All posts
November 16, 202620 min read

Your Resume Won’t Get You Hired as an AI Marketing Engineer. These 5 Builds Will.

Five portfolio builds hiring managers recognise, the PROOF rubric, a copy-ready case study template, starter prompts, ROI formulas, an interview question bank and a 90-day plan.

Your Resume Won’t Get You Hired as an AI Marketing Engineer. These 5 Builds Will.

In October 2026, Figma posted a Marketing Engineer role that asked for five or more years of experience. Then it said it was open to less time in the seat if your AI builder portfolio was strong. That one line tells you how this job is hired. Nobody is asking which course you finished. They are asking what you built, who uses it and what changed.

TL;DR

An AI marketing engineer is hired on proof, not titles. The fastest proof is a portfolio of three to five live builds, each with a before-and-after number. This guide gives you five builds that hiring managers recognise, a rubric to judge each one, a case study template you can copy, and a 90-day plan to ship three of them.

The five builds:

  1. Speed-to-lead system with a voice agent that calls inbound leads within minutes
  2. Competitor intelligence agent that writes a weekly digest for sales
  3. Content repurposing pipeline that turns one webinar into a month of assets
  4. Sales tool such as an ROI calculator or a battlecard finder
  5. Weekly reporting agent that writes the numbers and the commentary

Also inside: a tool stack for every layer with 40+ tool examples, copy-ready starter prompts for each build, ROI formulas with a worked example, an interview question bank, and free practice data if you are between jobs.

New to the role? Start with my guide to what an AI marketing engineer does, then come back here to build your proof.

Why do hiring managers want a portfolio instead of a resume?

Because the role is too new for titles to mean anything. A resume says what you claim. A portfolio shows what you can build. Hiring posts in 2026 already say this out loud:

EmployerRoleWhat the listing says about proof
FigmaMarketing Engineer, AI Deployment (posted October 14, 2026)Asks for 5+ years, but is open to less time in the seat with a strong AI builder portfolio
GOAT GroupMarketing Engineer I (entry level)Requires a portfolio of AI builds; personal projects, internships and open-source work count
WonoloMarketing engineerWants 2 to 4 years in a technical role where you built things, whatever the title

Source for all three listings: CXL.

The wider market points the same way:

  • AI skills went mainstream fast. 40% of 878 US marketing job descriptions asked for AI skills in Q3 2026, up from 15% in Q1 (CXL). When everyone writes "AI skills" on a resume, only shipped work separates you.
  • Reviewers spend more time on portfolios. One guide for GTM engineers estimates about 90 seconds on a resume and 4 to 7 minutes on a portfolio (GTME Pulse).
  • Many builders are self-taught. The same source cites a 2026 survey in which 53% of GTM engineers were self-taught. A portfolio is the credential they have.
  • Hiring tests are practical. Clay's hiring guide for GTM engineers uses take-home tasks, such as building an outbound campaign in 60 minutes, and says the real signal is a willingness to learn by tinkering (Clay).

The takeaway: three measured builds will do more for you than any certificate.

What makes a build portfolio-worthy? The PROOF rubric

A build is portfolio-worthy when it solves a business problem, real people use it, and you can prove what changed. Here is a five-part rubric I call PROOF. Score each build from 1 to 5 on every letter before it goes in your portfolio.

LetterQuestion a hiring manager is silently askingWhat a 5 looks likeWhat a 1 looks like
ProblemDoes this fix something the business cares about?Tied to a metric leadership already tracks: pipeline, response time, win rate, hours"I wanted to try n8n"
Real usersDoes anyone use it besides you?A colleague or team runs it weekly without you in the roomA demo that only works on your laptop
OutcomeWhat changed, in numbers?4 to 6 before-and-after metrics with the time window stated"It saved a lot of time"
OperationsWhat happens when it breaks?Error handling, a human review step, logs, a named ownerNo plan for a failed API call or an empty CRM field
FluencyCan you explain the trade-offs?You can say what you would change, what it costs to run and when to retire itYou can only list the tools you used

How to read your score:

  • 20 to 25: lead your portfolio with it.
  • 14 to 19: include it, and fix the weakest letter first. It is usually Outcome or Operations.
  • Below 14: keep it as practice, not proof.

