Justifying an Upgrade With Graphs Management Believes
You have done the work. The register says the file server is full in five months, the lead time is ten weeks, and the trigger date was last Tuesday. Now you need someone with a budget to agree. That person has never seen your monitoring system, does not know what a daily maximum is, and has twenty minutes. What you put in front of them decides whether the disk arrives before the pain or after it.
This guide is about that one page. You will learn how to choose the single graph that carries the argument. You will learn how to annotate it so it explains itself. You will learn how to draw the projection honestly enough that it survives a sceptical question. And you will learn how to write the sentence underneath that turns a line on a chart into a consequence someone cares about. There is a one-page template to copy and a list of the things that make a manager quietly stop believing a graph. And there is a story about a proposal that failed on its axis.
The arithmetic lives in the rest of this pillar. The forecasting guide produces the date and the range, and the lead time guide produces the trigger. This guide shows you how to present them.
In this guide
- One graph, one line, one limit, one date
- Choosing the graph
- Annotating it so it explains itself
- The honest projection
- The consequence sentence
- Before-and-after graphs
- The one-page upgrade proposal
- Making the graph
- Things that make management distrust a graph
- The proposal that failed on the axis
- Checklist
- Where to go next
One graph, one line, one limit, one date
A proposal that works has exactly one graph on it, and the graph has one line, one limit, and one date. The line is what the resource has done. The limit is where it stops working. The date is when the line reaches the limit. Everything else on the page exists to answer the question the graph raises, which is "so what do we do?" A reader who is shown one line crossing one limit at one date does not need to be persuaded of anything. They can see it.
The temptation is to show more, because you have more. You have the CPU graph, the memory graph, and the backup duration graph, and each supports the case. Resist. Two graphs halve the attention each one gets. Five graphs turn the page into a dashboard, and dashboards are for people who already know what they are looking for. Your reader does not. Give them the one graph that makes the decision obvious and keep the others in an appendix. (They will thank you for the appendix. They will not read it.)
If you cannot pick one graph, the proposal is not ready. Either you have not decided which resource is the real problem, or you have two proposals and should write them separately. A manager can approve one thing in twenty minutes. Two things take a meeting.
Choosing the graph
The right graph is the busy-hour peak of the resource, one point per day, over at least a year. It is not the five-minute samples from last week, however dramatic they look. A week of raw samples shows the reader a jagged mess with a bad afternoon in it. And the reasonable response is "that was one afternoon". A year of daily peaks shows the reader a line going somewhere, and the reasonable response is "when does that hit the top?"
The busy-hour peak is the highest reading in the day. And it is the honest choice for the same reason it is the honest choice in forecasting. That is because the resource fails at the peak, not at the average. Use the daily maximum for links, CPU, memory, and scopes. For disks, the daily maximum and the daily average are nearly identical, so either is fine. Whatever you choose, say which it is in the axis label, because "utilisation" on its own means nothing.
A year is the minimum because it is the shortest window that makes the trend believable. Anyone who has seen a summer knows use dips in it. A line drawn through six months will get the question "does that include the summer?" If you have less than a year, show what you have, write the number of months on the graph, and expect the question. Two years is better still, because the reader sees the same season twice and can tell the climb from the season.
Annotating it so it explains itself
A graph that needs you standing next to it is half a graph. Annotate it so that a reader who finds it on a desk understands the argument without you. Five annotations do the work.
- The limit, as a horizontal line, labelled with the number and the unit: "2,000 GB volume size" or "100 Mbit/s circuit". Without it the reader has no idea how close to the top the line is.
- The projection, as a dotted continuation of the trend from the last real data point to the limit. Dotted, not solid, so nobody mistakes the future for the past.
- The crossing date, as a vertical marker where the dotted line meets the limit, labelled "projected full" with the month.
- The trigger date goes in as a second vertical marker to the left of the crossing, labelled "order by" with the month. Add a bracket between the two labelled "lead time". This is the annotation that turns the graph into a request with a deadline.
- Events that explain steps, as small labels on the line: "archive migrated", "new site added", "cleanup". A step in the line without a label looks like noise or an error. The same step with a label looks like history.
Two smaller marks finish it. Put the date range of the data in the title or the corner. And put the source in small text underneath: "daily maximum from monitoring, fs01 D:, twelve months". Both answer "where did this come from?" before it is asked. A reader who can see the source relaxes; one who has to ask starts wondering what else is missing.
The honest projection
Show the range, not just the line. A single dotted line arriving at a single date looks precise. Precision invites disbelief, because everyone in the room knows the future is not straight. Draw the projection as a narrow wedge instead. Use the dotted line for the most likely path, and two fainter lines for the earlier and later crossings from the forecast range. Label the wedge "between March and June, most likely May". It is the same statement the forecasting guide taught you to make out loud, drawn.
