Measuring the Shift Left: By Tier, Over Time, Honestly

Thursday, 14:00, at Bluewater Bank, the quarterly review. Esther, who runs the desk, has one slide for the trading floor's quote-screen ticket. It used to wait all day for Maren, the senior engineer, to spend four minutes on it. The slide says escalations of that type fell from thirty-one last quarter to four this quarter. She is pleased with it, and the operations director, Ingrid, reads it and asks two questions. "Is that because the front line fixes it now, or because they stopped sending it up?" And then: "What did Maren do with the time?" Esther has an answer to neither. The first would need the reopen rate for the type, which she has not cut by type. The second would need someone to have written down what the freed hours went on, which nobody did. The slide is true. It is also, Esther realises as she looks at it, the only number on it that would have gone the same way if the move had failed.

Measuring the shift left means showing, for each ticket type that moved, where its resolutions happen now compared with before, quarter by quarter. Three numbers go with that to say whether the move was real. They are the escalation rate for the type, the reopen rate for the type, and the hours the move freed. Those hours are counted after the enablement that bought the move was paid for. For types that reached self-service, the measurement adds a fourth, the deflected count carried as a range. Each number is counted under rules written down once, and the four are reported together in one line. Any one of them alone is a target the front line will hit for you. This article is part of our Shift-Left Strategy series. And it is the last one, because measurement is where the strategy finds out whether it happened.

It covers the three questions the numbers have to answer, resolution by step over time, and the escalation rate falling for the right reason. Then come the self-service share, specialist hours freed and what they bought, and reading the four together. Next are per-technician numbers and why they are for coaching, followed by the honest report line. A short war story and the shift-left scorecard to copy come last, in that order.

The Three Questions the Numbers Have to Answer

Ingrid's two questions and one more are the whole of the measurement. Did the resolution move: are tickets of this type now resolved at a lower step than before? Did it stay fixed: are the resolutions at the lower step as good as the ones at the higher step were? And what did the hours buy: did the specialist's freed time go somewhere the organisation can see? Every number on the scorecard serves one of those questions, and a number that serves none of them is decoration. The escalation rate serves the first, the reopen rate the second. The hours line serves the third, only when it says what the hours bought.

Write the counting rules before the first count, on one sheet. Apply them the same way every quarter whether or not the result flatters anyone. A figure without its rules is an opinion with decimals. The thinking behind that, with five rules for one metric, is in Measuring First Contact Resolution Honestly. The rules for shift-left are fewer and the discipline is the same. Which tickets are in the type? Which step counts as the resolving one? What counts as an escalation? What counts as a reopen and over what window? Which hours are subtracted? The scorecard at the end of this article carries the rules on its own page. That way, whoever counts next quarter counts the same way. The rules are the metric. The numbers are what the rules produced this quarter.

Count by ticket type, never by the desk as a whole, until the end. A desk-wide escalation rate falling can mean the front line is finishing more. Or it can mean three high-volume types moved while everything else got worse, or that a department left. A type's escalation rate falling means the front line is finishing that type. Add the types up on the scorecard's last line if a summary is wanted, and keep the lines above it. The lines are where the truth is and the sum is where it goes to hide.

Resolution by Step, Over Time

The core measure is a count, per type per quarter, of how many tickets were resolved at each step of the staircase. The steps are specialist, second line, front line, self-service, prevented. The resolving step is the step of the person who wrote the resolving note. It is not the step the ticket was first assigned to and not the highest step it visited. A ticket the front line escalated and the specialist fixed resolved at the specialist. And a ticket the specialist glanced at and sent back with a hint, which the front line then fixed, resolved at the front line. Decide that rule once; the honest version is "whoever wrote the fix".

Where the tool keeps the history of ownership and notes, the resolving step is a filter rather than a reading. If you run QuerySolv, the event history on every ticket records each claim, handoff and reply. So the step at which a type resolved is a question the ticket already answers. In any tool with an ownership history, export the type's tickets with the last owner before resolution and count. In a tool without one, read the resolving notes for the quarter, which for a single type is an hour. That is the hour that makes everything else on the scorecard honest.

Bluewater's quote-screen type, over five quarters, is the shape to expect, and the table below is its worked example, with the type's whole staircase walked. The quarter before the move, almost everything resolved with Maren. The first quarter after, the front line took most of it and the specialist still saw a fifth. That is normal and is the feedback loop at work. The second quarter, the front line took nearly all of it. The third, the user guide went on the portal and the ticket count itself fell. The fourth, Maren used some of the hours the move had given her to get the underlying fault fixed with the supplier. The type all but disappeared. Five quarters, five steps, one ticket type. (The slide for that would have been worth showing. It needed five columns.)

