Hey! 👋 

A quick note that I feel strongly about…

Some folks are saying that marketing ops has zero chance of being replaced by AI. I get the sentiment — I wish it were 100% true. But I also feel like saying such a thing would be misleading to you all. So here’s my honest take:

1. Whether AI can replace marketing ops professionals or not, executives and VCs are really pushing the message that it can. For that reason, we are seeing tons of push to consolidate ops under RevOps and to lay off folks. Even if we feel confident that we can’t be replaced, I am sad to see that many companies seem to be “trying it out” to see how it pans out, rather than heeding any warning.

2. I think campaign ops will be the hardest hit, due to the executional nature of the work. I KNOW, I love campaign ops and respect the experts in the space so much…but the sad truth is that many companies are allured by the idea of automating everything possible, including content. So if I was in campaign ops right now, I’d take a hard look at how I can leverage AI to run campaigns more quickly or elevate myself to a strategic partner space.

3. If you are doing ticket-taking work, you are at risk. Yes, much of our job is taking tickets one way or another, whether those are literal tickets or verbal requests. BUT we need to be strategic about which tickets we prioritize and which we push back on. If we’re only sending emails because we are overwhelmed with manual email work, see if you can get AI tooling to help…and again, try to elevate your role to be more proactive, not reactive ticket-taking.

Why do I tell you this? Not to be depressing (I promise). I think that our entire department is about to be upleveled. BUT I think that there will be pressure, so I want you to be prepared. This edition of the newsletter is one of the ways I want to arm you with tooling and prepare you for the upcoming “compression.”

Ok, back to the main topic…

Most MOPs professionals I talk to right now are somehow both exhausted by AI spam online and still unsure if they're using AI tools correctly.

Last month I built a report for the third Monday in a row. Halfway through, I caught myself doing the thing I'd tell anyone on my team not to do -- performing a task by hand that I'd already done enough times to hand off. So I stopped and built a Skill for it. It did take about an hour to perfect, but now I can save hours of production from here on out.

Once I stopped trying to "keep up with AI" and started treating it like any other ops prioritization problem, the overwhelm went away.

Here are some best practices:

Score before you build. The first move was getting honest about where my time was going, then scoring potential AI workflows the same way I'd score any ops project. How often does the task happen? Does it touch revenue? How hard is it to build? Most of the AI ideas I was excited about scored badly once I was honest about the build effort. The boring ones I'd overlooked scored highest. That's the whole game IMO. Not "what's the coolest thing AI can do," but "what's the highest-return thing AI can do for me?”

Start with one, NOT ten. I KNOW the temptation is to rebuild your whole stack the second you get excited. Resist it. One workflow running and trusted beats five in progress, every time.

Protect the build time. Same uncomfortable truth as always. This work doesn't happen in the cracks between tickets. You have to carve out the time, and if that makes you nervous, have the conversation with your manager. Frame it as what it is: a few hours now to save yourself days every quarter for the rest of the year.

So where do you start?

That's what the guide helps you decide. It's the scoring matrix I run before I build anything, turned into something you can fill in for your own stack in about ten minutes.

After a quick break, I’ll teach you some of my time-savers, including a guide you can use at work… 📺

This section is sponsored by Knak, a partner who supports this newsletter. 💌 

🤖 How to let Claude draft your launch email templates without breaking brand

Most MOPs teams are still in the "are we allowed to use ChatGPT for subject lines?" conversation. Others are wiring AI agents directly into the production layer, so brand guardrails enforce themselves at the asset level. Here's the play I've been recommending to enterprise MOPs leads since Knak shipped its MCP server in April:

1. Install Knak's MCP server in your AI tool of choice. Add the Knak connector with your enterprise auth and confirm it can see the template library you want agents pulling from.

2. Pick one campaign type to pilot. Internal comms or routine product update emails work best for the first run -- low judgment and high volume. Skip flagship campaigns until the team is comfortable.

