Jack

20 years in Sales Ops and a lifetime of coding. I build free tools and share practical ways to improve sales operations with data, automation, and AI.

[Jack @salesoptecs]

Made a free territory builder. draw sales territories right on a map, balance them by revenue, accounts, drive-time, or demographics, then hand them to a rep or load them straight back into salesforce. Would love some feedback! 👇

https://t.co/ZNuBcFx2Bi

[Jack @salesoptecs]

XGBoost and neural networks for predicting whether Salesforce opportunities will close, using CRM fields, call transcripts and sales emails. It also covers combining both models in one prediction.

[Jack @salesoptecs]

HubSpot Breeze as an agent router makes me ask which agents can change CRM data. See how to scope each agent's access so using one assistant doesn't mean giving every agent write permissions.
#HubSpot #RevOps
https://t.co/KAMCYgk0Dq

[Jack @salesoptecs]

Salesforce Account syncs with an ERP can overwrite newer edits or leave records locked. Version checks, event IDs to prevent duplicate updates, and clear field ownership help avoid those problems.

[Jack @salesoptecs]

HubSpot AI updates worry me when a confident overwrite can replace a correct value. See how to set source authority and retain audit history so pipeline changes stay traceable.
#HubSpot #RevOps
https://t.co/Yqz0HT2GqG

[Jack @salesoptecs]

Whatever happened to OpenClaw?
For a minute there, it felt like that was the only thing people were talking about.

[Jack @salesoptecs]

@PatSchaber Thanks for the read. Short video on this one coming out soon.

[Jack @salesoptecs]

Sales Ops territory exceptions need investigation before reassignment. See how I built an agent to check CRM issues and report daily, leaving ownership decisions to a human.
#RevOps #AIAgents
https://t.co/aaTzDThaBs

[Jack @salesoptecs]

Adding more compute can slow a Spark data pipeline through extra data shuffling and coordination. Uneven customer transaction counts can leave workers waiting, so fixing data skew and filtering earlier may help more than a bigger cluster.

[Jack @salesoptecs]

@mikekaufmann000 @kushalbyatnal Owner and a review date on every page, and the page displays its own staleness rather than waiting for an audit. Then never edit an old page into a new truth, publish the new one and point forward from the old.

[Jack @salesoptecs]

Sales Ops AI agents need clear job ownership, whether one agent handles many systems or several each own a task.
See how specialist agents and one broad agent offer two ways to organize work beyond chat.

[Jack @salesoptecs]

Excel sales calculations become hard to trust when manual steps make the logic difficult to explain or repeat.
See how moving recurring calculations into maintained code with checks makes results easier to verify.

[Jack @salesoptecs]

@ZMacromaniacs @deanwball Is a backhanded compliment?

[Jack @salesoptecs]

Persistent AI agents monitoring pipeline need an owner when they get stuck. See how I scope ongoing CRM work and exception handling so unresolved issues reach a person.
#RevOps #AIAgents
https://t.co/U9kbalq5Km

[Jack @salesoptecs]

Sleep is for the weak

[Jack @salesoptecs]

@KomalWCode CPU under 30 percent means it is waiting, not computing. At peak that is almost always lock contention or a connection pool with fewer slots than you have concurrent requests. Look at what the sessions are blocked on before you touch anything about the query itself.

[Jack @salesoptecs]

@LeakerApple Right, and granular is the actual fix. All or nothing permissions force everyone to grant everything, which means the dialog stops carrying any information. The useful question was never whether access got approved, it is which files were read and when.

[Jack @salesoptecs]

The constraint worth covering here is that an agent working across Sales, Service and Marketing writes to objects three teams each believe they own. Does field level security and the sharing model evaluate against the agent or against the running user? That answer decides whether any of this is safe to turn on.

[Jack @salesoptecs]

@infinterenders It cannot own the failure is the whole thing. Writing the code was never the expensive part. The expensive part is that a person has to be accountable when the field it writes to turns out to mean something different than everyone assumed it meant.

[Jack @salesoptecs]

Good list. The one I would move near the top is what makes two entries the same entry. Retries and webhook replays are the normal case rather than the edge case, so every movement needs a caller supplied key the ledger rejects on the second attempt. Append only does not save you if you append twice.

[Jack @salesoptecs]

Lost update. The second write lands on a value it never read, so the first one disappears with no error anywhere. Read committed does not stop it. You either take the row lock at read time with select for update, or you keep a version column and check it on write so the stale one fails loudly instead of winning.