Read the table down a column before reading it across a row. Quarter one's column says the move happened and was not yet settled. There were nine still with Maren, eight with the second line, and seventeen escalations, the cases the runbook had not covered yet. Quarter two's says it settled. Quarter three's says the type started leaving the desk altogether, and the deflected count is a range because that is the only honest form it has. Quarter four's says the cause went away. Across the rows, reopens never rose, which is the answer to Ingrid's first question. The table is the answer to the second, once the last two rows are read together, and the next sections say how.

The Escalation Rate, Falling for the Right Reason

The escalation rate for a type is the share of its tickets that were handed up from the front line to a higher step. That applies whoever eventually resolved them. It is the number that goes down when a move works. It also goes down when the front line simply stops handing tickets up. The two cases look identical on a slide. They are told apart by the reopen rate for the same type over the same quarter, and by the type's resolution time. A front line that has stopped escalating and started fixing shows reopens flat and resolution time falling. A front line that has stopped escalating and started closing shows reopens rising and resolution time that has not moved. The mechanics of counting an escalation so that the count means the same thing every month are in Measuring Escalations. They include the quiet reassignment that is an escalation in everything but name.

First contact resolution rises with shift-left, for the type, and the desk-wide figure is the one to be careful with. When the front line finishes a type it used to hand up, that type's first-contact share goes from nearly nothing to most of it, which is real. Suppose the desk reports its whole first contact resolution rising on the strength of three moved types, and does not say so. It has reported an improvement in the mix as an improvement in the desk. The way to raise the figure by widening what the front line can finish is in Raising FCR Without Cherry-Picking. That replaces picking the easy tickets and calling it progress. Shift-left done well is that article's method applied type by type. Report the type's figure as the type's. Let the desk-wide figure follow, and say what moved it.

The quarter after a move, expect escalations of the type to fall by something like half to two thirds, not to nothing. And treat the ones that remain as the feedback they are. A type's escalations may go to zero in its first quarter. In that case, the type has a front line that is not sending up the cases the limit names, or a runbook that was perfect first time. And I have met one of those two things.

The Self-Service Share

For a type that has reached the portal, the scorecard adds the deflected count. It is carried as a range from a low count to a high count, never as a point. The method that produces it is an estimate with two defensible ends and no defensible middle. The low count is the responses that say the article solved it, or the drop in the exact question the article answers. The high count is the whole drop in the type after the control type's drift is removed. The method, with its baseline, its control and its two columns, is in Deflection Math Without Fooling Yourself. The scorecard borrows the result and writes it in one cell. Bluewater's quarter three says eleven to seventeen, against a control type that did not move. Both ends are reported. The reader picks a point, which is the reader's job.

The self-service share for a type is the deflected count divided by the deflected count plus the tickets that still arrived. It is a share of an estimate, so it inherits the range. Bluewater's quarter three share was somewhere between a quarter and a third of the type's demand. Report it that way, in words, with the counts beside it. A desk-wide self-service share can be built by summing the deflected ranges across every type with an article and dividing by the sum plus all tickets resolved. And it is a useful trend line for a desk that has been doing this for a year. In the first two quarters it is a small number with a wide range; leave it as counts.

Never let the self-service share be improved by the counting rules. A "sent the link" reply is a ticket resolved by the desk, not a deflection; a ticket raised after the user tried the article is a ticket; a view is a view. The fifth article in this series set those rules out of fairness to the user; here they are rules of measurement. The share goes up when the article answers the question first, and not otherwise.

Specialist Hours Freed, and What They Bought

The hours line is two numbers, and the second is the one that answers Ingrid. The first is the hours freed. Start with the type's tickets times the people time per ticket before. Subtract the tickets times the people time after, then the enablement hours spent that quarter. The result comes from the desk's own counts, as a range because the per-ticket figures are middle estimates. For Bluewater's quote-screen type in quarter two, that is forty-six tickets at roughly forty-five minutes before against forty-six at roughly twenty-two after. That is about seventeen hours freed, less the one hour of feedback reading: about sixteen hours in the quarter, most of them Maren's. The first quarter's eleven enablement hours came out of that quarter's saving and left it close to break-even. That is normal and should be shown rather than smoothed. The two figures for an hour, the hands-on one and the loaded one, are in the ROI series this library already has. So is why the desk reports in hours rather than money. The scorecard uses hands-on hours and says so.

