Which marketing dashboard should a team build first?

Start with a recurring decision the team struggles to make. Define its owner, metric, denominator and source, then build the smallest view that can support that decision.

How should we measure an AI marketing workflow?

Compare accepted outputs, revision rounds, review time and failure types with a comparable manual baseline. Include input preparation and maintenance costs, not just draft generation speed.

Can AI explain why a marketing metric changed?

It can suggest hypotheses from the data you provide. Check those hypotheses against source data and changes in tracking, audience or campaigns before presenting a causal explanation.

Playbook · 2026-W45 · 07 · 11 · 2026 · 2 min read

Marketing dashboards worth building for 2027

By Eddie Espriella (Eduardo de la Espriella) · Experience and CV

Start with the decision a team needs to make. Five useful dashboard views for pipeline, content, follow-up, AI workflow quality and experiments.

A useful marketing dashboard changes a decision. Before choosing a chart, write down who will look at it, what they can change, and when that decision happens. For 2027, I would prioritize five views: pipeline quality, content contribution, follow-up reliability, AI workflow quality and experiment outcomes. This is a proposed operating approach, not a forecast of which tools will win.

1. Pipeline quality: where should the team focus?

Count qualified opportunities by source and cohort, then show stage progression, time in stage and reasons for disqualification. Define what qualified means with sales. Put the denominator beside every conversion rate. A high volume of enquiries can hide poor fit; a small source can look unusually strong when only a few opportunities have had time to progress. The decision is whether to change targeting, messaging or the handoff, not whether to celebrate a larger lead total.

2. Content contribution: what should we make next?

Pair the question each piece answers with meaningful actions: engaged visits, relevant enquiries and sales conversations where the content was useful. Keep attribution limits visible. An assisted touch is evidence of participation, not proof that one article caused a sale. Combine quantitative signals with questions from prospects and sales. The decision is whether to improve, repurpose, distribute or retire a piece. Avoid turning this view into a publishing-volume leaderboard.

3. Follow-up reliability: which promise is at risk?

After an event or campaign, track the next agreed action, its owner, due date, completion and response. Separate records needing clarification from actions ready to execute. Report overdue actions as a proportion of actions due, not as a count without context. A response is different from a booked meeting, and a booked meeting is different from revenue. This view helps the team fulfill promises without pretending that every contact belongs in an automated sequence. The free event follow-up guide provides a starting routine.

4. AI workflow quality: is the system reducing work?

Track accepted outputs, revision rounds, review time and failure types against a comparable manual baseline. Include the time spent preparing inputs, checking claims and maintaining the workflow. Faster draft generation is not a saving if the reviewer has to rewrite everything. Record exceptions such as unsupported claims, missing sources, inappropriate tone or actions blocked by permissions. The decision is whether to keep, narrow, redesign or stop a workflow. A small reviewed sample can be more informative than thousands of unclassified outputs.

5. Experiment outcomes: what did we learn?

Before launch, record the hypothesis, audience, primary metric, guardrails and decision date. Show the current result next to the baseline and the amount of evidence available. If a comparison is observational, label it that way. If the sample is too small or tracking changed mid-test, say so. A dashboard should make it possible to conclude that the result is inconclusive. The decision is to continue, stop or run a better test, rather than select whichever metric improved.

A minimum specification before you build

For each view, write a one-page brief: decision, owner, review frequency, metric definition, denominator, source, refresh delay, quality checks and the action each result could trigger. Use a simple table before adding charts. Add a source-health note when data is incomplete. Agree how long to retain personal information and who can see row-level records. Tool choice comes after those decisions.

Where AI belongs

An AI assistant can summarize verified figures, suggest questions and draft an explanation of a change. It should not silently redefine metrics, invent a cause, hide missing data or publish a recommendation without review. Supply the metric definitions and source tables; ask it to separate observations, hypotheses and proposed checks. Keep calculations reproducible outside the generated prose.

A practical first week

Pick one recurring meeting where a decision is currently difficult. Define one view with its owner, build it from existing data, and use it in that meeting. Record which decision it changed and what information was missing. Improve that view before building the other four. If the team cannot name a decision it supports, another chart is unlikely to fix the problem. The marketing AI audit can help identify a first workflow to evaluate.

Published 7 November 2026 · 2 min read · 60 words