Starting Metrics on a Small Desk, One Quarter at a Time

Wednesday, 16:20, at Dunmore Plastics. Ines and Rob are the IT department, all of it, for a factory and an office of about a hundred and forty people. Until the spring their ticketing tool was a shared mailbox and a good memory. They have had a proper tool for six weeks. This afternoon Gavin, the managing director, has sent a two-line e-mail: "Could you put together some numbers on the help desk for the quarter-end? Nothing fancy." Ines opens the tool's reporting page, which offers her fourteen numbers, and then a blank spreadsheet. She sits looking at both. Rob, from the next desk, says "nothing fancy" in a voice that suggests he has heard the phrase before. Neither of them has ever produced a help desk metric. They have about seven weeks.

Starting small means measuring fewer things than you could, exactly, for long enough that the trend means something. It means adding the next number only when the current ones have become routine. A small desk has less data than a large one, less time to produce numbers, and a lead who is also the technician. The method that fits it is a quarter long. It starts with three numbers and a habit in month one, the clock in month two, and the requester in month three. This article is part of our Helpdesk Metrics That Matter series. It is the one to read if the desk is you, or you and one other person, and the tool is new or cannot report.

It covers month one, then the spreadsheet fallback when the tool cannot produce a number. Month two and month three follow. Then we cover when to add the next metric and when not to, what to stop measuring, and the traps particular to small numbers. We finish with a short war story and the first-quarter measurement plan to copy.

Month One: Three Numbers and a Habit

The first month is for three numbers and the habit of writing them down. The three start with tickets created and resolved each week. Next come the age of the oldest open ticket and the count of tickets that came back after being marked fixed. All three are deliberately the simplest measures a desk can keep. The case for starting a brand-new desk with a set that small is made in the first metrics for a new desk. This article does not re-argue it. It assumes you have a first handful, or will have by the end of the month, and builds the rest of the quarter on top. Each of the three is a count or a date, which means each can be produced from any tool, or from no tool, in minutes.

The habit matters more than the numbers in month one. Pick a time, say Friday at four, and spend ten minutes writing the week's three figures into a sheet with the date beside them. Do it when the week was quiet and nothing moved. Do it when the week was dreadful and you would rather not know. The point of month one is to find out whether you will do it when you do not feel like it. A desk that measures only in the weeks it has time to is measuring its own calm. Its calm is the least interesting thing about it. Ines and Rob did it on Fridays, alternating, and the second Friday Rob forgot. That is why the plan at the end of this article has a line for whose turn it is.

Write the definition of each number down on the day you first produce it, in a sentence. Do this even though the number seems too simple to need one. "Created means the ticket's created timestamp falls in the week; walk-ups logged on the day count." "Oldest means the open ticket with the earliest created date, in whole days to today, including anything waiting on a requester." "Came back means the requester told us it was not fixed after we marked it resolved." These sentences are the start of the definition sheets in defining each metric exactly. Writing them in month one costs five minutes. Writing them in month six, after two people have been counting differently, costs an argument.

The Spreadsheet Fallback When the Tool Cannot

Many small-desk tools, and most shared mailboxes pretending to be one, cannot produce the six metrics in this series. The fallback is a spreadsheet with one row per ticket. The columns are the facts the metrics are made from. They start with ticket number, the date and time created, the requester's department, and the channel it arrived by. Next come the date and time of the first reply a person wrote, and the date and time it was marked resolved. Then the columns record whether it was reopened and when, the rating if one came back, and the status on the last day of the month. Nine columns. If the tool can export its tickets to a spreadsheet file, as most can in some form, the rows come from the export. In that case, the month's work is pasting. If it cannot, the row is typed when the ticket is created and finished when it is resolved. On a desk of sixty tickets a month, that is two or three minutes a day.

Every metric in the series is then a formula over those columns. And all of them can be written in plain words before anyone looks up the syntax. Tickets created in the month is a count of rows whose created date falls in the month. Tickets resolved is a count of rows whose resolved date does. The backlog at month end is a count of rows created on or before the last day whose resolved date is blank or later than that day. Age is the month-end date minus the created date, in days, and the oldest ticket is the largest of those among the open rows. Time to resolution is the resolved date and time minus the created date and time. The month's figure is the median of that column over the rows resolved in the month, with the 90th percentile beside it. Most spreadsheets have a median function and a percentile function that take a range and return the value. Reopen rate is the count of rows reopened within the window divided by the count resolved. None of this needs a macro, and a reader who can write "count the rows where" can write all of it.

