Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Social Media Marketing › Social Media Management

Building a Social Media Performance Baseline

How to build a social media baseline from your own posts: fix definitions, export history, tag posts, find the typical range, mark outliers and put it to use.
Tutorialintermediate17 min readUpdated

Why a social report needs a record of its own

Someone may ask whether a post did well, whether the month was good or whether a provider's work is paying off. A report that lists figures cannot answer, because a figure has nothing to be read against until the brand holds a record of what its own posts have done.

The reporting guide's section on comparing against yourself puts the brand's own history, grouped so that like is compared with like, ahead of published benchmarks, and the guide to using industry benchmarks also puts that history first. The pillar's first ninety days put reporting in place ahead of the first post, so that there is a baseline to read against, and compare the account with it in the third month.

The reporting guide asks the brand to build the baseline first, and the pages around it assume the file exists. None of them builds it for individual posts. This tutorial does, and it stops there: it sets no targets, borrows no benchmarks and says nothing about how any platform distributes posts. It gives no posting frequency and no export fields, because those belong to each platform's help pages and to the brand's own history.

Before you start

  • A goal and a goal metric for each platform. The baseline is built on the metric each platform was given in the reporting guide, so settle those first if they are not yet written down.
  • A calendar with a row for each post, if the brand has one. The social content calendar tutorial keeps a row per post per platform with a pillar, a format and a result note. Where it exists, it supplies the list of posts and their tags from the day it started; where it does not, the list of posts is made by hand from the accounts and the tags go in the same way.
  • Access to each platform's own analytics, or to whoever holds it, because the export comes from there and a provider's dashboard is not the brand's record.
  • A place the brand owns. A spreadsheet in business storage is enough. Each step works in one, and none needs a named tool.
  • One named owner. As general practice, one person refreshes the file and answers for it, so that the history does not depend on whoever happens to remember.

The steps

  1. Write down the questions the baseline must answer

    Begin with what people ask, not with what the platforms export. The questions that come up include whether this post did well, whether this month is ahead of the earlier ones, whether a format is earning its slot and whether a provider's work is moving the goal metric. Each fixes what the file has to hold: the first needs a usual level for posts like it, the second needs the same for a month, the third needs a group for the format, and the fourth needs a record that predates the provider and sits in the brand's hands. Write the questions at the top of the file with the name of the person who asks each. A question the file cannot yet answer is not a failure; it shows what to record from now on.

    Leave two kinds of question out. Reply speed and moderation are read against the standards the brand set for itself, which the community management guide covers, so they have no place in a file of post results. And any comparison with other brands cannot be answered from the brand's own posts at all.

  2. Fix the metrics and note each platform's definition

    Take each platform's goal metric from the reporting guide, the KPI that follows from the goal the platform was given, such as reach for an awareness goal, with the supporting metrics that explain it, and write a plain-words definition beside each from the platform's own help pages, such as the Meta Business Help Centre cited in this page's sources. Keep each platform on its own sheet and do not add figures across platforms, because definitions differ; copy the same definitions into the footer of each report built on the file, and leave out any vanity metric that no decision would change.

  3. Export what the platforms still hold and start a monthly export

    Export what each platform still holds today, because each decides for itself how long it keeps history, and as general practice store the raw files untouched in the brand's own storage. Then start the monthly export the reporting guide asks for, in the same place, so that the record stops depending on a window nobody on the team controls; which fields and file formats a platform offers is for that platform's help pages to say.

  4. Build one row per post and tag it

    This table is what the rest of the file is made from. Make one row for each post on each platform, so that a post shared to two platforms has two rows, as in the calendar's own layout, and fill the columns from the calendar where one exists and from the platform's own analytics and the post itself where it does not.

