KPI Design

Thinking Tools6.4 Conducting Follow-Up and Taking FeedbackAlso known as Input and output metrics

Draws on the management by objectives of Peter F. Drucker (1954), the objectives and key results of Andy Grove (1983), the Balanced Scorecard of Robert S. Kaplan and David P. Norton (1992), and Amazon's input metrics (Colin Bryar and Bill Carr, 2021).

Pairs each result metric with the input drivers a team can move and the guardrails that must not slip, then ties a decision to every threshold.

Build a support team's measure set from an output, an input, a guardrail, and a decision

It opens on the set the team had, and the lines under the rows say what each choice invites. Pick one option per row until nothing is flagged: the objective is customers' problems solved quickly and staying solved. Illustrative.

Output: the result

Input: what the team moves

Guardrail: must not slip

When the output misses

  • Nothing flagged in this combination.Tickets closed counts activity: it rises when agents close early and customers come back.Not an input: the team cannot move renewals directly, and the number arrives months after the work.No guardrail: nothing on the page shows what the team gives up to hit the output.Open to gaming: close on the first reply, let the customer reopen, and the number improves every week.A threshold with no decision: the miss is reported, and nothing changes.A working set: a result, a driver the team moves weekly, a guard on the cheap route to the result, and an action tied to the miss. Add satisfaction as a second guardrail.A working set: a result, a driver the team moves weekly, a guard on the customer's experience, and an action tied to the miss. Add reopen rate as a second guardrail.
  • Nothing flagged in this combination.Tickets closed counts activity: it rises when agents close early and customers come back.Not an input: the team cannot move renewals directly, and the number arrives months after the work.No guardrail: nothing on the page shows what the team gives up to hit the output.Open to gaming: close on the first reply, let the customer reopen, and the number improves every week.A threshold with no decision: the miss is reported, and nothing changes.A working set: a result, a driver the team moves weekly, a guard on the cheap route to the result, and an action tied to the miss. Add satisfaction as a second guardrail.A working set: a result, a driver the team moves weekly, a guard on the customer's experience, and an action tied to the miss. Add reopen rate as a second guardrail.
  • Nothing flagged in this combination.Tickets closed counts activity: it rises when agents close early and customers come back.Not an input: the team cannot move renewals directly, and the number arrives months after the work.No guardrail: nothing on the page shows what the team gives up to hit the output.Open to gaming: close on the first reply, let the customer reopen, and the number improves every week.A threshold with no decision: the miss is reported, and nothing changes.A working set: a result, a driver the team moves weekly, a guard on the cheap route to the result, and an action tied to the miss. Add satisfaction as a second guardrail.A working set: a result, a driver the team moves weekly, a guard on the customer's experience, and an action tied to the miss. Add reopen rate as a second guardrail.
  • Nothing flagged in this combination.Tickets closed counts activity: it rises when agents close early and customers come back.Not an input: the team cannot move renewals directly, and the number arrives months after the work.No guardrail: nothing on the page shows what the team gives up to hit the output.Open to gaming: close on the first reply, let the customer reopen, and the number improves every week.A threshold with no decision: the miss is reported, and nothing changes.A working set: a result, a driver the team moves weekly, a guard on the cheap route to the result, and an action tied to the miss. Add satisfaction as a second guardrail.A working set: a result, a driver the team moves weekly, a guard on the customer's experience, and an action tied to the miss. Add reopen rate as a second guardrail.
  • Nothing flagged in this combination.Tickets closed counts activity: it rises when agents close early and customers come back.Not an input: the team cannot move renewals directly, and the number arrives months after the work.No guardrail: nothing on the page shows what the team gives up to hit the output.Open to gaming: close on the first reply, let the customer reopen, and the number improves every week.A threshold with no decision: the miss is reported, and nothing changes.A working set: a result, a driver the team moves weekly, a guard on the cheap route to the result, and an action tied to the miss. Add satisfaction as a second guardrail.A working set: a result, a driver the team moves weekly, a guard on the customer's experience, and an action tied to the miss. Add reopen rate as a second guardrail.
  • Tickets closed counts activity: it rises when agents close early and customers come back. (out: Tickets closed)
  • Not an input: the team cannot move renewals directly, and the number arrives months after the work. (inp: Renewal revenue)
  • No guardrail: nothing on the page shows what the team gives up to hit the output. (grd: None)
  • Open to gaming: close on the first reply, let the customer reopen, and the number improves every week. (out: Tickets closed, grd: None)
  • A threshold with no decision: the miss is reported, and nothing changes. (dec: Discuss at the review)
  • A working set: a result, a driver the team moves weekly, a guard on the cheap route to the result, and an action tied to the miss. Add satisfaction as a second guardrail. (out: Solved on first contact, inp: Help articles rewritten, grd: Reopen rate, dec: Two agents to triage)
  • A working set: a result, a driver the team moves weekly, a guard on the customer's experience, and an action tied to the miss. Add reopen rate as a second guardrail. (out: Solved on first contact, inp: Help articles rewritten, grd: Customer satisfaction, dec: Two agents to triage)

An example

The support team at a scheduling software company was measured on tickets closed per agent, and the number had risen every month for a year, along with complaints. A sample of a hundred tickets showed why: agents were closing tickets on a first reply, and one in five came back reopened, often twice.