Business hours are the one thing the spreadsheet does badly, and a small desk should say so rather than fake it. Counting only working hours between two timestamps is possible in a spreadsheet and fiddly. The common approach for the first quarter is to use wall-clock hours and write "wall clock" on the definition sheet. This approach switches to business hours when the tool can compute them, recomputing the earlier months at the switch. The wall-clock figure will be larger and the weekends will show in it. That is fine as long as every month is counted the same way. A spreadsheet is not an apology for a tool. It is a tool, and for a desk of this size it may be the right one for a year. If you run QuerySolv, the standard reports already produce volume, backlog by age, response and resolution times, reopens and satisfaction. In that case, the spreadsheet is mostly unnecessary and the plan is the same with less typing. In any tool, the plan is the plan, and the spreadsheet is where the gaps go.

Month Two: Add the Clock

Month two adds the two time metrics, first response and time to resolution, and turns the single backlog count into age bands. The habit from month one carries them. The Friday ten minutes becomes fifteen, and the sheet gains two columns of medians and two of 90th percentiles. First response is the time from creation to the first reply a person wrote, excluding the automatic acknowledgement. Time to resolution is the time from creation to the ticket being marked resolved, not closed. Both are reported as a median and a 90th percentile and never as an average alone. The vanity metrics article in this series spends a section on the reasons. Both time metrics get a definition sheet before the first figure is written down, with the hours decision, wall clock or business, made once.

The age bands are the cheap addition that makes the backlog mean something. Instead of one count, there are three: open tickets under a week old, one to four weeks, and over a month. The oldest ticket's age and number sit beside them. On a small desk the bands will be small numbers, nine and four and two, and that is the correct reading. Two tickets over a month old on a sixty-ticket desk is two people, and the desk lead can name them. Include every status that means the desk still owes somebody something, waiting statuses included. A small desk is especially prone to parking tickets as "waiting on the requester" and forgetting them. There is nobody to notice except the person who parked them.

Month two is also when the first hand check happens. Pick five resolved tickets, open each one's history, and compute the two times by hand with the definition sheet in front of you. Then compare with the sheet or the tool's report. On a small desk the check takes ten minutes and finds something about half the time. Usually it finds a reply that was counted as a first response but was an internal note. Or it finds a resolved timestamp that is actually the closed one. The end of month two is also the first time two numbers can be read as a pair. The pairs are created against resolved beside the backlog, and resolution time beside reopens. Ines read them on the last Friday of month two and found that resolution time had fallen and reopens had risen. Both were Rob, who had been marking tickets resolved when he sent the fix. Rob said he had been doing that for six weeks and nobody had told him it mattered. Nobody had been measuring it, so nobody could.

Month Three: Add the Requester

Month three adds satisfaction, which is the last of the six because it is the only one that needs something from outside the desk. The mechanism is one question, sent with the resolving reply, with a comment box, and no login, no reminder and no second question. And the design of a survey people actually answer is the subject of surveys people actually answer. Where the tool sends it automatically, the desk's work is reading what comes back. Where it does not, a small desk can add a line to the resolving reply template. That line can ask the requester to reply with a word, good or not, and the desk can tally the replies. It is cruder than a button and it works, and a desk of this size will get back a handful a week.

The reading on a small desk is counts and comments, never a score. With sixty tickets a month and a response rate of perhaps one in six or seven, the desk will see eight to twelve ratings a month. Eight ratings cannot produce a percentage that means anything; one annoyed person moves it by a tenth. Write the number received, the number good, the number of requests sent, and then read every comment. The comments are the part that tells you what to fix. The response rate itself is the figure to watch across the quarter. A falling response rate on a small desk usually means the requesters have stopped expecting anything to come of it. That is worth knowing before the score is.

The end of month three is the end of the quarter, and two things happen. The first is the desk's first scorecard, the one page with the three pairs on it from reading metrics together. It is filled in with three columns, one per month. That way, the quarter is visible as a trend rather than a month. The second is the first report to whoever asked for numbers, and it is not the scorecard. The report is one page, with a handful of figures, three sentences about what they mean and one thing the desk wants. Its form is set out in the monthly report they will actually read. Ines gave Gavin volume in and out for the quarter, the backlog with its oldest ticket, and the reopen count. She included a sentence explaining why the reopen count had risen in month two and fallen in month three. He read it. It was, she said afterwards, slightly fancy, and he did not seem to mind.

