Home › Topics › Automating Routine Reports › Reports People Read

Reports People Actually Read: Design, Narrative, and Feedback


You can automate a report perfectly, with correct numbers, on time, to the right list, and still have nobody read it. Open rates are not the measure. The measure is whether the reader could tell you, an hour later, what the report said. Most monitoring reports fail that test, and not because the numbers are wrong. They fail because they are data dumps rather than answers. They also fail because they show one snapshot with no direction, or because a graph dips and nobody says why. And they fail because the layout changes just often enough that the reader never learns where to look.


Readable reports are a craft with a short list of rules, and every rule is something a script can do or a template can enforce. This is not about making reports pretty. It is about making them answer the question the reader opened them to ask, in the first sentence. Reports need enough context to be trusted and enough consistency to be skimmed. I have sent the other kind. It was thorough.


This guide covers the rules. Those are to lead with the answer, show trends over snapshots, annotate the dips, and keep the layout constant. They also call for translating numbers into plain language, and one page for anyone senior. Then the guide turns to the part most teams never do: asking whether the report is still useful, and pruning the ones that are not. There is a before-and-after rewrite of a report page and a design checklist.


In this guide

  • Why reports go unread
  • Lead with the answer
  • Trends over snapshots
  • Annotate the dips
  • Consistency: the same layout every time
  • Plain-language numbers
  • One page for anyone senior
  • Asking for feedback deliberately
  • Pruning unread reports
  • Before and after: a report page rewritten
  • The design checklist
  • Where to go next

Why reports go unread


A report competes for attention with everything else in the reader's morning. The reader does not open it to learn. They open it to answer one question — "is anything wrong?", "did we meet the target?", "do I need to do something?" They give it about as long as it takes to find that answer. If the answer is in the first line, they read the rest, because they now trust the report to be worth their time. If the answer is buried under a table of interface counters, they close it, and next month they do not open it at all.


Everything in this guide follows from that. The report is not a record of what the monitoring system knows. It is an answer, supported by evidence, delivered in a shape the reader has learned to navigate. The evidence and the navigation are what make the answer believable; the answer is what makes the report worth opening.


Lead with the answer


The first sentence of the report says whether things are fine. It gives the verdict, not "this report covers the period" — the reader knows that from the subject line. For example: "All monitored services met their availability targets this month; one capacity risk needs a decision by the end of next month." Or: "Two services missed their target because of one storage failure on the ninth; fixed, detail below." Everything that follows supports that sentence.


This is harder for an automated report than it sounds, because the verdict requires judgment. The practical compromise is a two-part headline. The script generates a factual line from the numbers. It says how many services met target, how many did not, and whether any capacity figure crossed its warning line. A person adds one sentence of interpretation before it is sent, if one is needed. Most months, the generated line is the whole headline. When it is not, that single human sentence is the most valuable thing in the report.


Put the headline in the email subject too, as far as it fits. A subject that says "Monthly summary: all targets met" gets the answer to a reader who never opens the message. That is a success, not a failure.


Trends over snapshots


A single number cannot be judged. An illustrative example: "Disk usage on the file server is at 78 percent" invites the question "is that bad?" and gives no way to answer it. "Disk usage is at 78 percent, up from 71, 66, and 62 over the previous three months" answers it immediately. The trend is steady, the pace is about four to six points a month, and the warning line will be crossed within a quarter. The same figure, with context, becomes a decision.


So every key number in the report appears alongside its recent history — this period and at least the previous three. Where a target exists, the number also appears alongside the target. In a table, that is three or four extra columns. In a chart, it is the difference between a bar and a line. The rule applies with particular force to availability, where a single month's percentage hides whether things are improving or slowly getting worse. It also applies with particular force to capacity, where the trend is the information and the current value is almost incidental.


The other extreme: a year of history for every number is a data dump again. Three to six periods is enough to show direction. If someone needs more, that is what the archive and the dashboard are for.


Annotate the dips


A graph with a dip in it raises a question, and an unanswered question erodes trust. The reader sees the availability line drop in the second week and wonders: was that an outage? Did they know about it? Is it fixed? If the report does not say, the reader either asks (costing everyone time) or assumes the worst (costing you credibility). Either way, the graph has created work rather than answering a question.


The fix is the annotation: a short note attached to the dip. It goes in the chart if the tooling allows, or as a numbered footnote beneath it if not. "Dip on the ninth: storage controller failure, forty minutes, resolved, incident reference below." "Increase from the fifteenth: new branch site added to monitoring." "Gap on the twenty-second: planned maintenance window, excluded from availability by agreement." Three causes — the maintenance window, the incident, and the new site or device — account for most unexplained features in a monitoring graph. All three are things the monitoring system or the ticket system already knows about. A script can pull the period's maintenance windows and incident records and list them under the chart; a human adds anything else.


Annotations do a second job: they show the reader that you saw the dip too. A report that explains its own anomalies reads as coming from someone in control. A report that does not reads as a printout.


