How to Run a Campaign Post-Mortem With AI: A Worked Example

TL;DR

• Most post-mortems fail because the data-assembly sprint takes so long that the team moves on—not because anyone lacks curiosity.
• A 6-step framework with explicit AI and human lanes cuts a post-mortem from a half-day to under 90 minutes while producing deeper diagnostics than manual spreadsheet work.
• Databox Genie identifies which channels or sources are driving any drop in pipeline, names them in highlighted callouts, and quantifies the dollar impact so the post-mortem meeting starts with specifics instead of guesses.
• AI surfaces patterns humans miss, but it cannot access deal-room context, staffing realities, or the reasoning behind mid-campaign creative pivots.
• Action items without named owners and real deadlines are why most post-mortems produce no behavior change.erned metric definitions, computation that runs in tested code rather than in a language model, and a system honest enough to say when it does not know.
• A marketing leader planning next year’s spend should track how much it costs to defend a number, not how fast the team can get one.

Introduction

A marketing director sits down ten days after her campaign closed. Six browser tabs are open: LinkedIn Ads, HubSpot, GA4, Mailchimp, an attribution spreadsheet, and a blank doc that is supposed to become the post-mortem narrative. The meeting is in two hours. She knows something broke in the middle of the funnel (pipeline came in below target), but she cannot prove where or why until she reconciles numbers across all six sources. She will spend the next 90 minutes copy-pasting data and still walk into the room without a clear story.

The post-mortem fails because nobody had time to find it.

Teams either skip the review or run one that produces a document nobody reads. 88.5% of marketing teams already use AI for content and strategy work (per Databox’s The Role of Generative AI in Marketing research), but most treat AI as a content production tool, not an analytical one. The campaign post-mortem is where those change.

What follows: six steps, a full worked example showing what Databox Genie actually returns, and the prompts to use. Under 90 minutes, end to end.

What a Campaign Post-Mortem Actually Is (and Why Most Teams Do It Wrong)

A campaign post-mortem is a structured review comparing what a campaign was designed to achieve against what it actually produced, with the explicit goal of changing future behavior. Not documenting what happened, but changing what happens next.

Two failure modes dominate. The post-mortem gets skipped because a new campaign is already in motion and the data is scattered across tools nobody wants to reconcile. Or the review runs but produces a deck nobody references and action items assigned to “the team” with no deadlines. Same root cause: the cost of running a thorough review is high enough that teams skip it or rush through it.

The blameless framing is non-negotiable. The meeting’s job is to understand process failures, not individual ones. When the team trusts the format, they bring honest data. When findings get sharper every quarter, the post-mortem library becomes proprietary intelligence no competitor can replicate.

Before You Start: The 15-Minute Data Pack

The data sprint kills more post-mortems than any other factor. Two hours of tab-switching, and the leader still walks in without a connected story. Cutting that to fifteen minutes requires knowing which six sources matter and querying them simultaneously instead of one at a time.

Connect your ad platforms, CRM, GA4, and email tool to Databox once; from then on, every post-mortem starts from a single query across all of them. Before any query, confirm UTM integrity—a clean UTM chain is the prerequisite for any funnel story.

The six sources:

1. Ad platform analytics: impressions, CPM, CPC, CTR, ROAS by channel and creative.
2. Paid roll-ups: total spend, CPL, cost per MQL.
3. GA4: sessions, goal completions, landing page conversion.
4. CRM: leads, MQL-to-SQL rate, pipeline, closed-won.
5. Email platform: open rate, CTR, reply rate.
6. UTM integrity: confirm campaign IDs passed cleanly from spend sources into the CRM.

Databox Genie queries across all connected sources in a single prompt:

Genie returns a structured HTML report—not a chat reply, an exportable document. The report includes top-line KPI cards with variance vs. the prior period built in (total spend, total leads, pipeline generated, MQL rate), a full-funnel visualization from impressions through closed deals, and a channel-by-channel breakdown with CPL vs. target for each platform. Where Genie spots a problem—a sharp drop in one channel, a stage where the funnel is breaking—it writes a short, highlighted callout naming the specific channels driving the drop and the dollar impact. Below the report, Genie suggests follow-up questions you can click to drill into anything it flagged.

