Walk into almost any support or managed services team and ask how long a typical ticket takes, and you will get a confident answer. Ask how that answer was arrived at, and the room gets quieter. This piece takes a relaxed look at the current state of estimating in support work: how teams usually forecast effort, where the guesses and the real numbers drift apart, and what a growing number of teams are doing about it. No statistics are quoted here, because the aim is to describe patterns anyone can check against their own queue.
How support estimates usually get made
In most teams, an estimate is a mix of three things. The first is memory of similar tickets: “we fixed something like this last month”. The second is a rough category: password reset, deployment failure, performance complaint. The third is a sense of the customer, because some customers send tidy requests and others send a screenshot and a single line of text.
All three are reasonable inputs. The trouble is that none is measured. Memory favours the cases that went smoothly, categories hide huge variation, and customer impressions are coloured by the last conversation. The result is a forecast that feels informed but has never been compared with a record.
Trend one: from single numbers to categories with history
A change gaining ground is the move from a flat promise to one that depends on category. Instead of telling every customer that a request takes “about a day”, teams tag tickets by type and look at how long each type has really taken. A quick configuration change and a cross-system investigation obviously differ, and the data soon shows by how much.
This is not a sophisticated technique. It is a spreadsheet with a column for type, a column for the guess and a column for what happened. But it changes conversations. A team that knows its network-related tickets usually run about twice as long as first guessed will plan staffing differently from a team that treats all tickets alike.
Trend two: counting waiting separately
Support work is full of waiting: for a customer reply, for a change window, for another team. A second visible trend is separating hands-on effort from elapsed time. A ticket may need ninety minutes of work and five days to close, and both numbers are true. Teams that record only one of them either make unrealistic promises or look slower than they are.
Once the two are tracked side by side, a useful ratio emerges: how much of the calendar time was actual work. That figure tends to be small, which is a relief to engineers who feel busy but rarely finish things. It also points to the real lever for faster resolution, which is usually to shorten the waiting rather than to type faster.
Trend three: a closer link to engineering principles
The old line between building and supporting software has blurred. Support teams increasingly borrow ideas from delivery engineering: small batches, fast feedback, shared ownership of results. Pieces such as a short read on core DevOps principles for getting software out the door well describe a mindset in which measurement is routine, and the same logic applies to forecasting ticket effort. If feedback is a principle, then a forecast that is never checked breaks it.
Likewise, the shift from manual operations towards scripted infrastructure, discussed in why ops people are well placed to pick up site reliability skills once their infrastructure is code, changes what a ticket contains. Routine requests shrink because they are automated, while the remaining tickets are harder and less predictable. That makes estimates for the rest more uncertain, and all the more worth tracking.
Where the guesses still drift
Even teams with tidy records see consistent patterns of error.
- Diagnosis is underestimated. Finding the cause of a fault is the least predictable stage, and gut feel usually assumes it will be quick.
- Handover is forgotten. Time spent writing up what was done, so the next person can understand it, is rarely in the forecast.
- Interruptions are ignored. A support engineer rarely works on a single ticket in peace. Context switching lengthens everything, but it is seldom budgeted.
- Peaks are averaged away. A forecast based on a typical week fails in the week when several urgent problems arrive together.
None of these errors is dramatic on its own. Taken together they explain why a team can feel busy, deliver good work, and still be late on a surprising share of its promises.
What the more careful teams are doing
The most useful practice we see is surprisingly plain. At the end of each ticket, the engineer records the time spent and compares it with the figure that was given at the start. Once a month someone calculates the ratio by ticket type and shares it in a short meeting. The numbers lead to small, concrete changes: a new line in the intake form that asks for the information that most often causes delay, a standard checklist for the most common faults, or a different promise for the category with the worst ratio.
There is also a human side. Engineers feel freer to answer “not sure yet” when the team has a shared understanding that estimates are provisional. The tracking habit gives them a vocabulary for uncertainty that is less awkward than a guess pretending to be a fact.
Another small but telling habit is writing the promise on the ticket itself, in the first reply to the customer. A figure that lives only in somebody’s head is impossible to audit later. Once it sits on the ticket, a monthly review can pull the promised and actual values side by side without any detective work, and nobody has to rely on recollection.
Finally, some teams add a field for “what we missed at intake”. After a few weeks the free-text answers cluster around a handful of causes, such as absent log access or unclear environment names, and those clusters tell the intake form what to ask next.
A short example of correcting a promise
Imagine a team that tags its incoming tickets and records effort for a quarter. For routine access requests the ratio is close to one. For performance complaints it is nearly two. The team responds by keeping the quick promise for access requests and extending the stated window for performance issues, with an offer to update the customer partway through. Customers see fewer broken promises, and the engineers see fewer evenings spent catching up.
Notice that nothing about the work itself changed. Only the forecast did, and the forecast changed because someone compared it with reality.
A wider habit of looking back
Looking back at results is not unique to support teams. Many groups that work on education, research and technology treat reflection as part of the job. One example is the site run by the Cotocus Foundation, a nonprofit foundation whose pages cover technology, innovation and education, arranged as program, research, community and blog areas. That mix of doing, studying and writing about it is a reminder that regular review is a normal part of getting better.
What to try this month
- Pick the three most common ticket types in your queue.
- For the next twenty tickets, note the promised effort and the real effort.
- Record waiting time separately.
- Calculate a typical ratio for each type.
- Adjust your standard promises for the types that diverge most.
- Share the result with the team and repeat in a quarter.
The overall picture of support estimating today is one of gradual improvement, driven less by new tools than by plain record keeping. Teams that guess, check and adjust are quietly making better promises, and their customers notice even if they never learn why.