The rubric follows what hiring guides already test. The Keyword suggests asking candidates which systems are live, who maintains them, what broke recently and how they found out (The Keyword). Clay lists over-engineering, chasing trendy tools and missing commercial awareness as red flags (Clay). GTME Pulse calls out portfolios with no metrics and no error handling as common mistakes (GTME Pulse).

The 5 builds at a glance

These five cover the areas hiring managers recognise: lead response, research, content, sales tools and reporting. Pick three that touch different areas, and your portfolio shows range as well as depth.

#BuildBusiness problem it fixesMain userHeadline metricRough time to first version
1Speed-to-lead system with a voice agentInbound leads wait hours and go coldSales and SDRsMinutes to first contact, lead-to-meeting rate2 to 3 weeks
2Competitor intelligence agentSales hears about competitor moves from prospectsSales and product marketingHours saved per week, battlecard freshness1 to 2 weeks
3Content repurposing pipelineOne webinar becomes one recording nobody watchesContent and social teamsAssets per source, days from event to assets1 to 2 weeks
4ROI calculator or battlecard finderReps improvise value claims and competitor answersAccount executivesWeekly active reps, deals where it was used2 to 3 weeks
5Weekly reporting agentHours every Monday spent copying numbers into slidesMarketing leadershipReporting hours saved, time to insight1 week

Time estimates assume no-code or low-code tools and a few hours a week. They are a planning guide, not a benchmark.

The AI marketing engineer toolkit

You do not need every tool below. You need one per layer, and most layers have a free tier or trial to learn on. Pick the ones your target employers already use: job posts usually name them.

LayerWhat it does in your buildsTool examplesUsed in builds
AI modelReads, classifies, summarises and writesClaude, ChatGPT, GeminiAll five
AutomationTriggers, schedules and connects appsn8n, Zapier, Make, GumloopAll five
App builderTurns a brief into a working web appLovable, Replit, Bolt, v04
AI coding helpWrites and fixes scripts, SQL and API callsClaude Code, Cursor4, 5
Data storeHolds inputs, outputs and logsGoogle Sheets, Airtable, Supabase, BigQuery4, 5
CRMSystem of record for leads and dealsHubSpot, Salesforce, Zoho CRM, LeadSquared1, 4
EnrichmentAdds company and contact data to a leadClay, Apollo1
Voice agentsCalls, qualifies and books meetingsVapi, Retell AI, ElevenLabs Agents, Exotel1
Change monitoring and competitive intelligenceSpots changes on competitor pagesVisualping, Distill.io, Klue, Crayon2
Transcription and contentTurns recordings into text, clips and visualsOtter.ai, Fireflies.ai, Descript, OpusClip, Canva3
Reporting connectorsPulls ads and analytics data into one placeSupermetrics, Windsor.ai, Looker Studio5
DocumentationRecords your proofLoom, Notion, Scribe, ExcalidrawAll five

Which AI model? Most marketers use more than one. In Basis's 2026 survey of 184 marketing and advertising professionals, 74.5% used ChatGPT and 72.3% used Claude, while 51.1% said their organisation runs AI agents (PPC Land on the Basis report). Use the one your employer has approved for company data.

A starter stack on a tight budget: Google Sheets as the database, n8n or Zapier as the automation layer, one AI model, Loom for the walkthrough and Notion for the portfolio. That combination covers builds 2, 3 and 5. Tool names and free tiers change often, so check current pricing before you commit.

The 5 builds, step by step

Each build below follows the same pattern: the problem, the stack, the steps, what to measure, and what to show a hiring manager. Swap any tool for one you already use. The logic matters more than the brand.

Build 1: Speed-to-lead system with a voice agent

The problem. Inbound leads go cold while they wait. In Harvard Business Review's audit of 2,241 US companies, only 37% responded to a web lead within an hour, and 23% never responded. Firms that made contact within an hour were nearly 7 times as likely to qualify the lead as firms that waited an hour longer (HBR, summarised by Fronetics).

The stack. A form or CRM trigger, an automation platform (n8n, Zapier or Make), an enrichment step, a voice agent platform, a calendar link, and the CRM.

