Latest / MBA Training AI / Connecting AI tools to your existing business stack with connectors

Transcript

Connecting AI tools to your existing business stack with connectors · MBA Training AI

• 0:00
  Welcome to the Leaders Insights podcast.
• 0:07
  Today's episode connecting AI tools to your existing business stack with connectors.
• 0:12
  Here's a number that should embarrass most of us.
• 0:14
  Professionals waste hours a week copying text out of chat GPT and pasting it into email,
• 0:20
  spreadsheets, project boards.
• 0:22
  Why is that still happening in 2026?
• 0:24
  Because the demo economy trained us badly.
• 0:27
  The vendors sell you a chat window, you type, it answers, everyone claps, but a chat window
• 0:33
  isn't connected to anything.
• 0:35
  So you become the connector, the meat and fingers integration between the AI and your actual
• 0:40
  work.
• 0:41
  The article calls that turning AI into a toy.
• 0:43
  Harsh.
• 0:44
  Fair?
• 0:45
  Completely fair.
• 0:46
  If the output has to be manually retyped to matter, it's a party trick.
• 0:50
  The fix is connectors, small bridges that let the AI read from and write to the software
• 0:55
  you already run, like Slack, Google Drive, HubSpot, your ticketing system.
• 1:00
  Define connector properly because people hear that word and glaze over.
• 1:04
  A connector is a permissioned pipe.
• 1:07
  It gives the model access to one specific system, say your customer records, so it
• 1:12
  can pull the real data and push results back without you touching the clipboard.
• 1:17
  OpenAI shipped this as connectors inside chat GPT.
• 1:21
  Anthropic pushed the standard underneath it, called MCP, Model Context Protocol, basically
• 1:27
  a universal plug so any tool can talk to any model.
• 1:30
  Anthropic released MCP and now OpenAI and Google both adopted it.
• 1:35
  When a competitor's standard gets adopted by everyone, I get suspicious.
• 1:39
  Who wins?
• 1:40
  You do, actually, for once.
• 1:44
  Before MCP, every integration was custom plumbing.
• 1:48
  We built a bespoke bridge for each app, each model, and it broke on every update.
• 1:53
  One shared protocol means a connector built once works across chat GPT, Claude, and Gemini.
• 2:00
  Anthropic gave away the standard because owning the standard beats owning one product.
• 2:05
  Fine, but walk me through it for someone who runs a normal business, not a data team.
• 2:10
  Start narrow.
• 2:11
  Pick one workflow that bleeds time.
• 2:14
  Say your sales team writes follow-up emails after every call.
• 2:17
  Step one, connect the AI to your CRM, who are the customer database, HubSpot, or Salesforce
• 2:24
  so it can see who the call was with and what was discussed.
• 2:28
  And that's the scary part, right?
• 2:30
  You're handing your customer list to a chatbot.
• 2:33
  That's the part you should be scared of, and most people aren't scared enough.
• 2:37
  Connectors run on scopes.
• 2:39
  Permissions that say, read these contacts, don't touch billing.
• 2:43
  Set them tight.
• 2:44
  Read only where you can.
• 2:46
  And check whether the vendor trains on your connected data.
• 2:49
  OpenAI and Anthropic both say enterprise data stays out of training, but that's their
• 2:54
  claim, so get it in the contract, not the blog post.
• 2:57
  Okay, CRM connected, then what?
• 3:00
  Then connect the destination.
• 3:02
  The email draft shouldn't appear in the chat.
• 3:04
  It should land in Gmail as a draft, ready to send.
• 3:07
  So you add a second connector to your mail.
• 3:10
  Now the loop is closed.
• 3:12
  The AI reads the call notes, writes the follow-up, drops it where the rep already works.
• 3:18
  Nobody copies anything.
• 3:19
  How much time does that actually save?
• 3:22
  Give me a real figure, not a vendor fantasy.
• 3:24
  Honest answer.
• 3:26
  The loud numbers come from the vendors, so discount them.
• 3:29
  Anthropic and Google both quote big productivity gains that are basically marketing.
• 3:35
  From teams I've watched, a tight connector on a repetitive task saves maybe 20 to 40
• 3:40
  minutes a day per person.
• 3:41
  Unsexy, but it compounds.
• 3:44
  And it's real because the human stopped being the copy-paste robot.
• 3:48
  What's the trap people fall into once they get connectors working?
• 3:52
  They overconnect.
• 3:53
  They wire the AI to 14 systems and give it right access everywhere, then act shocked
• 3:59
  when it edits the wrong record.
• 4:01
  Connect one system, read only, prove it, then expand.
• 4:06
  Treat every right permission like handing an intern your bank login.
• 4:09
  Next thing, if someone listening does exactly one thing this week, what is it?
• 4:14
  Open chat GPT or Claude, go to connectors and settings, watch team Toldoran, and link
• 4:20
  one read-only source, a lye, your Google Drive, or your calendar.
• 4:26
  Then ask it a question only that data can answer.
• 4:29
  The moment it responds with your real information instead of a generic guess, you'll understand
• 4:34
  why the clipboard days are over.
• 4:37
  Read only, one source, real question.
• 4:39
  That's the homework.
• 4:40
  That's a wrap.
• 4:41
  Fresh AI briefings drop daily at MBA-training.com.