Justin R.

totally depends on the objective and scale. i first attempt to nail precision and recall, and then once consistent, optimize for cost if doing at scale. claude writes the best. 4.1 mini is a great price point for reliable, bite size missions. if you’re going to one shot a large prompt, then you need a higher end model.

Hey Caitlin O. , I’ve done something like this for several companies, it’s all customized and involves a lot of experimenting to get it right for each ICP. You first need to map out what good looks like manually, and then it’s possible to create multi layered prompts and automations to do this at scale. I personally like to use a similar approach to find the influencers and those with authority at an account. Especially under the radar people who may have a lower title, but are known to care about their job, have a track record of getting shit done, and as a result have the ear of senior titles who get way more noise from vendors. Avoid the NINA’s (no influence, no authority). 😆 This approach requires a detectives mind to identify targets and build out a dossier on them. Clay allows this to be operationalized at scale across hundreds or thousands of accounts. If you’re looking for help to do this, I’m open to jamming on ideas together and showing what’s possible.

No bounces is impossible at scale and it doesn’t really impact things if under 5%. Being reported as spam is the only thing that matters. But to your main question, validate that the person still works at the place and email on record to improve delivery and reduce bounce rates.

If you’re suggesting the company size data is in one table and you want it show in another table, then do a write to table with company name being the identifier to map.

For one of my clients I recently setup some creative clay tables that feed into outbound emails for a 3d web design software company. Currently at a 5% reply rate and of this 87% positive reply who want to meet. The client is super happy because we are indexing more on trying to meet with qualified ICP people, not indexing on just replies or meetings. How did you filter for the target list? What about them makes a compelling and relevant fit? What’s the offer? Why do you believe it’s not performing? I use AI / Clay to narrow down an ocean full of prospects to small ponds of people in a situation that is relevant to my clients offering, and then help them create a free offer that the recipient may otherwise pay for, and eliminate all friction points and hidden objections why they wouldn’t take us up on it. Will personalize if it’s useful to the offer or to do a pattern disrupt. Focus on the target list and offer, and always monitor your inbox deliverability health status.

The target list and offer > personalization. Use Clay and your understanding of you ICP to find people who are in a situation in which your offer is both relevant and compelling. Post your copy here so we can try to be more prescriptive.

Have an end goal and approach it like you’re trying to create something useful at scale. For example create an evergreen playbook for your company that identifies your ICP from a list of companies, find unique insights and signals that indicates who is in a situation that may need what you offer, feed this into a dynamic variable email where you express this developed POV and attempt to create curiosity and conversation. Do this for a few rows, and then expand so that you can see how this initial approach breaks down at scale and requires validation steps, then brainstorm how to fix it. You’ll get a hang of it after doing this for a handful of playbooks and catching the idiosyncrasies.

Yes. It’s whale hunting. Use tools and techniques to make it easier to do unscalable prospecting. Be prepared to burn through a lot of inboxes and domains. Happy to chat further if you’re looking for a thought partner to ideate and then execute.

4.1 mini is better for well structured prompts. just keep in mind even the best prompts are 90% accurate, so you need to build validation and re-try logic.

The clay template/recipe for job change tracking is quite inaccurate and unreliable at scale. It will indicate changes if the person has things in their Linkedin profile like volunteer work, advisory or consulting work in addition to the current job/company on record. It also gets confused by M&A or subsidiaries. I custom built an entire job title and company change workflow within clay that is optimized for both recall (finding their correct linkedin profile) and precision (only indicating a change if it actually happens). There are also validation steps I built-in to confirm everything is operating as expected. Even the best prompts and LLMs are accurate at best 90% of the time, which is why I built this custom approach. The only downside is it uses a little more API calls (or clay credits). Let me know if you’d like this built and/or managed for you.