The steps.

  1. A form submission fires a webhook.
  2. The workflow enriches the lead with company size, industry and role.
  3. A routing rule decides: call now, email now, or send to nurture.
  4. The voice agent calls within minutes, asks three qualifying questions and offers meeting slots.
  5. The call summary and qualification fields are written to the CRM, and the owner gets an alert.
  6. Unqualified leads get a tailored email sequence instead of silence.

What to measure. Median minutes to first contact, contact rate, lead-to-meeting rate and meetings per week. Compare 60 days after launch with 60 days before.

What to show. A 3-minute recording of a test lead going from form to booked meeting, the workflow screenshot, and the before-and-after table.

Guardrails. Consent to call, calling hours, a clear opt-out, and human review of flagged calls. Rules on automated calls differ by country, so check them before you go live. In the US, the FCC ruled in February 2024 that AI-generated voices count as "artificial or prerecorded" voices under the TCPA: callers generally need prior consent, must identify who is calling and must offer an opt-out (Wiley). In India, check the TRAI rules on commercial calls and do-not-disturb preferences. The simplest safe design: only call people who filled in a form asking to be contacted, and say in the first sentence that it is an AI assistant.

Tool picks.

NeedStarter optionAdvanced option
Trigger and workflowZapier or Maken8n with error branches
EnrichmentApolloClay waterfall enrichment
Voice agentRetell AI or Vapi templatesVapi or ElevenLabs Agents with custom tools, or Exotel for Indian telephony
BookingCalendly or HubSpot MeetingsRound-robin routing in the CRM
CRMHubSpot free CRMSalesforce or LeadSquared

Starter prompt for the voice agent:

You are an AI assistant calling on behalf of [Company]. The person you are
calling filled in a form on [page] asking about [product] [minutes] ago.

Open with: "Hi [first name], this is an AI assistant from [Company]. You asked
about [product] a few minutes ago. Is now a good time for two quick questions?"

If they say no: offer to call back at a time they choose, then end the call.

Ask only these three questions, one at a time:
1. What made you look at [product] now?
2. Roughly how many [seats / agents / locations] would use it?
3. When are you hoping to have something in place?

Qualified if: [rule, e.g. 20+ seats AND timeline under 3 months].
If qualified: offer two meeting slots from [calendar tool].
If not: say you will email useful resources, and end politely.

Never discuss pricing, discounts or contract terms. If asked, say a specialist
will cover it in the meeting. If the person asks for a human or says stop,
comply immediately. Keep each turn under 25 words.

Measurement tip. Pull the CRM's created-time and first-activity-time for every inbound lead for the 60 days before launch. That baseline is the most important number on your case study page.

Build 2: Competitor intelligence agent

The problem. Sales learns about a competitor's price change or launch from a prospect, mid-call.

The stack. A scheduler, web page change monitoring or RSS, an AI model, a shared context document with your positioning, and Slack or email.

The steps.

  1. List 3 to 5 competitors and the pages that matter: pricing, product updates, careers, customer stories.
  2. Check those pages on a weekly schedule and capture what changed.
  3. Ask the AI model to sort each change into pricing, product, hiring or messaging, and to flag what sales should know.
  4. Draft a one-page digest with suggested talk tracks, using your positioning document as context.
  5. A product marketer reviews it in 15 minutes, then it goes to sales.

What to measure. Hours saved per week versus manual tracking, how fresh the battlecards are (days since last update), and how many reps open the digest.

What to show. Two real digests, the review checklist, and one example where a digest changed a sales conversation.

Tool picks.

NeedStarter optionAdvanced option
Spot page changesVisualping or Distill.io alertsScheduled scraping in n8n
Summarise and classifyClaude or ChatGPT inside Zapier or Maken8n with a structured output schema
Store historyGoogle SheetsAirtable or Notion database
DeliverSlack or email digestKlue or Crayon battlecards updated from the digest

Starter prompt for the weekly digest:

You are a competitive intelligence analyst for [Company].
Context: our positioning document is below, followed by this week's detected
changes on competitor pages.