3. Brief the AI like you'd brief a contractor. Give it the campaign goal, the audience, the on-brand Knak templates to pull from, and the launch date. The MCP server returns a draft email pulling from your real templates with brand guardrails enforced. You review, edit, approve, ship.

This is how OpenAI and Meta are already running it -- Knak confirmed both at the launch!

🔗 See Knak's MCP server in action -- book a demo.

…back to our regularly scheduled programming. 📺

The 6 workflows currently saving my week

Because I know you want to see the actual stack, here are the five automations I use every week at Vector, so they can help you brainstorm.

1. Wispr / Fathom to Claude to Notion. Wispr Flow catches my walking brain dumps, Fathom catches what's said on calls, and both pipe into Claude, which turns the raw transcript into structured notes or a first draft and drops it into the right Notion documentation page. This is the one I'd tell you to copy first -- easy to build, used daily, and it helps me keep up with documentation.

2. Claude Skills for repeatable tasks. Skills are reusable, named procedures Claude follows the same way every time, so I never re-explain the context or output format for anything I do more than twice. I’ve set up a Pipeline Forecast Skill, a HubSpot Data skill, and skills to give me an up-to-date status report on 3 of my top clients: Marketing, Sales, and Solutions.

3. Clay Functions + Claude. Calling Clay directly inside a Claude chat is 🔥. When we get a shoddy event list, I can mostly automate enrichment, qualification, and routing because I just have Claude pull in my pre-built Clay Function (which is basically just a Clay table of enrichment flows).

4. Meeting prep. I let Claude prep me for calls with the latest Slack and meeting context — so the constant context switching is not so severe. It’ll search Slack for updates, pull in the last meeting summary, and anything else I feed it as context.

5. The RevOps bot (Claude Code + Notion + Slack). This is my most ambitious build, as it involves some engineering work. But I’m hyped on the ROI here — basically, it’s a Claude bot that has all of my documentation + read-only access to HubSpot, so it can help answer basic Q&A questions for users.

6. Claude HubSpot analyzer. Claude as a thinking partner against exported HubSpot data -- audit reports, segmentation logic, automation review, naming convention cleanup. The real value is asking it to help you decide what work to do, then doing it yourself with better thinking behind the wheel.

What I'd tell you if we were getting coffee

Don't get overwhelmed trying to do everything at once. Pick the one workflow where you feel you’re wasting the most time. Never half-a** two things, whole-a** one thing.

(cue infomercial voice) but wait, there's more! ↓

I’ve created this guide you can steal and use at work to help you decide where to start! Check it out here. ✍️ 

What I’m up to/what I’m studying 💭

Lately, I’m working on getting our new sales leader set up and running. Our offerings are getting more complex, which means a little bit of a rework of the new business, renewals, and expansion processes.

I also had a new team member join, Jesse Adelberg! He’s a rockstar, diving right in to get us set up with tools like Quo for texting tracking in HubSpot and FirstTouch for social marketing tracking and playbook activation.

We’ve been working on building an MCP to enable persona-based AI + HubSpot permissions for our users. This enables us to allow certain roles to write to department-owned fields, while avoiding accidental overwrites of critical fields! It’s a bit wild that we all seem to need to build our own custom MCPs, but alas… 🙄 😒

On one hand, AI + automation can be an almost immediate value-add in terms of saving teams time. On the other hand, it does incur more build and maintenance time and I feel strongly that I don’t want to “bet the business” on AI too much. CRM/MAP/DW should be the spine, whereas AI should be an add-on that can help us move faster, but that we could live without. This way, if AI surges in price (as we suspect it will, once VCs stop subsidizing it…), we are not stuck with a $1,000,000 CRM bill instead of a $50,000 one.

Perhaps a topic for an upcoming newsletter — would love any thoughts you have or notes from the field!