When to Add the Next Metric, and When Not To

After the first quarter the six are in place, and the question becomes whether to add a seventh, and the answer is usually not yet. Three tests have to pass, all of them. The current numbers must have been produced the same way for three months without anyone having to think about how. A desk that is still arguing about whether a note counts as a first response is not ready for anything new. The decision the new number changes must have a name; "it would be interesting" is not a decision. And the new number's partner must have a name: the metric that would catch it if it started to lie. A metric added without a partner is a metric that will be read alone. The whole of this series is about why that goes wrong.

The candidates that usually pass come in roughly the order small desks reach them. First is first contact resolution once reopens are being counted properly, because it needs the reopen window to be honest. Next is volume cut by department and by category once the categories have settled, because that is where the hotspot reading starts. Then come the slowest ten tickets each month as a named list. That is less a metric than a habit and is worth more than most metrics. The candidates that usually fail include anything per person on a desk of two and anything the tool offers that nobody asked for. They also include anything with the word "utilisation" in it. Adding a metric because the tool shows it is the commonest way a six-number desk becomes a fourteen-number desk. The vanity metrics article in this series has the test to run before it happens.

Add one metric per quarter at most. Add it the same way the first six arrived. That means a definition sheet first, a month of producing it before anyone reads it, and a partner on the scorecard from the first day. A desk that adds a number a month will have twenty by the end of the second year and will read none of them. A desk that adds one a quarter, when the tests pass, will have perhaps nine after a year and will read them all. That is the only thing that was ever the point.

What to Stop Measuring

Measurement has a weight, and every number on the sheet is a small promise to look at it. So the end of each quarter is also when a desk drops something. The rules for dropping are the mirror of the rules for adding. A number nobody acted on in the quarter, that changed no decision and prompted no conversation, has shown it is not a metric and goes. A number whose definition nobody on the desk can state without looking goes, or gets a sheet by the end of the week. A number that was added for a project that has since finished, the migration count, the rollout tally, goes with the project. And anything per person on a desk of one or two goes immediately. A per-person figure on a two-person desk is a sentence about a colleague wearing a decimal point.

Dropping is harder than it sounds, because the number is on the sheet and the sheet has a column for it and the column has a history. Drop it anyway. Move the old column to a tab called "retired" with the date, so the history is kept and the sheet is not. Tell whoever reads the report that the number has gone and why, in one sentence. Ines and Rob dropped tickets-by-channel at the end of their second quarter. It had said "mostly e-mail" three months running and nobody had done anything about it. They reinstated it a year later when the portal went live and the question had become real. A dropped metric is not a dead one. It is a metric waiting for its decision to come back.

The test for the whole set, once a year, is to read the sheet. Ask of every column what the desk did differently because of it in the last twelve months. Columns with an answer stay. Columns with a shrug are the next quarter's drop. A desk that has done this honestly for three years tends to have the six from this series. It also tends to have two or three it added for its own reasons, and a retired tab that is longer than the live one. That is what a measurement habit looks like when it is working.

Small Numbers on a Small Desk

Everything in this series is harder with sixty tickets a month than with six hundred. On a small desk one ticket moves a median and one bad week is a sixth of the quarter. The first adjustment is to prefer counts to rates wherever the denominator is small. "Three of fifty-five resolved tickets came back" is a sentence. "Five point five percent" is the same fact pretending to be more precise than it is. The pretence is what gets quoted. Write the count, write the denominator, and let the reader divide if they want to.

The second adjustment is to read the quarter rather than the month for anything that wobbles. Volume on a small desk swings with one department's project; satisfaction swings with one annoyed requester; reopens swing with whoever was on leave. The three-column scorecard exists for this: a number that moved one month and moved back is weather. A number that has moved the same way for three months is the thing to act on. Where a month is genuinely too small to read, a rolling three-month figure gives a steadier line. That is the median over the last ninety days of tickets rather than the last thirty. The definition sheet says that is what it is.

The third adjustment is about people, and it is absolute. On a desk of one, every number is about you. On a desk of two, every per-person cut is about the other person. Neither reading is useful or kind. Measure the desk. The two-person desk at Dunmore knew perfectly well that Rob's early resolves were Rob's. There were only two of them and Ines was the other one. The scorecard said "the desk's reopens rose", the conversation was between colleagues, and the fix took a week. That is what the per-person rule in the larger articles of this series comes down to on a small desk. The rule is coaching not ranking and never below a minimum sample. There is no sample, so there is no cut, and the number belongs to the two of you.