The support director rebuilt the measures around the objective in plain words: customers get their problem solved quickly and stay solved. The output became the share of issues solved on first contact. The input, something the team could move directly every week, was the number of help articles rewritten for the twenty most common issues. Two guardrails sat beside them: reopen rate under eight percent and satisfaction above 4.3 out of 5. Each measure carried a decision. If first-contact resolution fell below seventy percent two weeks running, two agents moved to triage; if reopens passed eight percent, a lead reviewed that week's closures.

Within a quarter, first-contact resolution rose from 58 to 74 percent, and reopens fell to six percent.

Why it works

A single number aimed at a result invites people to move the number by the cheapest route, which is often not the route to the result. Pairing the output with guardrails makes the cheap route visible: any gain bought with reopens or unhappy customers shows up on the same page.

The input metric fixes a different problem. Outputs arrive late and are moved by many things at once, so a team watching only the output knows it is behind and not what to do. An input the team controls, such as rewriting the articles behind the commonest issues, gives it something to act on this week.

Tying a decision to every threshold is the test of whether a measure belongs on the dashboard at all. If nobody would do anything different when it crossed the line, it is decoration.

The thinking behind it

Peter F. Drucker Peter F. Drucker, the Austrian American consultant, educator, and author whose books helped found management as a discipline, popularized management by objectives and self-control in The Practice of Management (1954). Managers and their superiors agree on measurable objectives, and the manager then steers their own work against them. Starting a measure set from the objective in plain words, before any metric is chosen, is his order of operations.

Robert S. Kaplan Robert S. Kaplan, a Harvard Business School professor, created the balanced scorecard with David P. Norton, introduced in The Balanced Scorecard: Measures That Drive Performance (1992) in the Harvard Business Review. It pairs financial measures with customer, internal process, and learning and growth measures, so managers can see what drives future results instead of only what has already happened, the same reasoning that sets guardrails and inputs beside an output.

Colin Bryar Colin Bryar, a longtime Amazon executive, described the company's management practices with Bill Carr in Working Backwards (2021), among them separating controllable input metrics, which a team can move directly, from the output metrics such as revenue that they are meant to move. The book is a practitioner's account of one company's methods, and the input metric in this tool comes from it.

How to run it

  1. State the objective in plain words. Write what good looks like for the customer or the business, before naming any metric.
  2. Pick the output closest to it. Choose the one result measure that moves when the objective is met, such as issues solved on first contact.
  3. Add inputs the team can move. Choose one or two measures of work the team controls this week that should drive the output.
  4. Add at least two guardrails. Name what must not get worse as the output rises: quality, cost, or customer experience.
  5. Define each measure exactly. Write the calculation, the data source, and how often it is read.
  6. Tie a decision to every threshold. Say what happens, and who does it, when each measure crosses its line.
  7. Ask how the set could be gamed. Imagine winning on the numbers while losing on the objective, and add a guardrail for each route.

Where it goes wrong

The set is all output. The dashboard shows revenue, churn, and satisfaction, and the team learns each month that it is behind with no idea what to change. Add an input the team controls and review it weekly.

The guardrails are missing or decorative. A measure pushed hard degrades whatever nobody watches, and the damage shows months later. Ask what a team optimizing this number would sacrifice, and put that on the page.

Thresholds carry no decisions. Numbers are reported in green and red at every review, and nothing happens when one turns red. Write the action and the owner beside each threshold, or take the measure off the dashboard.

When to use something else

  • Goodhart-Aware Metric Selection. When the main worry is that a measure will stop meaning anything once people are rewarded on it.
  • OKRs. When the task is setting a quarter's ambitious goals across teams, with key results graded at the end.
  • Construct Validity. When the question is whether a measure captures the thing it claims to measure at all.

Where it sits in SOLVE eX

Go to the source

  • The Practice of Management, 1954
  • High Output Management, 1983
  • The Balanced Scorecard: Measures That Drive Performance, Harvard Business Review, 1992
  • Working Backwards, 2021

Back to the catalog