P.S. If you want to follow along with our wider GTM strategy and tactics, follow the Vector blog! You can subscribe to our newsletter too! 👻

This section is sponsored by Zapier, a partner who supports this newsletter. ⚡

🧠 How to turn call transcripts into product intelligence in HubSpot

Most teams know what their prospects said on a call. Almost none of them have that information living anywhere useful in their CRM, which can be a huge miss when it comes to targeting and nurtures. Here's how we fixed that at Vector using Zapier + AI.

1. Create a custom Product Interest field on the Company object in HubSpot. Open text or picklist both work.

2. Set up a Zap that sends new Fathom call transcripts to a Google Sheet. Set a 490,000 character limit on the transcript or the row will break. Zapier Tables works here too if you prefer.

3. Set up a second Zap that triggers on a deal stage update. Have it find the associated company in HubSpot, look up the transcripts from the sheet by domain, send them to ChatGPT with context on your product and features, and write the analysis back to the Product Interest field.

Now your GTM team can see which features a prospect actually cares about without listening to a single recording -- and you don't have to share your full CRM with the AI model to make it work.

🔗 Try it yourself with a free 14-day Zapier trial.

Dear Sara ✍️

New to marketing operations? On a team of one at your company? Shy/introverted? Wish you could ask a question to an experienced marketing operations professional, without them knowing who you are? Here’s your chance! Submit an anonymous question to me here and I’ll answer a new question in every issue.

Here’s my answer to a question from last week:

Oh man, what a complex question! First, we have to think about the data model. How do you want to track job changes in your CRM? The tools can adjust to whatever data model you want to create; the trickiest part is figuring out what you want that data model to be.

Clay will overwrite, append, or trigger a create depending on how you wire the HubSpot/Salesforce action. That means the hygiene decision is entirely yours to design, and you should design it deliberately rather than accepting the default "update record" action because it's the easiest button.

The typically recommended pattern for a job change:

Keep the old contact record intact. Don't change its company association or its historical activity. Instead, mark it — a "Left Company" or "Moved On" status, a checkbox, or a lifecycle flag. In Salesforce many teams flip the old contact to a "no longer at company" picklist value and optionally null out the direct phone/keep the old email. This preserves attribution and the account's historical relationship.

Create a new contact for the person at the new company, associated with the new account. This becomes a fresh prospect record at the new org, with its own lifecycle stage (usually reset, because they're a net-new opportunity at a net-new account — though warm, given the prior relationship).

Link the two if your system allows it. This is the part many teams skip and later regret. You want to preserve the thread that says "this new prospect is the same human who was a champion at their last company." In HubSpot you can do this with a custom property (ex: a "Previous Contact ID" or a "Known Relationship" field) or by using the association model. In Salesforce, Contact-to-Contact relationships or a custom lookup field accomplishes the same. That link is gold for sales — "you worked with us at Acme, now you're at Beta" is one of the highest-converting plays in B2B. You can use this for Champion tracking, for example!

On duplication specifically…the duplication risk isn't usually from creating the new-company record — that's intentional and correct. It comes from two other places. First, Clay re-running enrichment and creating a second copy of a record that already exists because your dedupe key is weak. Always dedupe on a stable identifier — in HubSpot that's the email or the record ID, in Salesforce ideally a managed dedupe rule — before any create action fires. Second, the same person showing up under two emails (old work email, new work email) with nothing tying them together, which reads as two strangers. The "link the records" step above is what prevents that from becoming invisible duplication.

One more practical guardrail: gate the write behind review for anything high-value. For your general prospect base, automated create-new-and-flag-old is fine to run hands-off. But for current customers, open opportunities, or named accounts, I'd have Clay route the detected change to a Slack channel or a review table for a human to confirm before it writes to the CRM. Job-change signals have a real false-positive rate (LinkedIn lag, ambiguous titles, people listed at two orgs during a transition), and you don't want an automated overwrite touching an open deal based on a stale signal. 🫵😅

One Salesforce-vs-HubSpot wrinkle. In Salesforce, the Contact is locked to one Account by default, which actually nudges you toward the correct behavior — you naturally create a new Contact under the new Account. In HubSpot, a Contact can be associated with multiple companies, which is more flexible but makes it easier to accidentally just re-point the association and lose history. So in HubSpot you have to be more disciplined about the "flag old, create new" pattern because the system won't force it on you.

