Soumyadeb Mitra

Today we're launching Lookout. With Lookout, a marketer starts from the campaign goal and gets the audience live in Meta, Google and Braze in minutes, built on the definitions their data team already maintains. We've been running Lookout with design partners for the last few months. Meghan Hilton, who leads RevOps and lifecycle marketing at PrizePicks, told us: "Acting on a new campaign idea used to take weeks. We'd make a request with our analytics team, wait for them to do analysis, add custom attributes if needed, and sync the data to Braze. Lookout eliminates the back and forth. Now my team can launch campaigns faster than ever." Thank you to our design partners for all the feedback that shaped it.

90% of the cold emails I still get are Hello <first name>, congrats on your raise <from this-time ago>. Even the warm emails from products I have signed up for look similar - <Welcome. Here is a irrelevant case-study> That was because it was still some human building that campaign/journey. Lot of scope for innovation here.

When buyers can ask a company's agent questions inside ChatGPT, a lot of research that used to show up as visits to your site moves off it. Some of it was already happening with agents visiting the website but this takes it to a very different level. The entire customer-data stack (from Analytics to Engagement) needs to be rebuilt for this new world.

For many of the marketing teams we talk to, a campaign idea still means a ticket to the data team and weeks of waiting for the audience. This session is about that wait, and how Meghan Hilton's team at @PrizePicks now builds and syncs audiences themselves. Dom Benton from our product team joins @humansofmartech to share what we learned building Lookout.

So true. The best teams are deploying proper tracking, rest are just burning VC dollars

Try it in a sandbox, no setup required: lookout.rudderstack.com/w/lo… Read the launch blog: rudderstack.com/blog/introdu…

A pattern from GTM teams we've talked to lately: the ones building their own agents get a demo working and then stall. Getting customer context to the agent reliably, running the agent with guardrails and approvals, measuring whether it worked and learning from it to improve the agent are some of the things that take a lot of time to get right!!

The next evolution of the GTM harness is something which can take actions mostly autonomously. The evolution from Claude to Instinct.

Useful list, but I would move stage 5 to the top. In B2B, cold email without signals isn't the strategy that works very well - from our experience. Use ads and intent signals to find the accounts that are somewhat warm and only then bring in email, LinkedIn and SDRs.

Building got easier because engineering went AI-native first. Marketing got harder for a related reason: everyone can now ship more content and email than buyers can read. The edge is deciding who to talk to, when, and what to say, and most GTM teams still do that by hand.

Customer data is the biggest pile of dark data most companies have. It lands in the warehouse and often just sits there (except some basic analytics), because acting on it is often blocked on data and data-science teams Cheaper intelligence like Jev changes that - enabling business teams like marketing to build models they want. Excited to be building that at @RudderStack

A data lead at one of our customers told us, "I've never seen a clean warehouse." In the same conversation, they said you can get it good enough to work with. We think that's right, and it shaped how we built AI for marketing teams. Every warehouse has some mess in it. One of our customers had a different lifetime value for the same customer in each tool they sent data to. Nobody did anything wrong. Each team built its own version, and over time they drifted. So the "best customers" behind a retargeting campaign in Meta and the ones getting an email from Braze weren't always the same people. This is the worry we hear from data teams when AI for marketing comes up. Point an agent at raw tables and it will happily come up with its own idea of an "active customer", and now there's one more number to argue about. Another data lead told us they don't want everyone building whatever audience they want. We understand why. Waiting for a perfectly clean warehouse doesn't work either, because campaigns can't be put on hold that long. So we started from what the data team already maintains: dbt models, LookML, semantic layers in Snowflake and Databricks, column descriptions. The data team keeps owning those definitions, and marketers ask questions and build campaigns on top of them. When both teams say "high-value customers", they mean the same people, in Meta, in Google and in Braze. That's the idea behind Lookout, we're launching it tomorrow.

Interesting, this split still matters when agents are running inbound and outbound. We have separate agent for running outbound and separate one for following up on inbound. The prompts, channels are very different #AIGTM

GTM needs a claudecode like product

RudderStack moves more than 300 billion customer events a month into our customers' warehouses. Lately we've been asking those customers what happens after the data lands, and how long it takes their marketing team to turn an idea into something live, like a retargeting campaign in Meta or Google, or an email or journey in Braze. The answer is usually weeks, so we started asking where the time goes. It starts with a marketer who has a campaign goal and a hunch about who to target. The data to check that hunch sits in the warehouse, and they can't explore it, so they write a brief and file a ticket or send a Slack. Someone on the data or marketing ops team picks it up, digs through the warehouse and comes back with a few options. Often the first cut is too broad, too small or just off, so it goes back and forth for a few rounds. Then the list gets exported, usually a spreadsheet of names and emails sent over email or a shared drive. Every ad platform wants that file in a different format, so someone reformats it once per tool and uploads it to Meta, then Google, then Braze, waiting for each to process. Only then can the team launch the retargeting campaign, send the email or push people into a journey. By the time the campaign goes live, weeks have passed, and the list is a snapshot from the day it was pulled. One customer put it as two weeks in the best case, sometimes a month or longer. We also hear that marketers stop asking for more interesting audiences, because each one means another ticket. Engineering went AI-native this year, and this part of GTM still runs on tickets and CSVs. What would marketing teams do differently if a campaign idea could go live in minutes instead of weeks?

Outbound calls are about to get as cheap as email, and email shows where that goes: buyers get flooded and stop answering. The hard part is deciding who should get a call, why now and what to say.

This works only when the loop closes fast: ad, click, purchase, ROAS. In B2B, the path from filling up a form to a deal closing can take months. An ad agent there needs the follow-up and the outcome data wired in, or it ends up optimizing for clicks.