5 Clay Releases Every GTM Leader Should Care About and How to Use Each
A GTM leader's field guide to what changed and how to run each one this week.
Clay committed to shipping a major product release every week this summer, and they’ve held to it. But keeping up with the latest is like a second job… that you don’t have time for.
Likely it’s part of my job, and I wanted to share 5 of the most recent releases that I think every GTM leader should be paying attention to. I’ll cover what the feature is, why you should care, and how to implement it this week.
Account Research Agents Turn Your Closed-Lost List Back Into Pipeline
This is the one I’d try first, and it’s the newest (open beta as of July 22). An Account Research Agent reasons over your GTM data (Gong calls, CRM records, emails, and signals unified in Clay’s new Audiences layer) and keeps a segment’s intelligence current on its own. Point it at your closed-lost list, your renewal book, or your expansion accounts, and it surfaces who’s ready for a re-engagement, an upgrade, or an outbound play, without an analyst chasing reps to update fields.
Arguably, a closed-lost is the highest-intent list you own. I ran a closed-lost program for a client last year, built by hand, and it took forever to get right. With this feature, an agent that re-reads the segment every week removes the exact bottleneck that killed that motion.
To put it to work:
Create an Account Research Agent and point it at one segment. Start with closed-lost, not your whole market.
Feed it the loss context you already have (CRM closed-lost reasons, the last Gong call, the final email thread).
Define the signals that flip an account back to live: new funding, a champion who changed jobs, a competitor renewal window, a leadership change.
Have it write a one-line “why now” per account and a recommended play.
Route what it flags into a sequence or a rep’s queue.
Re-run it weekly and watch what resurfaces.
The old loss reason is the highest-value input you have and the one people forget to feed it. An account you lost on price 14 months ago is a completely different play than one you lost to a competitor whose contract is now up for renewal. Start with one segment and one signal before you let it reason over everything, and confirm your plan has access first (open beta is live on Enterprise, Growth, and Launch) before you promise sales a resurrection engine.
Inbox-as-a-Source (AgentMail) Lets You Reply to Every Inbound Lead Automatically in Minutes
Email is now a column in Clay. Connect an email inbox straight to a Clay table and inbound leads get enriched, qualified, and replied to by Claygent, right from the table. For a team that can’t staff a dedicated inbound SDR, this is the closest thing to hiring one that also happens to work 24/7. Speed-to-lead is STILL as important as ever, and this reduces the time between form submitted and first reply down to about the time it takes to enrich a row.
To put it to work:
Connect the inbox that receives inbound (demo requests, contact-us, content replies) to a Clay table.
Add your enrichment waterfall so every inbound row fills in company, size, and role.
Put ICP scoring right after enrichment.
Branch the routing. High-fit goes to a rep or a fast reply, mid-fit to a sequence, junk dies before it costs you anything.
Let Claygent draft the reply, and send your highest-value accounts to a human-in-the-loop process.
Track reply time before and after you turn it on.
Do not auto-reply to everything on day one. Put your top-ACV band behind human approval so a bot never sends the first word to a six-figure logo. The scoring part matters more than the reply itself, because a fast, personal answer to the wrong lead is still wasted motion. And measure one number to prove the whole thing works: median time-to-first-touch.
Clay Functions Package Your Best Builds, Clay for Reps Ships It to the Whole Team
This is the one I’m most fired up about, because it’s what I’ve been waiting for and I’m sure a lot of other RevOps teams have as well. Two releases combine here. First, functions let you package a whole enrichment-and-logic chain into a single reusable column. The other part is that you change the function once and it updates everywhere you use it. Clay for Reps puts those Functions inside the tools your reps already live in (ChatGPT, Claude, Microsoft 365 Copilot), with per-rep permissions and credit budgets sitting on top.
Together, these fundamentally change how compound teams use Clay. Ops doesn’t have to be the team that builds tables on request. Just like a good ops function, they are now scaling systems and we can finally scale Clay for sales reps. Ops creates a Function once (e.g., an account brief, a persona lookup, an outbound draft, an ICP score), set who can use it and how many credits it can spend, and every rep calls it by name from their chat window.