    ColumnWhat goes in itWhy it is there
    Post and platformThe post's title or first line, and the platform it ran onA post is read only against posts on the same platform
    Date postedThe day it went outLets the file be cut by month, and lets a provider's start date be marked
    FormatOne value from the brand's format shortlist for that platformFormats are compared with themselves, not with each other's totals
    PillarThe content pillar the post was planned underShows which themes the audience rewards
    JobWhat the post was meant to do, such as start a conversation or earn a saveThe content hub reads results by job beside format and pillar
    SlotThe recurring slot it filled, where the calendar uses slotsLets a slot be judged apart from the topic that filled it
    ResultsThe goal metric and the supporting metrics, read from the platform's own analyticsThe figures each group is built from
    Campaign or paid pushA yes or no, with the campaign's name where there is oneAs general practice, keeps a post that had help from being read as bare content
    Link tagThe tag the post's link carried, if it had oneLets visits from social be separated from other sources
    ContextWhatever may explain the post, in a few words: a promotion, a piece of news, a platform change, a public holiday or a shift in who is watchingLets a later reading account for it
    Result noteThe one-line note the calendar already asks forWhere the placing in the eighth step is written

    Tag the way the social content hub recommends: each post carries a format, a pillar and a job, and results are read by those tags, not post by post. As general practice, three habits keep the tags usable. Write each tag's short list of allowed values before any history is tagged. Give each post one value for each tag, using the pillar it was planned under where it could fit two. And tag the back history by hand from the posts themselves, because the platforms cannot group posts by the brand's own categories unless the brand tags them. It is slow, and it is the step that gives each later step its value. Changing what a tag means afterwards is a change of method, which the ninth step deals with. As general practice, fewer tags give thicker groups, so add a tag only when a question from the first step needs it.

    The link tag follows the brand's naming convention so that social visits can be separated in analytics; the column only records which tag was used and does not replace the analytics. As general practice, read each post's figures at one agreed age after posting, so that a post read soon after it went out is not set beside one read long after; the age is the brand's choice and goes in the definitions sheet.

  5. Choose the groups to compare and the smallest group worth a figure

    A group is a set of like posts whose results are read together: a platform and a format, and where the posts allow it a pillar as well. Choose groups from the questions in the first step, not from all possible combinations of tags, because each extra tag splits the posts into thinner groups, and a thin group's figure moves for no reason except chance, which is the effect the guide to reading small numbers describes.

    QuestionGroup worth buildingIf the group is too thin
    Did this post do well?Platform, format and pillar togetherFall back to the platform and format group
    Is this format earning its slot?Platform and formatPool formats of the same kind on that platform
    Is this pillar worth its share of posts?Platform and pillar, across formatsShow the posts and their count, with any figure marked provisional
    Is the provider moving the goal?Platform and format, before and after the start dateShow the posts as they come, with any figure marked provisional

    The smallest group worth a figure is the brand's own choice, and the file records it with the reason. The small numbers guide rules out a universal number and a number borrowed from elsewhere, and sets the minimum from the team's own history. Apply that guide's method to the brand's own posts. A group under the minimum is pooled with a neighbour or shown with its count and any figure marked provisional, so that nobody reads a figure the data cannot carry as settled. Show the count beside each figure in the file.

  6. Work out the typical level and the ordinary spread for each group

    For each group, take the goal metric of each post in it and work out two figures. The typical level is the median: sort the values and take the middle one, or the mean of the two middle ones when the group has an even number of posts. The reporting guide gives the reason for using medians. As general practice, the ordinary spread is the range of the middle half of the group, from the value a quarter of the way up the sorted list to the value three quarters of the way up; where a position falls between two posts, take the value between them in proportion. Spreadsheets have functions for both, so name the function and the method in the file and the refresh repeats it. Work out the same two figures for each supporting metric, because a group's typical reach says nothing about its typical saves.

    Put the figures in one results table for each platform, with these columns: the group, the count of posts behind it, the typical level, the middle half, the same two figures with flagged posts left out (posts marked as outliers in the seventh step, so fill these two columns after that step), and a status of settled or provisional. As general practice, showing the figures with and without the flagged posts lets a reader see at a glance whether a few posts are carrying the group. A group under the minimum, whether new or restarted, is provisional until it holds it, as a control chart treats early limits.

    How far back a group reaches is the brand's choice. The reporting guide's own example reads a month against the earlier months' posts of the same format on the same platform, and a longer window steadies a small group while slowing the reading of a real change. The social media strategy template reads each platform's measures against the account's own median; this step is where that median comes from, one for each group so that like is read against like.