Also, when you build lists and workflows, watch the company association filter. By default HubSpot includes a contact if any of their associated companies match — so a contact can get pulled in based on a previous employer you've already flagged. Switch the filter from 'Any company' to 'Primary company' so your automation only targets people whose current primary company actually matches.

News of the week 🗞️

Salesforce went headless. Salesforce now exposes every capability on its platform as an API, MCP tool, or CLI command, so AI agents can operate the whole system without a human ever opening a browser. Co-founder Parker Harris framed it from the stage roughly as: once agents can handle the work directly, why would you choose to do it through a browser at all?

Why you should care: this makes me wonder how long the UI remains the center of gravity for the rest of our stack. If the biggest CRM in the world is betting that agents, not humans, are the primary users, it's worth asking which of your other tools go headless next, and what that means for a job that's been built around knowing the interface. The skill stops being "I know where every setting lives in HubSpot" and starts being "I know what to ask the agent to do, and how to check it didn't do something dumb."

Vibe coding is hollowing out the edges of the stack. Teams are rebuilding point tools with AI instead of buying them. One agency reportedly replaced 80% of its software subscriptions with vibe-coded alternatives, and roughly 63% of vibe coding users are now non-developers.

Why you should care: the tools most exposed are the thin ones -- form builders, dashboards, schedulers. HubSpot and Salesforce still own the orchestration layer, so this isn't "rip out your CRM." But "we've always paid for this" is no longer reason enough to renew the small stuff. Worth an audit before your next cycle.

74% of enterprises have rolled back a live AI customer agent. Nearly three-quarters of companies that put an AI customer communications agent into production later pulled it after a governance failure. The twist: teams with the most mature guardrails rolled back at an even higher rate, 81%.

Why you should care: the bottleneck has moved from getting an agent live to keeping it live. The higher rollback rate among mature teams doesn't signal worse performance -- it reflects better monitoring catching problems sooner. The takeaway for you is that governance alone won't save you. Knowing when to pull an agent, and having the instrumentation to see it coming, is the actual skill. 👀

HubSpot is rebuilding its Salesforce sync engine. The new "v2" engine adds de-duplication on any field (not just email), owner field sync, unique ID support, and fewer errors. Company sync is live now; deals come next; contacts aren't expected until a Q4 beta.

Why you should care: if you run both systems, I don’t need to explain this to you lol. This is a huge step in the right direction.

What else have you heard about recently? Reply back to this email to send me any other word on the street. 👂

Interesting tech of the week ⚙️

Luster.ai: AI roleplaying, enablement, coaching for sales.

My VP of Sales loves this tool for enabling sales, especially when it comes to pitch feedback and sales call training! If you don’t want to have to build a custom AI tool to do this, Luster is trained specifically in this use case.

This section is sponsored by Default, a partner who supports this newsletter. 🧑‍💻

🤔 Why your AI agents keep acting on the wrong data

Your agents aren't underperforming because the models are bad….they're acting on a data layer that was broken long before you deployed them. You have four versions of the same account spread across HubSpot, your warehouse, and last quarter's enrichment tool, and the Agent isn’t sure which is the right one.

Most vendors bolted agent features onto that shaky foundation. Default rebuilt from scratch instead -- identity resolution handled upstream, deterministic workflows, every action logged and reversible, MCP access on every tool. The practical version: a GTM engineer in Claude Code can wire up routing, scheduling, and enrichment in one session and ship a real agent, not a demo. Full reveal drops May 27.

🔗 See what Default built.

Meme of the week ⚙️

If you have a moment, will you let me know if you enjoyed this edition of the newsletter? Thank you!

Thanks for reading,

❤️ Sara

P.S. Special thank you to our sponsors for supporting the creation of this newsletter. 🫶