The Friday Export That Taught a Support Team to Script

The Friday Export That Taught a Support Team to Script

Meera ran a four-person support operations team at a mid-sized software company, and every Friday at three o’clock a particular dread settled over the room. The weekly service summary had to go out to the account managers, and producing it meant opening the ticketing tool, exporting a list, deleting four columns, sorting by customer, colouring the overdue rows, and pasting the result into an email. It took about forty minutes. It was never difficult. It was simply always there.

Dev, the newest engineer on the team, announced one Monday that he would script the whole thing over the weekend. Meera asked him to wait. What happened over the next six weeks became the team’s favourite story about automation, and it is worth telling in order.

Week one: doing it the slow way, on purpose

Meera’s request was unusual. She asked Dev and the others to take turns doing the Friday summary by hand, and to keep a shared note open while they worked. Every click, every decision, every moment of hesitation went into the note. Dev grumbled, because he could see the script in his head already.

The first note was messy. Dev’s version said “delete extra columns”. Priya’s said “delete four columns, but keep the priority column unless the client is on the premium list”. Nobody had realised the team was performing two slightly different procedures. If Dev had scripted his own version over that first weekend, the account managers would have received a different summary depending on who was on call.

Weeks two and three: finding what never changed

By the third Friday, the notes had converged. The team sat down on Monday and went through them with a pen. Some things were identical every single time: the same export screen, the same filter, the same set of columns, the same sort order, the same email layout. Others changed constantly: which tickets counted as overdue depended on customer agreements, and a handful of rows needed a quick message to the assigned engineer before anyone could decide whether to flag them.

That distinction became the project plan. The fixed parts were candidates for a script. The overdue judgment stayed with a person. Anyone who has worked on incident handling will recognise the pattern; a piece on resolving incidents more quickly with operational intelligence tools describes how routine triage can be prepared by machines while the actual call is left to responders.

Week four: the smallest possible script

Dev finally got to write code, but Meera set a rule: the script was only allowed to do the first three fixed steps, which were to export, trim the columns and sort. It could not send anything, and it could not decide anything. It would drop a tidy file into a shared folder and print a one-line message saying how many tickets were in it.

The script was twenty-odd lines long. It was also, on its first Friday, wrong. A new ticket category had been added the week before, and the sort order put it at the top. Because a person was still doing the review and the email, the team spotted the oddity in under a minute, and Dev fixed it that afternoon. Had the script been wired straight to the inbox of every account manager, the mistake would have travelled much further.

Week five: adding the formatting, keeping the judgment

With three Fridays of reliable output behind them, the team let the script apply the standard colours and generate a draft email. The draft contained a clearly marked line: “Review overdue list before sending.” Priya, whose turn it was, read through the highlighted rows, messaged two engineers, changed two flags and pressed send. Total time: eleven minutes, most of it spent on the part that needed a brain.

Week six: the monthly check

Meera added one more ritual. On the first Friday of each month, one team member would produce the summary entirely by hand and compare it with the script’s output. This felt like wasted effort for about two minutes, until the first comparison caught a subtle shift: the ticketing tool had renamed a column, and the script had silently started using a different field. A manual sample caught it before anyone downstream did. The team also noted that the hand-run kept the old skill alive, which mattered the day Dev went on holiday.

What the team learned

  • Doing the task by hand first exposed that the team had two different procedures.
  • Separating fixed steps from judgment gave the project an obvious, small scope.
  • A script that reports, rather than sends, keeps errors close to the people who can fix them.
  • A monthly manual sample catches drift that no alert will mention.

The same idea, spread beyond the team

Word got round the company. The partnerships group, which handled enquiries from overseas contacts, asked Meera to look at their own repetitive routine: replying to introductory messages with the same list of questions. She did not suggest a script. She suggested a standard enquiry form and a saved reply, and she pointed to how a medical tourism website, HipHospitals, collects free-quote and appointment requests from international patients through structured forms organised around hospitals and doctors by country. The partnerships team adopted a similar form for their own purposes, and the repeated back-and-forth shrank, while the personal replies remained personal.

A footnote on model upkeep

Months later, the data team wanted to apply the same idea to checking whether their models were drifting. Meera suggested the same sequence: do the check manually a few times, find the constant parts, script those. For background reading she shared a short primer on the fundamentals of keeping an eye on deployed models. The data team’s first script merely collected the numbers they were already looking at by hand. Interpretation stayed with them.

Where the story ends

The Friday dread did not vanish entirely; somebody still had to read the overdue list. But the forty minutes became ten, the summaries became consistent and, more importantly, the team trusted the script because they had earned that trust by watching it work. Dev, for his part, admitted that his weekend version would have been wrong in at least two ways. He now begins every automation request with the same question: have we done it by hand enough times yet?