  7. Mark outliers and context, and keep them in the table

    As general practice, add an outlier column, mark any post that sits well outside its group's middle half, write the context beside it from the context column, and leave the row in the table, because a baseline that quietly drops its best and worst posts stops describing the account; how far counts as well outside is the brand's call, written down before the posts are read. The control charts framework explains why a point outside the usual range is a prompt to investigate and not a verdict, and why the rules for what counts as one are agreed before the results are read.

  8. Place the latest posts and the month against the baseline

    For each new post, find its group and compare its goal metric with the group's typical level and middle half. Write one of three placings in the result note column: below, inside or above, with the count of posts behind the group. The middle half holds only half of a group's posts, so a post just outside it is ordinary; only one well outside, or a run of posts on the same side, deserves a look. Before crediting or blaming the content, read the flag and the context columns, because a paid push, a campaign or a news event can explain a post on its own.

    For the month, set out each group's posts as counts: how many fell below, inside and above, beside the month's own typical level and the baseline's. Counts serve better than percentages here, as the small numbers guide says of tiny bases. A group under the minimum is listed with its count and its figures marked provisional. A month that holds a holiday or a seasonal peak is read with that written beside it, not against a month that held neither.

    To judge a provider, place the posts made since the start date against groups built from the posts made before it, with the start date marked in the file. A record that begins when the provider does cannot show what changed.

  9. Handle breaks, new accounts and changes of method

    When a platform changes what a metric counts, or the brand changes how it tracks or tags, mark the date as a break, following the guide to comparing periods, record it in the measurement change log, and restart the affected groups from that date, as the control charts framework restarts a chart after a break. A new account, or a group restarted after a break, has no history to build on: as general practice, label its figures provisional, and decide in the meantime from the business's own economics and what it records as it goes, as the guide to marketing before data describes.

  10. Use it in the weekly note, the monthly report and the calendar review

    The weekly note already says which posts stood out; with the file, standing out means above or below a group's middle half, written with the flag and the context beside it. The monthly report puts each platform's goal metric against its baseline, and the monthly marketing report template leaves room for a note beside each row of results, which is where the placings and the counts go. The reporting cadence playbook decides who sees which numbers and when; the baseline feeds the packs it names and adds none of its own. Where the brand also runs social listening, that tutorial's monthly note feeds the monthly social report and does not replace it, so what listening finds about a news event or a wider conversation can be entered as context.

    At the calendar review, read the placings by slot and by pillar, not post by post. A slot whose posts keep landing below their group's middle half, and a pillar whose posts land above it, are what the review acts on. One weak post pulls no format, and the review changes one thing at a time, which lets a later reading say which change mattered.

  11. Keep the file and refresh it on a fixed rule

    Keep the file in the brand's own storage with one named owner, so that a change of provider, of reporting tool or of staff does not take the history with it. As general practice, refresh it on a fixed day after each monthly export: add the new rows and their tags, recompute each group's typical level and middle half, and keep the previous figures beside the new ones so that drift shows. At each refresh, check which groups have crossed the minimum or fallen back below it, restart any group that a break has affected, and recompute from the new level after a lasting change, as the control charts framework advises. Write the refresh rule and the date of the last refresh at the top of the file.

Reading the result

The finished file gives two views. The first is the results table for each platform, with its groups, counts, typical levels and middle halves. The second is the post table with its placings. Three of the first step's questions are read from them, and each reading stops short of a verdict; whether a format is earning its slot is answered at the calendar review in the tenth step.

A post. Inside the middle half is ordinary for that group, and just outside it is ordinary too, since half of a group's posts lie outside the middle half. Well outside is a prompt to read the flag and the context, and then to ask whether anything in the post can be repeated that did not depend on the push or the news. One post changes no format.

A month. Read each group's counts below, inside and above against its range, with the context beside them. A month whose posts lean to one side of the range is worth a note; a month spread across it is a finding too, because no meaningful change is something a report can say. Frame the finding with what, so what and now what: what the groups did against their ranges, what that means for the goal metric once the context is allowed for, and what the owner will do and when it will be checked.

A provider. Read each group since the start date against the range from before it, and treat a group below its minimum as provisional. If the groups have been restarted by a break, say so in the reading.

The baseline says what is ordinary, not what is good enough. A group can sit steady at a poor level and read as normal, and whether a level is good enough is a question for the goal each platform was given. A rescore of the social media presence audit gives its reporting row a record to be read against; the row asks whether reports are tied to the account's goal and end in actions.