For each change:
- Category: pricing / product / hiring / messaging / customer win
- What changed, in one sentence, quoting the page where possible
- So what for sales: one sentence
- Suggested talk track: up to two sentences, in line with our positioning
- Confidence: high / medium / low, and why

Then write a 3-bullet summary at the top: the changes sales must know this week.
Do not speculate beyond what the pages say. If nothing important changed,
say so in one line.

[Positioning document]
[Detected changes with URLs]

Measurement tip. Time yourself doing the digest by hand once. That is your baseline for hours saved.

Build 3: Content repurposing pipeline

The problem. A one-hour webinar turns into one recording and a single LinkedIn post.

The stack. A transcription tool, an AI model with your brand voice guide and customer language, a document or content calendar, and a design tool for visuals.

The steps.

  1. Transcribe the recording.
  2. Pull out the strongest claims, stories, data points and customer questions.
  3. Draft a set of assets: a blog post, 5 LinkedIn posts, an email, short-video clip scripts and an FAQ.
  4. Check every draft against the brand voice guide, then have a human edit it.
  5. Schedule the assets across 4 weeks.

What to measure. Assets per source, days from event to first asset, and engagement or pipeline influenced compared with the previous manual process.

What to show. One source asset with everything it produced, side by side, and the prompt and voice guide you used.

Tool picks.

NeedStarter optionAdvanced option
TranscribeOtter.ai, Fireflies.ai or DescriptDescript with speaker labels and edits
Pull clipsOpusClipDescript with manual clip selection
Draft assetsClaude or ChatGPT with a brand voice guideA Claude Project or custom GPT that holds your voice guide and past best posts
VisualsCanvaCanva brand kit with templates
ScheduleNative LinkedIn scheduling or BufferYour marketing automation platform

Starter prompt for the repurposing step:

You are a B2B content editor for [Company]. Below are our brand voice guide,
three of our best-performing posts, and a webinar transcript.

Step 1. List the 8 strongest ideas in the transcript: claims, stories, data
points and audience questions. Quote the transcript for each.
Step 2. From those ideas, draft:
- 1 blog post outline (H2s as questions, 1 table idea)
- 5 LinkedIn posts (hook in the first line, under 200 words each)
- 1 email to webinar registrants who did not attend
- 3 short-video clip scripts with timestamps from the transcript
- 5 FAQ entries
Step 3. Flag any claim that needs a source or legal check.

Use only what the speakers said. Do not invent statistics or customer names.

[Voice guide] [Best posts] [Transcript]

Measurement tip. Count assets and days-to-publish for the last two webinars done by hand. Compare engagement per asset, not just volume.

Build 4: ROI calculator or battlecard finder

The problem. Account executives improvise value claims and competitor answers, so the story changes from deal to deal.

The stack. An app builder such as Lovable or Replit, a small database such as Supabase, and your CRM for usage tracking. Add an AI assistant for the battlecard version.

The steps.

  1. Agree the inputs and formula with sales and finance before you build anything.
  2. Build a simple web app: inputs, assumptions you can see, and a result the rep can share.
  3. For the battlecard finder, load approved battlecards and let reps ask a question, such as "they say they are cheaper", and get the approved answer with its source.
  4. Track usage and link it to opportunities in the CRM.

What to measure. Weekly active reps, deals where the tool was used, and win rate or sales cycle length in those deals compared with similar deals.

What to show. The live tool, sign-off from finance on the formula, and usage data.

Tool picks.

NeedStarter optionAdvanced option
Build the appLovable, Bolt or v0Replit, or Cursor and Claude Code on your own codebase
Store inputs and usageGoogle SheetsSupabase with row-level security
Battlecard answersA Claude Project or custom GPT loaded with approved battlecardsKlue or Crayon with your own retrieval layer
Usage in dealsA CRM field the rep ticksEvent tracking linked to opportunity IDs

Starter brief for the app builder (ROI calculator):

Build a one-page ROI calculator for [product] for account executives.

