What this template is for
Use the specification whenever a dashboard will be built by someone other than the person who will read it, or will be read by more than one team. That covers most marketing dashboards: a weekly performance view for a leadership meeting, a channel page for a paid media manager, a pacing view for finance, a branch or market view for regional managers. A personal exploration that nobody else will quote does not need one; GA4 Explorations or a scratch report are the right place for that kind of work, and the governance tiers treat it as a draft.
The specification sits between three other documents. Above it, the KPI tree and the measurement plan decide which outcome the business is chasing and which drivers move it; a dashboard page sits at one node of that tree. Beside it, the KPI dictionary holds the definitions, and the specification refers to its entries rather than writing definitions afresh. After it, the dashboard governance guide covers the inventory, certification and review that keep a built dashboard trustworthy. The measurement plan and executive dashboard capstone practises the whole sequence as an exercise; this template is the working document for a real request.
Writing the specification is also the cheapest moment to say no. A request that cannot name a decision, or whose questions are already answered by an existing certified dashboard, should become a change to that dashboard or nothing at all. The specification exposes that before an analyst spends days building a copy.
The template
Copy the table into a document or a ticket and keep the field names. Fill it with the requester, not for them: the first two rows are theirs to answer, and the rest is a conversation between requester and builder.
| Field | What to write |
|---|---|
| Audience and decisions | Who opens the dashboard, how often, and the decisions they make with it, each as a sentence |
| Questions per page | One question per page in plain words, with the headline chart, the trend, the drivers and the breakdown that answer it |
| Metrics and definitions | Every metric on the dashboard with its key and definition from the KPI dictionary, and any metric that has no definition yet |
| Targets and comparisons | The target, prior period or year-on-year comparison beside each headline metric, and where the targets are stored |
| Data sources | Each source, the connector used, the credential mode, and any blend or join with its key |
| Filters and defaults | The date range and comparison it opens on, the filter controls offered, and the default state of each chart |
| Refresh and performance | How fresh the data must be for the decision, the freshness setting, and whether an extract or modelled table is used |
| Access and sharing | Who can view, who can edit, whether it is shared by invitation, link, schedule or embed |
| Owner and certification | The named owner, the certification tier sought, the review date and where the inventory entry lives |
Keep the specification to one or two pages. Mock-ups, a wireframe and sample queries go in an appendix, so the requester signs off the decisions and questions first and the layout second.
How to fill each section
Audience and decisions
Name people or roles, not departments, and say how often they open the dashboard. Then write each decision as a sentence: "decide on Monday whether any branch needs extra local spend this week". A dashboard serves decisions, and the dashboard design guide explains why a page that cannot name one fills up with charts chosen because the data was available. If the audiences split — a director who wants the answer and an analyst who wants rows — note both and plan separate pages rather than one page that serves neither.
Questions per page
Write one question per page, in words the audience would use, and under it the headline metric, the trend, the drivers and the single breakdown the viewer can act on. A second question gets a second page. This is the row that most changes what gets built: a request for "everything about bookings" becomes three or four pages, each with a reason to exist. The design guide covers chart choice for each comparison; the specification only needs to say which comparison each page makes.
Metrics and definitions
List every metric that will appear, with its key and the definition from the metrics layer or the KPI dictionary. Say which count is meant — a SessionA group of one user's interactions with a site or app in one visit, ended by inactivity., a user, a Key EventA GA4 event marked as important to the business, the property's measure of success. or a CRM record — because the same word in two tools usually means two counting units. Any metric without an agreed definition is an open item: the dashboard waits for the definition, not the other way round. Note known differences too, such as platform-attributed conversions against analytics-attributed ones, because the first question about a new number is why it does not match the platform.
Targets and comparisons
Every headline KPIA metric tied directly to an objective. needs something to be judged against: a target from the plan, the prior period, or the same period last year for a seasonal business. Choose deliberately and write the choice down, including how a partial period is compared. Targets need a home the dashboard can read, usually a small sheet or table keyed by period and the breakdown dimension; record where it lives and who updates it.
Data sources
Record each source, the connector, whether it is native or partner-built, the credential mode and whose account it runs on. The data sources and connectors guide explains the difference between owner's and viewer's credentials and why a source tied to one person's login stops working when they leave. Where two sources are blended — bookings from the booking system with spend from the ad platforms — write the join keyThe shared field, such as date and channel, a campaign ID or a click ID, on which rows from two data sources are matched in a blend or a query. and check that both sides share it; the blending data guide covers the join types and their traps.
Filters and defaults
State the date range the dashboard opens on and the comparison it shows, chosen to match the question: the last complete week for a weekly review, month to date for a pacing page. List the filter controls and keep them to the dimensions the audience actually cuts by. Defaults decide what the dashboard says to the viewer who touches nothing, which is most viewers, so specify sort orders and visible rows as well.
Refresh and performance
Ask how fresh the data must be for the decision, not how fresh it could be. Most marketing data changes once a day, and a weekly decision rarely needs intraday data. Record the freshness setting per source and whether an extract or a modelled warehouse table will carry the load; the performance and caching guide explains when an extract helps and what it cannot do.