The wedge does something a single line cannot: it makes the reader's objection part of the graph. When someone says growth might slow down, you point at the later edge and note that the trigger is set from the earlier one. When someone asks what if the new project lands, you point at the earlier edge. A range that already contains the objections is a forecast that has been argued with in advance.
Be equally honest about the data behind the wedge. If the fit rests on eight months, say eight months on the graph. If a step change means you fitted only the last five months, say that too, in the event label. The single most persuasive thing you can put on a capacity graph is a visible admission of what it does not know. Managers see optimistic charts every week. An honest one stands out.
The consequence sentence
Under the graph goes one sentence that says what the crossing means for someone who does not care about the resource. Not "the volume reaches capacity", which is the graph in words, but what stops: "At this rate the link is full in May; after that, afternoon file transfers to the depot stall and drivers wait for manifests." The first half is the date. The second half is the reason anyone should mind. The consequence sentence is the part of the page most readers will remember, so write it about their work, not yours.
Finding the consequence takes a minute of thinking about who uses the thing. A full file server stops the order export. A full DHCP scope means new visitors cannot get online in the lobby. A hot closet shortens the life of the switches in it and, on the worst afternoon, reboots them. Pick the one consequence the reader will recognise from their own week, and say it plainly. If the consequence is genuinely minor, say that too, and propose waiting. A proposal that sometimes says "not yet" is a proposal that gets believed when it says "now".
Before-and-after graphs
If you have done an upgrade before, you have the most convincing evidence there is: a graph of the last one. A before-and-after graph shows the old resource climbing to its limit. It shows the date of the upgrade as a vertical line, and the new resource with headroom on the far side. It proves three things at once. The forecast was right last time, the upgrade did what the proposal said, and the person proposing this one has been right before.
Keep the graph from every upgrade, with the proposal that preceded it, next to the register. The verify step of the planning loop produces the after-graph for free; all you have to do is not delete it. In a year you will have a small portfolio of kept promises. Nobody argues with the third graph in a row that turned out right.
The one-page upgrade proposal
Here is the page. The graph goes at the top, the consequence sentence under it, and the rest is this template, filled in and nothing added. Every field is one or two lines; if a field wants a paragraph, the proposal has grown a second decision and should be split.
UPGRADE PROPOSAL: <resource, plain name> Resource: <what it is, where it is, who uses it> Evidence graph: <title of the graph above; data range; source> Current use: <value> of <limit> (<percent>), busy-hour peak Trend: <+amount per month> over <n> months Projected full: between <early> and <late>, most likely <month> Consequence: <the one sentence: what stops, for whom> Lead time: <weeks>, measured from the last <n> purchases Order by: <month> (projected full minus lead time minus margin) Options: 1. <do nothing: what happens and when> 2. <shrink or shape: what it buys, how long, what it costs in effort> 3. <upgrade: what, to what size, headroom it gives in months> Recommendation: option <n>, because <one reason> Cost: <from the quote; one line; note if estimate> Decision by: <date, equal to the order-by date> Prepared by: <name>, <date>
Three of those fields carry more weight than the rest. "Options" must include doing nothing, with its date. That is because a reader who is not offered the option of waiting will invent it. A proposal that shows what waiting costs has already answered them. "Decision by" must equal the order-by date from the register, because a proposal without a deadline is a suggestion. And "Cost" is one line from the quote; if you do not have it yet, write "quote requested" and the date expected.
Tip: size the upgrade in months of headroom, not in units. "Two more terabytes" means nothing to the reader; "about three years at the current rate" is a decision they can weigh. Compute it with the same ruler: new headroom divided by monthly growth.
Making the graph
The graph can come from the monitoring tool or from a spreadsheet, and either is fine as long as the annotations end up on it. The monitoring tool is quicker and often draws a threshold line for the limit. Check the product documentation for how far back its graphs reach and whether it exports an image. The spreadsheet is slower but gives you full control of the projection and labels, and it works everywhere.
- Export a year of daily maxima as a CSV with two columns, date and value. Open it with dates in column A and values in column B, header in row 1.
- In column C, put the limit as a constant in every row:
=$F$1with the limit typed into F1. This becomes the horizontal limit line. - Extend column A below the last real date with future dates, one per day, far enough to pass the crossing. Leave column B blank for those rows.
- In column D, compute the fitted line for every row, past and future, with
=TREND($B$2:$B$366,$A$2:$A$366,A2)dragged down. This is the projection; the same formula on the real dates shows how well the line fits the past. - In column E, repeat step four using only the last six months of B and A; this gives the second edge of the wedge. Skip it if you only want one projection.