The second number is what the hours bought, and it is a sentence, not a figure. It names the project that moved or the prevention work that got done. Or it names the hard tickets that were resolved faster because the easy ones were no longer in front of them. Maren's sixteen hours in quarter two went on the supplier case for the quote screen's underlying fault, which is why quarter four's column looks the way it does. That is a sentence Ingrid could take to the trading floor. Freed hours with no sentence beside them are a saving nobody can find. A manager who cannot find a saving will, reasonably, conclude it was absorbed into the same work as before. Ask the specialist each quarter what the hours went on, and write the answer on the scorecard in their words.

Watch the specialist's resolution time for the hard cases as a second sign the hours were real. If the tickets that still rightly go up are being resolved faster than they were, the easy types that used to sit in front of them have genuinely gone. In that case, the requesters of the hard cases feel it. If they are no faster, the hours went somewhere, and the sentence had better say where. The tier-level view of whether the structure is working, with the specialist's queue as one of its signals, is in Measuring Whether Tiers Work. Shift-left's measurement is that view cut by type.

Reading the Four Together

The four numbers for a type are read as a pattern. And the patterns are few enough to put in a table, which is below. The short version: escalations down with reopens flat and a sentence beside the hours is a move that worked. Any other combination names its own problem. The reopened tickets, the enablement plan or the specialist's queue is where to look for that problem. Satisfaction for the type is a fifth number where the desk has it by type, the users' vote on the whole pattern. It is not usually available at type level on a small desk, and the scorecard leaves the cell blank rather than inventing it. If you run QuerySolv, escalation and reopen rates and satisfaction are on the standard reports already. So the scorecard's columns exist before the scorecard does. The work is the cut by type, which is a filter per type and an hour per quarter.

Per-Technician Numbers: Coaching, Not Ranking

Shift-left raises the front line's resolution counts by design, because types they used to hand up they now finish. That rise is the strategy's doing, not any individual's. A per-person count after a move is therefore mostly a measure of which technician was given which type. And a league table built from it would rank the desk's allocation decisions with people's names on them. The library's rule for per-technician figures applies in full. They are read for coaching, never for ranking, by the person's own lead in a one-to-one. They are never read from fewer than forty or so comparable tickets. The reasoning, and the care needed when a figure is cut by agent, is in Satisfaction by Category and, Carefully, by Agent. It is the same care here.

The per-person figure that does help is the escalation rate of a moved type for one technician against the type's rate for the front line as a whole. Read it with that technician, privately, after a quarter. A technician may send up twice the share of the others for one type. Usually, they have not been told the sentence, have met a case the others have not, or are applying the limit more carefully than the runbook asked. All three are conversations and none is a performance problem. Nobody has ever been coached into finishing more tickets by seeing their name at the bottom of a list. They have been coached into closing more.

The Honest Report Line

The report of one moved type is one line. And it reads something like: "The quote-screen fault now resolves on the front line in about four cases in five, with escalations down from forty-four a quarter to seven and reopens unchanged at one; the move freed roughly fifteen to twenty hours a quarter of senior engineering time after enablement, which went on the supplier case that has since removed the fault." Every figure in it is on the scorecard. Every one is a count or a range, and the last clause is the sentence. It is longer than "escalations down by four fifths". It is the line Esther sent after the review, because it survives both of Ingrid's questions. Ingrid, like most people who read desk reports, had learned to ask them.

The format the line goes into, with its one page and its three or four numbers chosen for the reader rather than the desk, is in The Monthly Report They Will Read. Shift-left earns a line on that page once a quarter, not a page of its own. Translate the step names before they leave the desk. "Resolved by the front line" is meaningful to Ingrid; "resolved at step three" is not. And "freed senior engineering time" is a thing the trading floor can picture while "specialist hours" is not. Do not claim what the desk cannot evidence. Offer no headcount conclusion, no money unless someone else supplies the rate, no "the business saved". The desk reports hours and what they bought, and lets the reader draw the next line.

Report the types that moved back up, in the same line format with their one-sentence reason, and the types that did not move at all. A scorecard with only successes on it will be asked for the rest, and the rest is where the credibility is. "Two types moved to the front line and held; one went back to the second line while the runbook is rewritten for the new client; two are waiting on console access from the supplier" is a report of a programme that is being run. A report with only the first clause is a slide.