A toy library reads its own posts (illustrative)

Illustrative example — not a real result

A toy library reads its own posts

Hyvorsane Toy Library, an invented library that lends toys, games and puzzles to member families from one hall, posts short videos and photo carousels on two platforms. A new volunteer asks whether the how-to videos are working, and the only answer anyone has is that engagement is down.

The volunteer exports the year of results that both platforms still hold, saves the raw files untouched and builds one row for each post, tagged by format, pillar and slot. The library sets its smallest group worth a figure at 8 posts, a minimum it chose from how much its own groups moved in months when nothing changed, and it offers that as its own choice and not as a rule. The how-to videos on the video platform form a group of 11 posts. Views per post is the goal metric the library gave that platform, and the group's values are 880, 1,050, 1,200, 1,300, 1,400, 1,500, 1,600, 1,750, 1,900, 2,100 and 21,000. The typical level is the median, 1,500 views, and the middle half, taking a value between two posts in proportion, runs from 1,250 to 1,825. An average would read about 3,240, which is why the library does not use one.

The 21,000 was a video carried by a local news share. It sits far outside the middle half, so its row is marked as an outlier with the context beside it, and it stays in the table. With that post left out, the typical level is 1,450, so the group is not being carried by it.

Four new how-to videos then come in at 1,900, 2,300, 1,650 and 2,700 views. Three sit above the middle half and one inside it. Four posts are too few to call the format better, so the result notes read above, above, inside, above, each beside the group's count of 11, and the next review looks again.

Later in the year the video platform changes what counts as a view. The library marks that month as a break in its definitions sheet and its change log, restarts the video groups from that month and labels their figures provisional until each group again holds 8 posts. The flat statement that engagement is down gives way to a monthly note that says, for each format, whether it sits inside, above or below its own range, with the count of posts behind each answer.

The social media management hub shows where reporting sits in the routine for pages, community and scheduling, and this baseline is the record that reporting reads against.

Common mistakes

  • Building the file from monthly totals. Totals hide the groups; the baseline needs one row for each post, with its tags.
  • Skipping the export of history. Each platform decides how long it keeps history, so what is not exported is lost, and a provider's dashboard is not the brand's record.
  • Adding figures across platforms. Definitions differ, so each platform is read on its own and the definitions go in the footer.
  • Tagging after seeing the results. Tags chosen to rescue a weak post or flatter a strong one make the groups say what the team hoped.
  • Cutting groups too thin. Each extra tag leaves fewer posts in each group; show the count, and set the minimum from the brand's own history, not from somewhere else.
  • Using an average. A post that travelled flatters it; use the median and show the figure with and without flagged posts.
  • Dropping the outlier. Keep it in the table with its context, so that the file still describes the account.
  • Reading a break as a change in the business. Mark the date, record it in the change log and restart the group.
  • Treating inside the range as success. The baseline says what is ordinary, not whether it is good enough.
  • Letting the file lapse. A file nobody refreshes on its fixed day stops matching the account and loses its owner.

Frequently asked questions

How many posts does a group need before it has a typical level?

There is no universal number, and a figure taken from another account or a published source is a guess. Set the minimum from the brand's own history, by the method the small numbers guide gives. Until a group reaches it, show its count with any figure marked provisional, or pool the group with a neighbour. Write the minimum and the reason at the top of the file so that it can be revisited.

What should I do with a post that travelled?

Keep it in the table and mark it as an outlier, with the context beside it, such as a news event, a campaign or a paid push. As general practice, show the group's typical level with and without it, so that a reader can see whether a few posts are carrying the group; a median already holds steadier than an average when one post travels. Do not build a format or a plan on one post; the content guidance warns against following one viral post into a format the team cannot keep producing.

Can a new account build a baseline?

Not at once, because a baseline is made from the brand's own results and a new account has few of them. As general practice, start the table with the first post, record each post as it goes, label each figure provisional until the groups hold the brand's minimum, and decide in the meantime from the business's own economics and what it learns, not from published benchmarks, which cannot say whether a month was good. The same label applies to a group restarted after a break in a definition or in tracking.

Terms used on this page

Frameworks

Sources

  1. Meta Business Help Centre — Meta

Related