Inputs (with defaults): [number of agents], [average handle time in minutes],
[cost per agent hour], [current monthly tool spend].
Assumptions (visible and editable, with a note on where each comes from):
[e.g. 20% reduction in handle time, from our customer study].
Outputs: monthly savings, annual savings, payback period in months.
Show the formula under each output in plain words.
Add "Copy summary" (plain text for email) and "Download PDF" buttons.
Log each use (timestamp, rep email, inputs, outputs) to a table.
Brand: [colours, font]. No login needed; works on mobile.

Measurement tip. Agree with sales ops on how to tag deals that used the tool before launch, or you will not be able to compare win rates later.

Build 5: Weekly reporting agent

The problem. Someone spends Monday morning copying numbers into slides, and the commentary arrives too late to change anything.

The stack. Connections to your ad, analytics and CRM data, a spreadsheet or warehouse, an AI model, and Slack or email.

The steps.

  1. Agree the 8 to 10 metrics leadership actually uses, with their definitions.
  2. Pull the data on a schedule into one sheet or table.
  3. Have the AI model write a summary: what moved, by how much, the likely reason, and one recommended action.
  4. A human checks the numbers before the report is sent, at least for the first month.
  5. Post it to the leadership channel every Monday at the same time.

What to measure. Reporting hours saved per week, number of errors caught, and how many decisions were made from the report.

What to show. Three consecutive weekly reports, the metric definitions, and the checks you run before sending.

Tool picks.

NeedStarter optionAdvanced option
Pull dataNative Google Sheets connectors, or Supermetrics or Windsor.aiScheduled exports to BigQuery
OrchestrateZapier or Make on a weekly schedulen8n with checks that stop the run on missing data
Write commentaryClaude or ChatGPTThe same model with last month's reports as examples
ShowA Slack post with a link to the sheetA Looker Studio dashboard plus the written summary

Starter prompt for the commentary:

You are a marketing analyst writing the Monday report for the leadership team.
Below are this week's metrics, the previous 4 weeks, and our metric definitions.

Write:
1. Three-line summary: the most important change, with numbers.
2. A table: metric, this week, last week, 4-week average, change in %.
3. For any metric that moved more than [15]%: the likely reason, based only
   on the data provided, and how confident you are.
4. One recommended action, with the owner and the expected effect.

Rules: never invent a cause the data does not support. If data looks broken
(zeros, duplicates, a move of more than [50]%), flag it at the top instead of
explaining it.

[Definitions] [This week] [Previous 4 weeks]

Measurement tip. Keep a simple log of the decisions leadership made from each report. "Decisions made" is a stronger portfolio metric than "hours saved".

This is the easiest of the five to start, because it uses data you already have and needs no customer-facing step.

How to put an ROI number on every build

Most teams never do this, which is exactly why it sets you apart. In Basis's 2026 survey, only 6.5% of marketing organisations measured the return on their AI spending against defined benchmarks (PPC Land on the Basis report). Salesforce's State of Marketing 2026, a survey of 4,450 marketers, found that high performers save about 8 hours a week through automation (Salesforce). Turn your own hours and meetings into money, and your case study speaks the language leadership uses.

1. Time-saving builds (competitor agent, content pipeline, reporting agent):

Annual value = hours saved per week × working weeks × loaded hourly cost

2. Revenue builds (speed-to-lead, ROI calculator, battlecard finder):

Added pipeline = extra meetings × meeting-to-opportunity rate × average deal size
Added revenue  = added pipeline × win rate

3. For every build:

ROI = (annual value − annual running cost) ÷ annual running cost

Worked example (illustrative numbers, not benchmarks):

BuildInputsResult
Weekly reporting agent3.5 hours saved per week, 48 working weeks, Rs 1,500 loaded hourly cost, Rs 30,000 a year in tool costsRs 2,52,000 annual value; ROI of 740%
Speed-to-lead system8 extra meetings a month, 40% become opportunities, Rs 6,00,000 average deal, 25% win rateRs 19.2 lakh added pipeline and Rs 4.8 lakh added revenue a month

Three rules that keep your numbers credible:

  • Use the company's own conversion rates and deal sizes, not industry averages.
  • Show the inputs, so the reader can redo the maths with their own numbers.
  • Say what else changed in the same period, such as a new campaign or a pricing change. Hiring managers trust a candidate who points out the caveats.

Which build should you start with?