The analyst’s role shifts from data assembly to data interpretation. The two-hour scramble becomes a fifteen-minute review of what Genie already pulled. And because Genie remembers context across prompts, the next question you ask builds on the report you just got—you don’t have to re-explain the campaign each time.

The 6-Step AI-Assisted Post-Mortem Framework

Every step runs on two lanes: what AI does with the data, and what the human decides with context AI cannot access. The lane structure isn’t novel; it’s the standard division of labor in any AI-assisted workflow. What is novel is where the lanes intersect in a post-mortem: AI’s strength is multi-dimensional cross-tabs that exceed human working memory, and that strength surfaces a class of finding (channel compression, segment misallocation, anomaly clusters) that manual review almost always misses.

Step 1: Lock the Objectives and Hypotheses

Document what the campaign was supposed to achieve: MQL target, pipeline goal, ICP definition, CPL threshold, and the hypothesis being tested—stated clearly enough that someone could disagree with it.

Genie: Retrieves baseline targets and historical benchmarks from Databox. You: Confirm the hypothesis was stated before the campaign, not reconstructed after. No documented hypothesis is itself a finding.

Step 2: Run the KPI Scorecard

Generate a structured scorecard across four categories: outcome (pipeline, revenue, ROMI), efficiency (CPL, CAC, cycle time), quality (MQL rate, SQO rate, win rate), and operational health (UTM integrity, delivery pacing).

Genie returns the scorecard as a set of tile cards—one for each metric—each showing the plan target, the actual result, the variance, and a status label so you can see at a glance where the campaign stands. A metric within 5% of plan reads ON TRACK; a metric 5–15% over plan on a spend or efficiency dimension reads OVER PLAN; a metric drifting toward the flag threshold reads NEAR FLAG; and any metric more than 15% off target reads FLAGGED. A visual chart underneath shows percent-of-plan attainment across every metric in one frame.

If you didn’t log documented plan figures at campaign launch, Genie shows a callout disclosing what assumptions it used to set the targets (typically industry-median benchmarks per channel) and offers to recalibrate against your actual plan. The transparency is intentional—you always know exactly what Genie is comparing against.

You: Read the FLAGGED metrics and decide which variances are signal versus noise. A CPL 8% over target when CPMs rose industry-wide is context. A CPL 54% over target when CPMs held flat is a red flag that demands explanation.

Step 3: Run the Funnel Diagnostic

Map the funnel from reach to revenue and find the stage where performance degraded most sharply. This is where the post-mortem earns its keep—manual reviews almost always shortcut this step.

When Genie finds a sharp drop, it writes a short, highlighted callout naming the channels or sources driving the decline and quantifying the dollar impact. Specific channels, specific dollars—no room for the post-mortem meeting to drift into “LinkedIn underperformed” hand-waving.

You: Decide whether the break is a targeting, offer, timing, or handoff issue. Genie surfaces the pattern; only you know what was happening in the sales org.

Step 4: Dissect Creative and Channel Performance

Rank every channel and creative variant by CPL, MQL rate, and pipeline contribution. Pipeline contribution, not CTR—high-CTR creative that doesn’t convert is the budget sink.

Genie: Generates the ranked matrix and flags combinations consuming over 20% of budget with below-average MQL conversion. You: Interpret why. Audience-creative misalignment, weak offer, competitor launch mid-campaign—context Genie can’t see.

Step 5: Root Cause Analysis

For the top two or three gaps from Steps 3 and 4, run structured RCA. AI accelerates hypothesis generation; the human validates using qualitative knowledge.

Genie might surface a Tuesday-vs-Thursday segment pattern buried in aggregate numbers. It cannot know that Thursday’s subject lines were rewritten last-minute by the brand team. The pattern is the starting point. The team in the room provides the ending.