I built a few Functions last week to see how real this was, and I absolutely love that governance lies inside the Function itself. The compliance and the credit rule are created once by whoever owns the system. The rep can’t skip or overrule the rule because it was never a step they could see.
To put it to work:
Inventory the logic your team has rebuilt more than twice (ICP score, account brief, competitor snapshot). That list is your first product roadmap.
Build each one as a Function, with your definition encoded as explicit weights.
Set permissions and a credit budget per Function, per rep.
Publish them into your reps’ chat tools so they call them by name instead of filing a request.
Version and update centrally. Change the logic once, and every rep inherits the fix.
And I would treat this like product management. Pick an owner, ship a small library (five Functions people trust beats fifty nobody opens), and write a one-line “what this is for” on each. Put deterministic rules in code columns (not AI columns) so they cost nothing per row and never drift, unlike a lot of AI. And use the credit budget as a governance tool, not just a cost cap. It’s how you stop one rep’s runaway prompt from torching the team’s month.
Audiences Lets You Score Your Whole Market
Clay’s old ceiling was 50,000 records at a time, but Audiences removes it. You get unlimited people and company searches, millions of rows imported from your CRM and data warehouse, and enrichment applied in place. Alongside it came a redesigned Company Lookalikes source (cluster from up to 15,000 inputs) and a brand-new People Lookalikes enrichment for finding net-new contacts. This means you can score your entire addressable market instead of scoring a sample and calling the sample a TAM.
To put it to work:
Import your full CRM and warehouse into an Audience, not a 50k slice of it.
Cluster lookalikes off your closed-won list so “good fit” is defined by real data.
Apply your ICP score across the whole set.
Tier it, then route only Tier 1 into active plays.
Refresh as new records land.
You can build another model on closed-won. What you actually close beats what you wrote down in your ICP docs at the last planning offsite. So go ahead and score the whole market (but watch your spend!). Or just enrich the top tier deeply and leave the long tail cheap until it hits scoring thresholds. And re-baseline every quarter, because your best-fit cluster changes as the product and the buyer change, and these days, it feels like it’s monthly!
Cost Engineering Give You Obervability and Lets You Run Clay for a Fraction of the Credits
You’re building clay a lot, you also are always paying attention to your Clay unit economics. If you’re the one who owns the budget, this is actually a big opportunity for you. Three releases stack up here. March’s pricing overhaul cut data marketplace costs by 50% to 90% and split credits into a clearer system. Open-weight models (Kimi K2.6 and GLM 5.2) now run in Claygent for cheap long-running research. Sandbox Mode lets you validate an AI column before it burns a single credit, and observability shows you credits-per-row and fill rates in production.
To put it to work:
Turn on Function Observability and find your most expensive columns by credits-per-row.
Move high-volume, low-stakes research steps from a frontier model to an open-weight one.
Sandbox-test every large run before you deploy it, so a broken prompt doesn’t bill you 5,000 times.
Read the fill rates. A column that only returns data 40% of the time is a provider problem, not your problem.
Publish your own before-and-after math to the team so the savings actually stick.
Keep the frontier model where taste and judgment matter (final outbound copy, compliance-sensitive research) and send the grunt work to open-weight. Sandbox Mode is free insurance, and the one run you skip it on is the run that costs you a month of credits. And put deterministic scoring in code columns, which cost nothing per row and give you the same answer every time.
You Don’t Need All Five – Pick One and Ship It This Week
Yes, I get it, you’re busy and you can’t run all 5 of these this quarter. So, my advice is to just pick the one that helps clear your worst bottleneck and get it working before you touch the next. For most teams I talk to, that’s either Account Research Agents for the closed-lost list, or the Functions library if your ops person spends every week rebuilding the same scorer. Start there, prove it on ten accounts you know cold, then earn your way to the second one.
The shipping isn’t slowing down either. Kareem walked through what’s coming on last week’s roadmap stream (Sequencer 2.0, topic intent signals, account execution agents), so whatever you build now, build it so you can swap a piece out later without starting from scratch.
If you want to try any of it, you can start on Clay here. And if a GTM leader you know is drowning in the same changelog, forward this to them.