it. When it misread a page, I saw exactly which page. The bug report and the founder are the same person, so problems get fixed in days instead of months.

There’s one more thing I’d recommend to any founder, though it’s uncomfortable. My agent runs in public – anyone can open my site and ask it hard questions. Using your own product where everyone can watch means you can’t hide behind a polished demo video. Every weakness I find, a prospect could have found first. That pressure improved the product faster than any internal review.

Two things I’d tell a business leader deciding what to automate:

First, automate where waiting kills the outcome. Not everything needs an AI agent. My invoices are filed by boring rule-based automation, because the inputs are predictable and rules work fine there. Support is different: a real person is waiting, in their own words, at their own hour, and every hour of delay can cost you that person. If you can write the if-then rules in advance, you don’t need an AI agent. Save agents for conversations.

Second, use what you sell. If your company makes an automation product and your own operation doesn’t run on it, ask yourself why. Whatever you were unwilling to deal with, your customers will have to deal with instead.

The midnight questions still come in. The difference is that now somebody answers, and it knows my product better than anyone who’s awake.

Author bio: Aarohi Kulkarni, Founder & CEO, DeskClone AI – deskclone.ai