This is the permanent division of labor between AI and humans in a post-mortem. Genie sees the data. The leader sees the context the data doesn’t contain: which deal-room conversations changed mid-campaign, which decisions were made under pressure, who was understaffed that quarter, which mid-flight creative pivots happened because the brand team rejected the original work. AI surfaces patterns. The leader explains why the patterns exist and which explanations are actionable. The post-mortem meeting itself is the place where that explanation happens, which is exactly why AI compresses the prep but doesn’t replace the conversation.

Step 6: Build the Action Plan

Convert findings into a prioritized set of changes with owners and due dates. Without this step, the review produces insight but no behavior change.

You: Assign real owners (a person’s name, not “the team”). Set real deadlines. Cut anything that won’t survive the next planning meeting.

The Worked Example: A 30-Day Multi-Channel Performance Review

Frameworks are useful in theory. Worked examples are useful on Monday morning. The example below mirrors the structure Genie actually produces—a single 30-day campaign report across four ad platforms (Google Ads, LinkedIn Ads, Microsoft Ads, Meta Ads) plus organic and email sources.

Steps 1–2: The Scorecard

Genie returns the scorecard with six metric tile cards and a percent-attainment bar chart. Two metrics come back flagged at greater than 15% off target:

Spend overran by 7.6%. CPL drifted 14% over the benchmark but stayed inside the flag threshold. MQL volume is down 14%. But the two metrics Genie explicitly FLAGGED are downstream of marketing: MQL→SQL conversion fell to 27.5% against a 35% plan, and ROMI came in at 0.89x against 1.10x plan. Spend efficiency is not the problem, conversion is.

The channel breakdown confirms it:

Every channel’s CPL beats its industry-median benchmark (Google +10%, Meta +25%, LinkedIn +16%, Microsoft +19%). The top of the funnel is operating well. The leak is somewhere below it.

Step 3: Where the Funnel Broke

The VP runs the funnel diagnostic prompt. Genie surfaces two findings as highlighted callouts in the report—the kind of short, attention-marked summaries the reader can extract without parsing the underlying data:

Pipeline drop concentrated in two channels
Website leads dropped from 349 to 294 (-16%) while Email Campaign leads fell from 223 to 138 (-38%) period-over-period. Both are top pipeline contributors. Combined, they account for $4.56M of the $8.79M pipeline—their contraction is the primary driver of the 12.1% pipeline gap vs. plan.

Critical miss is the MQL-to-SQL handoff
MQL-to-SQL rate is the critical miss at 27.5% vs. a 35% plan (-21.3%). Only 76 of 276 MQLs converted to SQLs. Email Campaign has the best MQL→SQL rate at 39.5%; Social Media has the lowest at 20.0% despite the highest MQL rate (38.2%)—suggesting a volume-vs-quality trade-off in social lead nurturing. Improving MQL→SQL conversion to plan would add ~20 SQLs and close the ROMI gap without additional spend.

What Genie surfaced that manual review would have missed: The paid channels all look healthy. CPL beats benchmark across every platform. A scorecard-only review would conclude “the campaign performed well, fix the volume drop.” But the real story is downstream of marketing entirely—MQL→SQL conversion collapsed. The biggest single lever to recover ROMI isn’t more ad spend or better creative; it’s lead routing, ICP scoring, and sales follow-up SLAs.

Steps 4–6: Diagnosis and Action Plan

Creative dissection by pipeline contribution confirms the paid channels are doing their job. RCA on the two flagged metrics surfaces three correlated patterns: organic landing page conversion rate dropped from 4.2% to 2.9% after a site redesign in week two; the email nurture sequence was paused for ten days during the redesign migration; and Social Media MQLs (the highest-volume source) showed a 20% MQL→SQL rate vs. 39.5% for Email Campaign MQLs—suggesting either weaker lead quality or weaker SDR follow-up on social-sourced leads.

