The 20/60/20 Rule: AI Cannot Do 60 Percent of the Work Without Your Business Context

AI cannot do 60 percent of the work with 0 percent of your business context.

The 20/60/20 rule is the cleanest way to describe how a person and an AI should split the work. It is also the rule that breaks the first time you try it inside a real business.

What the 20/60/20 rule says

You handle the first 20 percent. Define the problem, set the direction, decide what matters.

AI handles the middle 60 percent. Research, analysis, drafting.

You handle the final 20 percent. Verify the work, apply judgment, make the decision.

The split is sound. The ratio is not the problem.

Where the rule breaks

Then you run it inside a business.

Your financials are in QuickBooks. Your pipeline is in a CRM. Your marketing performance is in Google Analytics. The reasoning behind those numbers sits in email threads and documents.

The AI sees whatever you remembered to paste into the chat.

So the middle 60 percent turns into preparation. You export the reports. You upload the files. You explain the business again. You hunt for the context the model is missing. You check whether the answer matches the numbers you already had.

You were supposed to hand off 60 percent of the work. You picked up a second job instead, feeding the AI and re-checking what comes back.

Context is the missing input

The middle 60 percent is not a model problem. Frontier models are already good at research, analysis and drafting.

It is a retrieval problem. An assistant that cannot read your systems can only reason about what you hand it. Every gap in the input becomes a gap in the answer, and you will not always see which gap produced which claim.

This is also why a stronger AI subscription does not close the distance on its own. Access to a better model changes the quality of a sentence. It does not change whether the model can read the invoice, the campaign or the deal.

What closes the gap

CorpusIQ connects AI to your business data and returns source-cited answers, private by design. You can see where an answer came from and check the evidence behind it, inside the AI client you already use.

External source connectors retrieve without writing back. They do not create campaigns, change bids or budgets, send email, or edit records in the systems you connect.

CorpusIQ works in ChatGPT, Claude and Perplexity. One connection answers the same question in any of them, with the same citations attached.

Two properties matter more than the connector count:

• Every answer carries its source. A figure you cannot trace is a figure you have to re-verify by hand.
• Access is granted per system and can be revoked. You approve each connection, and you can remove it.

What the middle 60 percent looks like with context

• The question and the answer stay in one window. No exporting between a report and a chat.
• Each figure arrives with the system it came from, so the final 20 percent starts from evidence instead of suspicion.
• The first 20 percent gets sharper, because a better question is easier to ask when the context is already there.
• The work repeats. The same question on Monday returns Monday's numbers, not last month's export.

Where this earns its keep

Recurring work benefits first. Closing the books, reviewing ad spend, checking pipeline coverage, preparing a board update, or answering why margin moved in one pass instead of five tabs.

The pattern holds for any question whose answer lives across more than one system. That is most of the questions a business owner actually asks.

What to watch out for

• CorpusIQ reads the systems you connect, not every system you own. Data outside those connections stays invisible.
• The final 20 percent is still yours. The rule does not remove judgment. It protects it.
• Freshness belongs to the source. A connector can only return what the system it reads has recorded.

See also

• Read-Only OAuth and Scoped Data Handling
• Why CorpusIQ Is a Business System, Not a Chat
• Connect Google Analytics 4 to Claude