Access and sharing
Write who views, who edits, and how the dashboard reaches them: named invitations, a restricted link, a scheduled email or an embed. Check the combination of credential mode and sharing before anything is shared widely; the sharing and embedding guide explains why owner's credentials with permissive link sharing exposes the owner's data to everyone with the link. Accessibility belongs here too: the audience includes people reading on phones and people who cannot rely on colour, so state that status will be shown with labels as well as colour and that every chart has a table behind it.
Owner and certification
Name one person as owner, the tier the dashboard is meant to reach, the first review date and the inventory entry. The dashboard governance guide sets out the certified, team and personal tiers and what promotion requires; a dashboard intended to be quoted in reviews should be specified as certified from the start, which means every metric reads from the agreed definitions.
Filled example
Below, the specification is completed for an invented business. The branches, figures and names are illustrative.
Ravensholt Opticians: weekly eye test bookings dashboard
| Field | Ravensholt's specification |
|---|---|
| Audience and decisions | Priya Lal, Head of Marketing, and the 12 branch managers, every Monday morning. Decisions: whether any branch needs extra local search or social spend this week; whether a channel's cost per booked test has drifted enough to raise at the Tuesday trading meeting; whether a branch's no-show rate needs a reminder-message change |
| Questions per page | Page 1: "Are we on track for eye test bookings this month?" (bookings scorecard against target, weekly trend over 13 weeks, drivers: sessions, booking rate, spend; breakdown by channel). Page 2: "Which branches are behind?" (sorted bar of bookings against branch target, table of branch, bookings, gap, no-show rate). Page 3: "What is each channel costing per booked test?" (scorecard per paid channel, trend, table by campaign) |
| Metrics and definitions | Booked eye tests (booked_tests): appointments created in the booking system, excluding staff test bookings and cancellations within 10 minutes. Booking rate: booked tests from the website divided by website sessions. Cost per booked test (cost_per_booked_test): paid media spend including agency fee divided by booked tests attributed to paid channels in the warehouse model. No-show rate: attended tests subtracted from booked tests, over booked tests, per branch. Open item: "attended test" has no dictionary entry yet; owner Callum Reid to add it before build |
| Targets and comparisons | Monthly booking target per branch from the trading plan, held in a Google Sheet keyed by month and branch, updated by finance on the first working day. Cost per booked test compared with the prior four weeks. Bookings also shown against the same week last year, because demand peaks before school terms |
| Data sources | Booking system export loaded nightly to BigQuery (fct_bookings); GA4 via the warehouse export for sessions; Google Ads and Meta cost via the warehouse connector (fct_spend_daily); targets sheet. All Looker Studio sources on the analytics service account with owner's credentials. Blend on date and branch code; the GA4 landing page is mapped to branch code in dim_branch_pages |
| Filters and defaults | Opens on the last complete Monday-to-Sunday week, compared with the previous week. Controls: region (North, South) and branch. Branch table sorted by gap to target, showing all 12 rows |
| Refresh and performance | Data needed by 07:00 Monday. Warehouse tables refresh nightly; Looker Studio freshness set to 12 hours. Pages read pre-aggregated weekly tables rather than event-level data |
| Access and sharing | View: branch managers and the leadership team by named Google group. Edit: Callum Reid and one analyst. Weekly PDF of page 2 scheduled to branch managers at 07:30 Monday. Link sharing restricted. Status labels ("behind", "on track") shown in text as well as colour |
| Owner and certification | Owner Callum Reid, Marketing Analyst. Target tier: certified, replacing the three branch spreadsheets. First review 30 days after launch, then quarterly. Inventory entry: "Certified – Bookings – Branch weekly" |
At the first review, branch managers were using page 2 every week and page 3 rarely; page 3 moved to the paid media team's own team-tier dashboard, and the certified dashboard kept the two pages its audience read.
Common mistakes
- No decision named. A dashboard specified as "all the booking data" becomes a gallery that nobody acts on. Start from the decisions and work back to the questions.
- Several questions on one page. The page grows until no one can find the answer. Give each question its own page and link them.
- Definitions written into the specification. A definition invented for one dashboard drifts from the dictionary within weeks. Reference the agreed definition and treat a missing one as an open item.
- Numbers with nothing to compare against. A headline metric without a target or comparison is trivia. Specify the comparison and where the targets live.
- A source on one person's login. When that person leaves, every chart goes blank. Record the credential and put production sources on a shared or service account.
- Fresher than the decision needs. Intraday refresh for a weekly decision costs speed and quota for nothing. Match freshness to the decision.
- Sharing decided after the build. Owner's credentials plus an open link can expose more than intended. Settle access in the specification, before the link exists.
- No owner or review date. An unowned dashboard drifts into the sprawl the governance guide describes. Name the owner and the first review before launch.
The monthly report that often draws on a dashboard like this, and the other planning documents, are in the templates library.
