Cancelling repetitive tasks

Walk into any public office and watch for an hour. Someone is copying numbers from one sheet to another. Someone is checking that a name on one document matches a name on a second. Someone is typing the same status update into the same template for the fortieth time this week. These are not jobs. They are tasks that have been attached to jobs because, until now, there was no other way to get them done.

Now there is. So the question for every head of department is not whether to automate. It is which tasks to cancel first.

Find the repetition

Repetition hides in plain sight because it is normal. The way to see it is to count. Take a team for a week and have them note every task they do more than three times a day. Not the important work. The repeated work. The list is always longer than anyone expects, and it is always the same kinds of things:

• Reading a document to find out what it is and where it should go.
• Comparing two records to see if they agree.
• Moving information from one system, form or spreadsheet into another.
• Writing the same letter, notice or report with different names in it.
• Answering the same question from the public for the thousandth time.

Every one of these is now within reach of a well-built AI system, run on infrastructure the government controls.

Build the cancel list

Rank the list by hours. Not by how interesting the automation would be, and not by which vendor is in the building this month. Hours. The task that eats the most of the team’s week goes first.

Then, for each task, decide one of three things. Cancel it entirely because nobody needs the output. Automate it because the output is needed but the labour is not. Or keep it human because it carries judgement, risk or a relationship. Most tasks fall in the middle category. More than you expect fall in the first.

Count the hours. The task that eats the most of the week goes first.

Automate the right things, properly

Classification, extraction, matching, drafting and routing are solved problems. A system can read an incoming letter, work out what it is, pull the facts from it, check them against the register and draft the reply for an officer to approve. That is one afternoon of work per hundred letters turned into one minute per hundred.

Do it properly. The system runs on data that stays in the country. Every automated action leaves a record. An officer can see what the machine did and overrule it. Nothing goes out to a citizen without a human name on it. These are not constraints on the ambition. They are what make it safe to be ambitious.

Give the hours back

Here is the part that gets missed. Cancelling repetitive tasks is not a cost-cutting exercise, and the public service should not sell it as one. It is a redeployment. The officer who spent three days a week on data entry now spends three days a week on the public: resolving the hard cases, visiting the district, answering the phone. The service gets better because the people are finally doing the job they were hired for.

That is why we train. BUILD AI Academy exists to put practical AI skills into the hands of hundreds of thousands of Ghanaians, including the ones sitting in those offices. The people who cancel the tasks should be the people who were doing them. They know where the repetition is. Give them the tools and they will find it faster than any consultant.

Measure one thing: hours returned to service. Publish it. Then do it again next quarter.