A Short War Story: The Falling Line at Lockwood Transit

Lockwood Transit, a bus operator with a desk of four, reported its escalation rate falling for three consecutive quarters after a shift-left programme. The rate was drawn as a single line on a single slide, and the line was admired at each review. The front line had been told, in the programme's first week, that escalations were "what we're trying to get rid of". They had taken it to heart. By the third quarter they were handing almost nothing up, including the depot timetable faults the limit said to escalate. They were resolving those by restarting the timetable display and closing the ticket. The displays failed again within days each time. The depot managers raised fresh tickets rather than reopening, which kept the reopen figure flat and the escalation line falling. The operations director learned about the problem from a depot manager at a bus stop, who asked why the desk kept "fixing it the same way". The honest recount showed resolution time for the type had roughly doubled. The real reopen count, with the fresh tickets linked back, had tripled. The desk lead spent a fortnight rebuilding three quarters of figures and a year being asked whether the next number was "the slide one or the real one". An escalation rate reported on its own is a target. And a front line told that escalations are bad will stop escalating, which is not the same as starting to fix.

The Shift-Left Scorecard

The scorecard below is one page per quarter for the whole desk, with one block per moved type. The counting rules are on the same page so that next quarter's count is a reproduction rather than an argument. Section A is the rules, filled in once and changed only in writing with a date. Section B is one block per type, five rows of counts and one sentence. Section C is the types that moved back up or did not move, which are on the page on purpose. Section D is the one summary line per type that leaves the desk. Fill it in a couple of weeks after the quarter ends, so that the reopen windows have closed. Keep every quarter's copy, because the trend across copies is the thing a manager will eventually ask to see.

THE SHIFT-LEFT SCORECARD — ONE PAGE PER QUARTER
Desk: ______________________   Quarter: ____________   Counted by: ______________   On: ____________

A. COUNTING RULES (set once; changes dated here: ____________)
   A type's tickets are identified by: the fix, read from the resolving note (not the category alone)
   The resolving step is: the step of the person who wrote the resolving note
   An escalation is: any handoff from the front line to a higher step, including quiet reassignment
   A reopen is: a requester reply or a fresh same-fault ticket within ____ working days, linked back
   Hours are: hands-on people time per ticket, middle estimate from a sample of ____ tickets
   Enablement hours are: taken from the enablement plan's "actual" column, this quarter only
   Deflection is: low count (solved-it responses or the exact-question drop) to high count (whole drop
   after the control type's drift), from the deflection sheet
   Per-technician figures: coaching only, read privately, never below ____ comparable tickets

B. ONE BLOCK PER MOVED TYPE
   Type: _______________________________   Moved from: ____________  to: ____________  on: ________
   Tickets of the type this quarter: ____      (last quarter: ____   before the move: ____)
   Resolved at: specialist ____  second line ____  front line ____  self-service ____ to ____  prevented ____
   Escalations of the type: ____   (last quarter: ____   before the move: ____)
   Reopens of the type: ____       (last quarter: ____   before the move: ____)
   Queue-wide median first response this quarter: ____  (last quarter: ____)
   People time per ticket: before ____ min   now ____ min
   Hours freed this quarter: ____ to ____   minus enablement hours ____   = net ____ to ____
   What the hours bought, in the specialist's words: ______________________________________
   Satisfaction for the type, if available by type: ____   (otherwise leave blank)
   Pattern (from the reading table): ______________________   Action: ______________________

   Type: _______________________________   (repeat the block)

C. TYPES THAT MOVED BACK UP, OR DID NOT MOVE
   Type: ____________________  Back from ________ to ________  Reason (one sentence): ________________
   Type: ____________________  Did not move; waiting on: ____________________________________

D. THE LINES THAT LEAVE THE DESK (one per type, counts and ranges, ending with the sentence)
   "__________ now resolves on the ________ in about ____ cases in ____, with escalations down from ____
   to ____ a quarter and reopens ________; the move freed roughly ____ to ____ hours a quarter of
   ________ time after enablement, which went on ______________________________."

Esther filled in the first scorecard the week after the review. It took an afternoon because the counting rules had to be written before the counting could start. Ingrid's two questions were answered on it in two cells: reopens of the type, one, unchanged; what the hours bought, "the supplier case for the quote-screen fault". Two quarters later the supplier's fix went in and the type's column went to nine tickets. The trading floor's quote screen stopped being a thing anyone on the desk talked about. Maren has the five columns pinned above her desk. She says it is the only chart of her own work she has ever seen. And she would like it noted that the fourth column was hers.

Frequently Asked Questions