The VP supplies what Genie can’t: the site redesign was a brand team initiative that nobody flagged to marketing ops as a conversion-rate risk, the email sequence pause was an unintended consequence of the migration that nobody caught for a week, and the SDR team was running at 60% capacity that quarter—social MQLs were getting deprioritized in favor of higher-intent inbound. The data shows three symptoms; the VP supplies three causes.

Genie drafts the action plan. The VP edits it, cuts one vague item, and tightens the owner assignments. Five action items survive:

1. Restore email nurture sequence and run a 30-day recovery analysis on the channels that paused. (Marketing Ops Lead, 5 business days.)
2. Roll back or A/B test the landing page redesign on the highest-converting pages until the conversion rate returns to 4%+. (Web Lead + Demand Gen Manager, 10 business days.)
3. Set an SLA with SDRs for sub-8-hour response on Social Media-sourced MQLs, conditional on staffing returning to plan; track MQL→SQL rate weekly. (VP Marketing + Sales Development Manager, next QBR.)
4. Add a marketing ops review gate to any future site redesign or platform migration that touches conversion paths. (VP Marketing, end of quarter.)
5. Audit Non-Brand Search and Display CPL underperformance (9–13% over target); consider shifting that budget to Video/Shopping (24% below target CPL). (Paid Lead, before next campaign launch.)

Total time from data pack to published action plan: 74 minutes. The findings are specific enough to change next quarter’s budget allocation, the cross-functional review process, the sales handoff SLA, and the partner program. That is a post-mortem that compounds.

Your Six-Prompt Reference

Copy these prompts into Genie with your campaign name and date range. Run them in order: Genie carries context from one prompt to the next, so the scorecard builds on the data pack, the funnel diagnostic builds on the scorecard, and so on.

1. Data Pack: “Show me [Campaign Name]’s full-funnel performance from first touch to pipeline, broken out by channel.”
2. Scorecard: “Generate a campaign scorecard comparing actuals to plan for [Campaign Name]. Flag any metric more than 15% off target.”
3. Funnel Diagnostic: “Show me the drop-off rate at each funnel stage for [Campaign Name]. Highlight the largest variance from historical average.”
4. Creative and Channel: “Rank all channels and creative variants in [Campaign Name] by pipeline contribution. Flag any combination consuming over 20% of budget with below-average MQL conversion.”
5. RCA: “[Metric] was [X%] below benchmark in [Campaign Name]. What patterns in the data are most correlated with the underperformance?”
6. Action Plan: “Based on [Campaign Name], draft top 3 wins, top 3 failures with root causes, and 5 prioritized action items.”

Export each report as a shareable HTML document and send to stakeholders before the meeting—not during. Teams that walk in having already read the scorecard and the highlighted findings spend meeting time on judgment calls instead of orientation.

Campaign Post-Mortem Checklist

Run within 5–10 business days of every campaign close.

Before the meeting

• Confirm data sources connected in Databox (ad platforms, CRM, GA4, email)
• Run UTM integrity check: campaign IDs passed cleanly from spend sources to CRM
• Run the Data Pack Genie prompt; review the HTML report and any highlighted callouts before distributing
• Retrieve original hypothesis, MQL target, pipeline goal, CPL ceiling, ICP definition
• Export the Genie report and send to stakeholders 24+ hours before the meeting

During the meeting

• Confirm the pre-campaign hypothesis was documented before launch
• Identify which red flags on the scorecard are connected vs. independent
• Walk the highlighted findings—name the channels driving any pipeline decline and quantify the dollar impact
• Review channel and creative ranked by pipeline contribution, not CTR
• Run RCA on the top two or three gaps; layer in qualitative context
• Run every action item through the three checks (sample size, confounders, actionability), then assign a real owner and deadline

After the meeting

• Publish the Genie HTML report in a shared location with all stakeholders tagged
• Enter action items into your project management system with owners and deadlines
• Add ICP refinements and creative learnings to the campaign playbook
• Schedule a 30-day check-in on action items
• File the post-mortem in a searchable library for future campaigns to draw on