- Insert a line chart with A as the horizontal axis and B, C, D (and E) as series. Format B as a solid line, C as a dashed line, D as a dotted line, and E as a thin grey line.
- Set the vertical axis to start at zero and end a little above the limit. Label it with the unit. Label the horizontal axis with the date range.
- Add vertical lines or text boxes for the crossing date and the order-by date, and text labels on any steps. Put the source and data range in the chart title or a footnote.
- Export the chart as an image, put it at the top of the proposal page, and write the consequence sentence beneath it.
Save the workbook next to the proposal. After the upgrade, the same workbook with a year more data becomes the before-and-after graph.
Things that make management distrust a graph
Managers are not fooled by graphs as often as people assume. They see a great many, and they have learned the signs that a chart is trying too hard. Avoiding all of them is most of what "a graph management believes" means.
| Sign | What the reader thinks | Fix |
|---|---|---|
| No dates on the axis | "When was this?" | Date range in the title; dated ticks on the axis |
| No units | "Percent of what?" | Unit on the vertical axis and on the limit label |
| Vertical axis starts above zero | "That climb is exaggerated" | Start at zero, end just above the limit |
| Short or convenient window | "Why only these three months?" | A full year or more; state the range |
| Solid line into the future | "They are presenting a guess as data" | Dotted projection, clearly starting at the last real point |
| Single crossing date | "Nobody knows that precisely" | Show the range; name the most likely month |
| Unexplained step | "Was that a glitch?" | Label the event that caused it |
| No source | "Where did this come from?" | Source and collection method in small text under the graph |
| Five graphs | "Which one matters?" | One graph; the rest in an appendix |
The truncated axis deserves its own warning because it is the one people do by accident. Most charting tools pick the vertical axis range to fit the data. So a volume that has moved from seventy to eighty percent gets an axis from sixty-eight to eighty-two and looks like a cliff. The reader notices, and then reads every other element as a possible trick. Start at zero, always, even when it makes the climb look gentle. A gentle climb that reaches the limit in May is still a full disk in May.
The cherry-picked window is the one people do on purpose and regret. Choosing the three steepest months makes the case stronger, right up until someone asks for the full year and finds the steep months were the migration. The window you use for the proposal must be the window you used for the forecast, and both must be the longest you have. The monitoring data for management pillar covers the wider craft of reporting monitoring numbers upward; the rule here is simpler. Never show a manager a window you would not want them to see the edges of.
The proposal that failed on the axis
Tollgate Foods had a warehouse link that was genuinely running out. The administrator's proposal had the right forecast, the right lead time, and a quote. The graph was exported from a spreadsheet in a hurry. It had a vertical axis that started at sixty percent and a horizontal axis with no dates. It showed six weeks of five-minute samples with a spike on the Thursday a backup ran at noon. The finance director looked at it for ten seconds and asked what the spike was and when the six weeks were. The director put the proposal at the bottom of the pile with a note saying "come back with a year". The link filled four months later, in the middle of the seasonal peak. The same proposal was approved in a day with a year-long graph of daily maxima on it. The graph had been right the first time. It just had not looked it.
Checklist
- One graph on the page: busy-hour daily peaks over at least a year.
- Limit drawn as a labelled horizontal line with its unit.
- Projection dotted from the last real point; range shown as a wedge with the likely month named.
- Crossing date and order-by date marked, with the lead time bracket between them.
- Every step in the line labelled with the event that caused it.
- Vertical axis from zero; dates on the horizontal axis; source and data range on the graph.
- Consequence sentence under the graph, written about what stops for whom.
- Template filled in, with a do-nothing option, a cost from the quote, and a decision-by date equal to the order-by date.
- Upgrade sized in months of headroom as well as units.
- Workbook and graph saved next to the register for the before-and-after next time.
Where to go next
You now have the whole loop from Capacity Planning From Monitoring History in one place. You have the register from the basics guide, the date from the ruler, and the trigger from the lead time. You have the rows beyond bandwidth from capacity beyond bandwidth. And you have the page that gets the order signed. For the bandwidth case specifically, utilisation to purchase timing is the companion to this guide, with the same graph read for a circuit. Make the first proposal for whichever register row has the nearest trigger. Then keep the graph, because in a year it will be the before half of the most persuasive page you own.
Put this into practice with SysPeep Monitor
The graph at the top of the page is the hard part. It only exists if the monitoring system kept a year of history and can show it. SysPeep Monitor keeps long-term graphs and reports in one web console. It stores the history in SQLite, MySQL/MariaDB, or SQL Server. It lets you put a threshold on utilisation at the order-by level so the proposal is prompted by an email rather than a memory. Check the product documentation for how far back its graphs reach. The free trial installs on any Windows machine.
Download Free TrialLearn About SysPeep