Consistency: the same layout every time


Readers learn a report the way they learn a route to work. After three or four issues they know the verdict is at the top and the scorecard is the first table. They know the capacity chart is halfway down, and the decisions are at the end. They stop reading and start looking, which is faster, and they notice changes in the numbers precisely because the layout has not changed. Every time the layout does change, the reader has to read again from scratch, and some will not bother. That change might be a section moving, a column being added, or the chart switching from bars to lines.


This is the same principle that makes a good dashboard work. The value is in the reader's learned map of where things are, not in any single panel. If you are designing dashboards as well as reports, the dashboards pillar treats that principle in depth; for reports, the rules are simple. Fix the section order and keep it. Keep the same charts in the same positions with the same axes. Add new sections at the end, never in the middle. And when a change is necessary, announce it in that issue's headline so the reader knows to re-read.


Automation helps here more than anywhere. A hand-built report drifts because a human rebuilds it and improves it a little each time. A generated report is identical in shape every period by construction. And the temptation to "just move that chart up" has to survive a code change, which is exactly enough friction.


Plain-language numbers


Some numbers mean nothing to most readers in their native form. Availability is the classic case. A percentage with several decimal places is precise and honest, but it does not land. The difference between two figures that look almost identical can be a good month against a bad one. And a reader outside operations cannot tell. The translation that lands is time lost: "the email service was unavailable for three hours and ten minutes this month, all of it during one incident." Give the percentage as well, for the record and for the contract, but lead with the hours. (Getting the underlying figure right, so the hours are honest, is the subject of honest availability math.)


The same translation applies elsewhere. Bandwidth utilization becomes "the internet link was busy enough to slow things down for about two hours on most weekday afternoons." Disk capacity becomes "at the current rate, the file server fills in about three months." Alert counts become "the team was interrupted about a dozen times, four of them at night." None of these replaces the number; each sits next to it and explains what it means for the reader.


One rule for the translation: it must be derivable from the data, not decorative. "About three months to full" comes from the trend; "the service was unavailable for three hours" comes from the downtime minutes. If a plain-language line cannot be traced to a figure in the report, it is opinion, not data, and a sharp reader will notice.


One page for anyone senior


Anyone above the ops team gets one page. This is the rule that forces every other rule. A page is not long enough for a data dump, a snapshot without context, an unexplained chart, or a layout that wanders. It is exactly long enough for the essentials: a verdict, a scorecard with trends, and one or two annotated charts. Those essentials also include a few one-line incidents and the decisions being asked for. If the page is too short for what you want to say, the answer is a link to the detail, not a second page.


The one-page constraint also protects the reader from the report's author. Every report writer believes their extra section is the important one. The page does not care. It is a fair and impersonal editor. Readers can feel the difference between a report that was cut to fit and one that was allowed to grow.


Tip: read your report on a phone before you send it. Senior readers open a surprising share of email that way. A report that needs pinching and scrolling to find the verdict has failed the first-sentence test. That holds regardless of how it looks on a large monitor. If the headline and the scorecard are readable on a phone screen without zooming, the design is probably right.


Asking for feedback deliberately


Reports get no feedback by default. Silence is ambiguous: it might mean "perfect, nothing to add," or it might mean "I stopped reading this a while ago." Since you cannot tell the two apart, you have to ask. Asking works best when it is built into the report rather than as an occasional survey.


The lightest version is a question at the end of every issue, the same one each time. The wording is: "Is there a number you wanted that is not here, or one here that you never use? Reply and tell me." Most issues get no reply, which is fine. The question costs one line and, over a year, produces a handful of replies that each improve the report. Vary the question occasionally — "was the headline this month accurate from where you sit?" Reply to every answer, so answering is seen to be worth it.


The heavier version is a quarterly "still useful?" check: a short, direct message to each named reader asking three things. Do you still open this? What did you last do because of it? What would you remove? The second question is the one that matters, because it separates reports that are read from reports that are acted on. A reader who cannot name a single action in a quarter is telling you, kindly, that the report can shrink or go.


At Northgate Retail the IT team sent nine scheduled reports and had never asked anyone about them. The first quarterly check found named actions behind three: the capacity page, the weekly status, and the client statement. The skip test on the other six produced one query, for a single number that is now a line in the weekly status. The team now sends three. The other six are missed by nobody, which the team has decided to take as feedback.


Pruning unread reports


When the feedback says nobody acts on a report, or when it says nothing at all, run the skip test. Leave out one cycle and see who asks. Tell your own manager first, so the silence is an experiment and not a lapse. If nobody notices within a full period, that is your answer, and you retire the report formally. In that case, announce that it has ended, archive the back issues, and remove the job. If someone does notice, you have found the real reader. Often, in that case, the real report is smaller than the one you were sending. That smaller report is one number, one line, folded into something they already receive.