A Short War Story: The Summer Numbers at Wrenfield Academy

Wrenfield Academy was a secondary school with one IT technician called Callum and about ninety staff. The deputy head, Ruth, asked in June for "proper metrics" to support a request for a part-time assistant. Callum set up eleven numbers in the first week: the tool's full default report, a spreadsheet with nine tabs, and a satisfaction survey with four questions. He kept all of it faithfully for six weeks of the summer holiday. The school was nearly empty and the desk saw about eight tickets a week. Then term began, and ninety tickets arrived in the first five days. The spreadsheet was not opened again until Ruth asked for the figures in late October. What existed was six weeks of summer data showing a quiet desk with a two-hour resolution median. There was nothing at all from the eight weeks of term. So the governors, reading the only numbers they had, concluded the desk was comfortably staffed and deferred the assistant for a year. Callum's backlog passed a hundred and forty open tickets by the following February. The exam-season printing failure that took him two days to reach cost the office roughly a week of staff time in reprinting and rescheduling. The assistant was approved the next summer on the strength of three numbers Callum had kept every single Friday since November. Three numbers kept through the worst week are worth more than eleven kept through the quiet ones. The worst week is the one the decision was about.

The First-Quarter Measurement Plan

The plan below is the quarter on one page, with the three months as three blocks. Two boxes at the end outlive the quarter: the tests for adding a metric and the tests for dropping one. Fill in the tool's capabilities at the top first, honestly. That decides how much of the plan is reading and how much is the spreadsheet. The "whose turn" line in month one is there because the second Friday is the one that gets missed. A plan that survives the second Friday survives the quarter.

FIRST-QUARTER MEASUREMENT PLAN — ONE PAGE
Desk: ______________   Quarter: ______________   Owner: ______________
Tool can produce:  [ ] volume   [ ] backlog by age   [ ] first response
                   [ ] time to resolution   [ ] reopens   [ ] satisfaction
Spreadsheet needed for: ______________________   Set up, one row per ticket:  [ ]

MONTH ONE — THREE NUMBERS AND A HABIT
  Measure:   tickets created and resolved, by week      oldest open ticket, in days
             tickets that came back after being marked fixed, count
  When:      every ________day at ____:____, ten minutes       Whose turn: ______________
  Definition sentence written for each:  [ ] volume   [ ] oldest   [ ] came back
  Walk-ups logged as tickets:  [ ]      Waiting tickets included in oldest:  [ ]
  End of month one: four weekly rows in the sheet, no report yet

MONTH TWO — ADD THE CLOCK
  Add:       first response, median and 90th percentile
             time to resolution, median and 90th percentile
             backlog by age band: under 7 / 7 to 30 / over 30 days, plus the oldest
  Hours:     [ ] wall clock for now, switch later   [ ] business hours ____:____ to ____:____
  Definition sheets written:  [ ] first response   [ ] resolution   [ ] age bands
  Hand check, five tickets recomputed by hand:  [ ]   agreed:  [ ] yes  [ ] no: ________
  End of month two: first pair read (volume with backlog; resolution with reopens)

MONTH THREE — ADD THE REQUESTER
  Add:       one question with the resolving reply, with a comment box
  Read:      ratings received / requests sent      good ratings      every comment
  Rule:      under ____ responses, read comments only and quote no score
  End of quarter:  first scorecard, three columns  [ ]
                   one-page report to ______________, due ________  [ ]

ADDING A METRIC (every box must be ticked first)
  [ ] The current numbers have been produced the same way for three months
  [ ] The decision it changes: ______________________________
  [ ] Its partner on the scorecard: ______________________________
  Next candidate: ______________________   Earliest it could start: ________

DROPPING A METRIC (review at the end of every quarter)
  [ ] Nobody acted on it this quarter
  [ ] Nobody can state its definition without looking
  [ ] It was added for a project that has finished
  [ ] It is a per-person figure on a desk too small for one
  Retired this quarter, with date, to the retired tab: ______________________

Ines and Rob's first quarter ended with six numbers, one page for Gavin, and a retired tab that was still empty. Their second quarter ended with the same six, a dropped channel column, and one addition, the slowest ten tickets as a named list. Rob reads the list aloud on the first Friday of the month in the tone of a man reading a weather warning. Gavin has asked for the numbers every quarter since. He has stopped saying "nothing fancy", which both of them have chosen to take as a compliment.

Frequently Asked Questions