What AI can run in a business today, and what to keep human

What business tasks can AI automate? The work that repeats and leaves a written record: answering, confirming, following up, invoicing and reporting. The research points to the same places in almost every company.

BlogAutomationUpdated 3 min read

Automatable60 to 70% of employee time
Where it sitsCustomer operations, sales and marketing
One companySix of seven steps, 29 hours a week
Keep humanPhysical work, judgment and the phone

Most lists on this topic are ten things you can ask a chat tool to do: summarize a meeting, tidy an inbox, draft a post. Those save minutes. The larger question is which whole pieces of work can leave your desk.

Where the research says the value sits

McKinsey analyzed 63 generative AI use cases across 16 business functions. About 75% of the value falls in four of them: customer operations, marketing and sales, software engineering, and research and development.

Across all work, it estimates that current AI and related technology could automate activities that take up 60 to 70% of employee time. The biggest jump is in knowledge work: decisions, collaboration and applying expertise, which used to be the hardest work to automate.

The work that moves first

Customer operations

Answering questions, confirming, reminding and following up. McKinsey estimates AI in customer care could raise productivity by a value of 30 to 45% of the function's current cost. In a study it cites of 5,000 customer service agents, AI help raised issues resolved per hour by 14% and cut the time spent on each issue by 9%.

Sales

Researching and ranking leads, writing follow-up, and keeping leads warm until they are ready to talk to a person. McKinsey sizes the gain at 3 to 5% of current sales spending, before counting the new leads that faster follow-up brings in.

Marketing

First drafts of social posts, emails, headlines and product descriptions, kept in one voice. McKinsey puts the gain at 5 to 15% of total marketing spend, with people still owning the strategy.

The work around the work

Asana found knowledge workers spend 58% of the workday on work about work: chasing status, coordinating and searching for information. Reporting, record keeping and handoffs are where the most hours come back.

What it looks like inside one company

Research describes averages. Here is one company. Tidy Up Services is a cleaning company in Jacksonville, Florida, and its whole business is seven steps.

StepWho does it
Leads and quotesSoftware. The customer picks the service and sees the price
Booking and schedulingSoftware. The customer books a time on the site
Cleaner assignmentSoftware. Each job goes to the right cleaner with its checklist
Confirmations and remindersSoftware. A text goes out at 10 AM the day before
Invoicing and paymentsSoftware. The invoice sends on completion and unpaid ones are followed up
PayrollSoftware. Pay is worked out from completed jobs
Marking the job completeThe owner

Six of seven run with nobody touching them. By hand they took 29 hours a week.

What to keep human

Physical work, relationships and judgment calls with real stakes. McKinsey found generative AI moved the automation potential of physical work far less than cognitive work. The clean still needs a cleaner.

Keep the final check on anything money depends on. At Tidy Up the one step a person still does is marking the job complete, and the invoice waits for it.

Keep the phone with people if your customers call. Xaedros does not answer phone calls.

Why a chat tool alone does not get you there

Most companies start by giving everyone a chat tool. It helps, and less than people expect. In a St. Louis Fed survey, workers who used generative AI saved 5.4% of their hours, about 2.2 hours a week.

The reason is the carrying. A chat tool does one task when you ask, and you still move the result to wherever it has to go. The steps at Tidy Up run without that, because each one hands its result to the next.

A four-part test for any task

Before you hand a task over, check it against these.

  • It repeats. Daily or weekly work adds up fast.
  • It leaves a record. The inputs and outputs are written down. Software can only run work it can read.
  • The decisions are small. Which slot, which price, which reminder. Leave the large ones with a person.
  • A mistake can be caught. Something or someone checks the output before it reaches a customer or the books.

How to find the tasks in your own company

Write down everything that happened last week that was not the work you sell. Circle each item that repeats and ends in something written. That circle is your list.

The hours only come back when the software holds the task itself. That takes AI agents that share one record of the business, which is different from a chat window you feed one task at a time. Then compare the list against what a hire for it would cost and what done-for-you automation covers.

Questions.

What business tasks can AI automate?

Customer questions, confirmations and follow-up, lead research and nurturing, first drafts of marketing content, invoicing, reporting and record keeping. McKinsey estimates current technology could automate activities that take up 60 to 70% of employee time.

Is AI only beneficial for large companies?

No. Tidy Up Services, a cleaning company in Jacksonville, runs six of its seven business steps on software, work that took 29 hours a week by hand.

What can AI not automate yet?

Physical work, relationships and high-stakes judgment still need people. AI takes over the repeatable written work around them.

More from the blog