Pruning is not a one-off. Reports accumulate, and every one you keep costs the attention of everyone on its list. The full method for deciding which reports to keep, merge, or delete is in which reports deserve automation. The skip test belongs here because it is the most honest form of feedback there is.


Before and after: a report page rewritten


Here is the top of a monthly management report, first as commonly produced, then rewritten by the rules above. Both use the same data; every figure is an illustrative placeholder, and the point is the shape, not the numbers.


Before

INFRASTRUCTURE MONTHLY REPORT - PERIOD YYYY-MM

1. Introduction
This report presents the monitoring statistics for the reporting period.

2. Availability
  Service            Availability
  Email              99.71%
  File server        99.94%
  Intranet           100.00%
  Internet link      99.88%
  VPN                99.97%

3. Interface statistics
  core-sw-01 Gi1/0/1  in: 412,338,101,220 bytes  out: 219,004,558,911 bytes  errors: 0
  core-sw-01 Gi1/0/2  in: 388,120,551,004 bytes  out: 190,882,113,776 bytes  errors: 14
  (continues for 46 more interfaces)

4. Alerts
  Total alerts this period: 217
  (table of all 217 alerts follows)

The reader has to work out from a table of percentages whether the month was good. They have no idea whether the email figure is better or worse than usual. They are shown forty-eight interfaces they do not know, and are invited to scroll through every alert. Nothing here is false. Almost none of it is useful to this audience.


After

MONTHLY SERVICE SUMMARY - YYYY-MM   (illustrative figures)

Four of five services met their targets. Email missed by a small margin
because of one storage incident on the 9th (fixed). One decision needed:
file server storage, see below.

SCORECARD               target     this month   previous three months
  Email                 99.9%      99.71%       99.95  99.93  99.96
                                   2 h 05 m lost, all in one incident
  File server           99.9%      99.94%       99.91  99.95  99.93
  Intranet              99.5%      100.00%      99.98  100    99.97
  Internet link         99.5%      99.88%       99.90  99.85  99.91
  VPN                   99.5%      99.97%       99.96  99.98  99.97

CAPACITY
  File server storage:  78% used, up from 71 / 66 / 62.
  At this rate it is full in about three months.

INCIDENTS USERS NOTICED
  9th, 2 h 05 m: email unavailable, storage controller failure. Resolved;
  controller replaced under warranty. Detail: ticket reference in ops report.

DECISION NEEDED
  Approve additional file server storage this month, so it can be installed
  before the projected fill date.

Full detail: ops weekly report and dashboard link. Something missing from
this page, or something you never use? Reply and say so.

The rewrite says whether the month was fine in the first line. It shows each figure against its target and its recent history. It translates the one miss into hours lost, explains the dip, and expresses capacity as time to full. It asks for a decision, points to detail rather than including it, and ends with the feedback question. It is shorter, and it took the same data.


The design checklist


Apply this to any report before it goes on a schedule, and again whenever someone asks to add a section.


#CheckPasses when
1Verdict firstThe first sentence says whether things are fine, and the subject line carries it too
2Trends, not snapshotsEvery key number shows this period, the previous three or more, and its target
3Dips annotatedEvery visible anomaly in a chart has a note: maintenance, incident, or change
4Layout constantSection order, chart positions, and axes match the previous issue; changes are announced
5Plain-language numbersAvailability shown as time lost; capacity as time to full; every translation traceable to a figure
6One page above opsManagement and client versions fit on one page; detail is linked, not included
7Readable on a phoneHeadline and scorecard legible without zooming
8Decisions explicitAnything the reader must decide is stated as a request with a date
9Feedback question presentThe report ends by asking what is missing or unused
10Quarterly usefulness check scheduledA reminder exists to ask each named reader what they last did because of the report
11Skip test on suspicionReports with no named action in a quarter are skipped for a cycle to see who asks
12Nothing decorativeEvery line either answers the reader's question or supports the answer

The checklist is short on aesthetics. Fonts, colors, and logos matter far less than whether the verdict is first and the trend is visible. A plain-text report that passes the twelve checks will be read; a beautifully styled one that fails the first three will not.


Where to go next


Design is the last step in the chain, and the one that decides whether the earlier steps were worth it. Lead with the verdict, end by asking whether anyone still needs it, and the reader can tell you an hour later what it said. The chain itself is the rest of this pillar. The guide to which reports deserve automation decides what exists. The guide to report templates per audience gives each reader the right depth. The guide to scheduled delivery gets the report to where it is read. And killing the hand-built spreadsheet replaces the manual ritual underneath it all. They live together under Automating Routine Reports.


Put this into practice with SysPeep Monitor

A readable report starts with a consistent source of numbers. The trends and annotations above are easiest when availability, capacity, and alert history are already in one place. SysPeep Monitor keeps live dashboards and reports in a single web console and can email reports to the people who read them. Check the product documentation for the report types and scheduling options. The free trial installs on any Windows machine in minutes.

Download Free TrialLearn About SysPeep