Is Your Salesforce Org AI-Ready?

Every Salesforce AE on the planet is currently pitching AI. Agentforce demos look incredible — an agent that qualifies leads, drafts case responses, and updates records without a human touching a keyboard. But here's what most of those demos don't show you: an AI agent built on top of a messy org just automates the mess, faster.

Before MK73 turns on any AI feature for a client, we run the same internal assessment. It takes about 30 minutes and tells you almost everything you need to know about whether you're ready. We're sharing it here.

Score yourself as you read: check the box next to each area you can honestly say is solid for your org today. Your score and next step appear at the bottom.

AI agents make decisions based on the data sitting in your org today. If your Lead Source field is 40% blank, if you have three ways of spelling "California," or if Opportunity Stage doesn't reliably reflect reality, an AI agent will confidently act on bad information. Pull a random sample of 50 records in your most AI-relevant object (usually Leads or Cases) and score them for completeness and accuracy. Below 80% clean is a stop sign, not a caution light.

Agentforce and other AI tools need a defined process to automate — a human doing the job "by feel" can't be replicated by an agent. If your best rep couldn't write down their qualification criteria in five bullet points, an AI agent can't either. Document your core process (lead qualification, case triage, whatever you're automating) in plain language before you build anything.

Someone needs to own AI outputs the way a manager owns a rep's outputs. Before go-live, define: who reviews agent actions in the first 30 days, what triggers a human handoff, and how you roll back a bad automation. Orgs that skip this step are the ones that end up with an AI agent that sent 200 duplicate emails before anyone noticed.

AI agents inherit the access of however they're configured — which means a sloppy permission set doesn't just risk a human clicking into the wrong record, it risks an agent surfacing or acting on data broadly. Audit your sharing rules and profiles before adding an AI layer on top of them.

The fastest way to kill an AI rollout is to launch it on a team that wasn't consulted. Reps who don't trust the agent will quietly work around it, and you'll never get the adoption data to justify the investment. Involve your front-line users in defining what "good" looks like before launch.

Score yourself

• 4-5 areas solid → you're ready to pilot Agentforce on a contained use case.
• 2-3 solid → fix data and process first, then revisit.
• 0-1 solid → an AI initiative right now will cost you more than it saves.