About Culture Monkey CultureMonkey is building HR tech for companies that take people seriously.

We help organisations listen to their people better through engagement surveys, lifecycle and pulse surveys, action planning, and people analytics. Over a million employees in 150+ countries trust us with something significant: the honest voice of how they feel at work.

About the Role Here is the honest version of the problem. We can pull 50,000 companies out of Apollo in about four minutes. Roughly 2,000 of them are worth a conversation. The other 48,000 are noise, and every hour our sales team spends on them is stolen from someone who actually needs us.

So this role is not about finding good accounts. It is about killing the ones that do not matter, cheaply, so everything we spend lands on the ones that do. You will live inside Clay. You will build the tables, the waterfalls, and the Claygent agents that read a company the way a good researcher would... their careers page, their Glassdoor reviews, their job postings, their new CHRO's first ninety days... and come back with a view, not just a row of fields.

Anyone can set headcount greater than 500 and call it targeting. That is a filter, and filters are a commodity. The interesting part is teaching a system to notice that a manufacturer with 900 people just posted five HR roles, has a Glassdoor rating sliding for three quarters, and hired a CHRO eleven weeks ago.

There is no API field for "this company is in a culture crisis right now." That lives in text, and reading it is what you and your agents will do.

What You Will Do Run the free checks before the paid ones. Dead domains, competitors, existing customers, suppression lists, headcount, geography. Nothing that costs a credit runs until the free work is done. Get properly good at Clay. Waterfalls, conditional runs, formula columns, HTTP columns, Claygent, and the HubSpot and Apollo integrations. You will learn fast which column costs a credit and which one is free.

Write

Claygent prompts that research, not retrieve. A prompt that returns "yes" is easy. A prompt that reads a careers page and tells you whether this company treats people as a cost or an asset, with the line it based that on, is the job.

Catch stale data before it reaches a rep. Apollo and Clay will confidently hand you someone who left four months ago. You will build the check that catches it. “Send a reason, not a score.” Every qualified account should reach a rep with a specific people-problem attached, in plain language, ready to open an email with.



“Find out if any of it is working.” We hold back a random sample of the companies we rejected and let them into outbound anyway.

Comparing them against the approved ones is the only honest test. You will help run it and report it even when it looks bad for us.

Who This Is For “Students, recent grads, or early switchers.” What you have built matters more than where you studied. A scraper, an automation for a college fest, a Clay table you made for fun... all of it counts. “You know Clay, or you will within three weeks.” Never touched it? If you have built things in Zapier, n8n, Make, Sheets scripts, or Apollo, and you learn fast, tell us. “You have extreme common sense.” The one we will not compromise on.

You should look at a spreadsheet and feel something is off. A "verified" email that does not match the person in the same row. A 900-person subsidiary sitting on a 58,000-person parent's domain. A work contact at a personal email address.

Tools will hand you confident nonsense every day, and this whole role rests on you noticing. “You use AI fluently but never blindly.” You can tell a sharp agent output from plausible filler. Fluent is not the same as correct. “Comfortable with data, not scared of a little code.” Spreadsheets, JSON, an API doc, a webhook. You do not need to be an engineer.

You do need to read the raw response instead of guessing what it said. “Follow-through.” You track your own open items and finish the boring last 10% without being chased. A Few Honest Things “We are all figuring this out.” Nobody here has run this exact playbook before, so there is no rulebook waiting for you. You will work closely with a small group across engineering, marketing, and sales who are building it as we go, and who will sit with you and be straight with you.

Fewer rules, more working it out together. “It is a startup, so it is a bit chaotic.” Priorities shift. A plan made Monday can look wrong by Thursday. You will wear several hats... some days it is Clay, some days it is cleaning a list by hand or sitting in on a sales call. “The problems are real.” This is not a sandbox project that gets thrown away.

What you build spends real money and reaches real prospects. That is the good part and the slightly scary part, and they are the same part. “Things will break.” An agent will return garbage on 200 rows.



An enrichment will burn credits on a list that should have been filtered first.

Everyone does a version of that. Just catch it early, say it out loud, and fix the design so it cannot happen twice.

What Three Months Looks Like

Roughly. It will move, because it always does. Month 1 - Understand the motion, then make one list defensible Sit with sales, learn who actually buys and why the ones that got away got away. Get fluent in Clay.

Then take one real list and rebuild it end to end: clean domains, catch the duplicates and subsidiaries, apply the free checks.

By the end: one gated list where every rejected row has a written reason next to it. Month 2 - Teach the agents to read Claygent prompts for the signals with no API field: who runs HR and how long they have been there, what the open HR roles actually say, which way the Glassdoor rating is moving, layoffs, reorgs, new CHROs. Expect to rewrite most of these several times.

By the end: Accounts that arrive with a written pain hypothesis, and an honest read on which signals were noise all along. Month 3 - Contacts, handoff, proof Enrich survivors only, never the full list. Two contacts per account minimum, verified, with a confidence floor below which we simply do not send. Push it to Apollo and HubSpot in a shape a rep trusts without double-checking. Then release the holdout and see what it says.

By the end: a handoff sales is actually using, a bounce rate you can defend, and a real answer on whether the qualification beats chance. You will walk away with something few people at your stage have: a system you built end to end, that ran on real money against real companies, and the numbers to show what it did.

You Will Love This If

You get a quiet thrill from watching 50,000 rows become 2,000 good ones. You would rather delete a column that does not work than defend it. You like the detective part... finding the one line in a careers page that says everything.

You want your work showing up as a reply in someone's inbox in two weeks, not two quarters. You would rather test an idea than wait for permission. The Kind of Person We Are Hoping To Meet Someone happier deleting 48,000 rows than adding 48,000.

Someone who reads the raw response instead of trusting the summary. Someone who sees a "verified" label and asks who verified it, and when. Someone curious about what AI agents can genuinely do, and equally clear about where they cannot be trusted yet.

Someone who believes good targeting is not about volume, it is about being able to defend every name on the list. If that sounds like you, we would love to meet you.

📌 GTM Engineer - Internship (Clay + AI Agents) (Chennai)
🏢 CultureMonkey
📍 Chennai