Start low on effort, then move toward revenue. Your first build should prove you can ship. Your last build should prove you can move pipeline.

The weekly reporting agent is the safest first build: it uses existing data and has no customer-facing step. The speed-to-lead system is the strongest finale, but it touches real prospects, so build it once you have guardrails you trust. If you can only do three, pick 5, then 2 or 3, then 1 or 4.

The case study template (copy this)

Every build in your portfolio gets one page in the same structure. Keep it under 600 words. Hiring managers skim, and a repeated layout lets them compare your builds quickly.

# [Build name]: [result in one line, with a number]

Example: Speed-to-lead voice agent: first contact in minutes, not hours

## 1. Snapshot
- Role: [what you owned, e.g. solo builder / built with sales ops]
- Timeline: [weeks from idea to first version, and how long it has run]
- Stack: [3 to 6 tools]
- Users: [who uses it and how often]
- Status: [live / retired / handed over to X]

## 2. The problem (2 to 3 sentences)
What was broken, who felt it and what it cost.
Include one baseline number.

## 3. The build (4 to 6 sentences + one diagram)
Trigger -> data -> AI step -> human check -> output.
Name the key decision you made and why.

## 4. Results (4 to 6 metrics)
| Metric | Before | After | Time window |
| --- | --- | --- | --- |
| [metric 1] | | | |
| [metric 2] | | | |
| [metric 3] | | | |
| [metric 4] | | | |

How measured: [data source, and what else changed in the same period]

## 5. When it breaks
- Failure I planned for: [e.g. API timeout, empty CRM field]
- Human review step: [who checks what, and when]
- Monitoring: [alert or log]

## 6. What I would do differently (1 to 2 sentences)

## 7. Proof
- Walkthrough video: [3 to 5 minute Loom link]
- Screenshots: [workflow, output, dashboard]
- Quote from a user: [one line from the person who uses it]

Why each section is there:

SectionWhich PROOF letter it proves
Snapshot and ProblemProblem, Real users
BuildFluency
ResultsOutcome
When it breaksOperations
What I would do differentlyFluency
ProofReal users, Outcome

The word limits and the 4 to 6 metrics follow GTME Pulse's advice for project pages, including a short video walkthrough for at least two projects (GTME Pulse).

A filled example: the weekly reporting agent

Here is what a finished page looks like. The company is a made-up B2B SaaS team, and the numbers are illustrative only. Replace them with your own measured results, and never publish numbers you cannot back up in an interview.

Weekly reporting agent: Monday report ready by 9 am, down from half a day of manual work

Snapshot. Role: solo builder, with sign-off from the head of marketing. Timeline: 1 week to first version, live for 10 weeks. Stack: Google Sheets, n8n, Claude, Slack. Users: marketing leadership team of 4, every Monday. Status: live.

The problem. The demand gen manager spent about 4 hours every Monday copying numbers from ads, web analytics and the CRM into slides. The report reached leadership on Monday afternoon, after the weekly planning meeting. Budget shifts waited a week.

The build. A scheduled n8n workflow pulls 9 agreed metrics into one sheet every Sunday night. Claude reads the sheet and the previous 4 weeks, then writes what moved, the likely reason and one recommended action. The demand gen manager checks the numbers in 10 minutes and approves the post to Slack. The key decision was to agree metric definitions with finance first, so nobody argued about the numbers.

MetricBeforeAfterTime window
Hours spent on the report4 per week0.5 per week10 weeks before vs 10 weeks after
Report deliveryMonday afternoonMonday 9 amSame
Data errors caught before sendingNot tracked310 weeks after
Budget changes made in the same week16Same

When it breaks. If an API call fails, the workflow posts an alert instead of a partial report. A human checks the numbers before sending. Every run is logged in a separate sheet tab.

What I would do differently. Add a confidence note when a metric moves more than 30%, because two of the early recommendations chased data glitches.

Proof. A 4-minute walkthrough video, a screenshot of the workflow, three sample reports, and a one-line quote from the head of marketing.

PROOF score for this page: Problem 5, Real users 5, Outcome 4, Operations 4, Fluency 5 = 23 out of 25. Lead with it.

For real examples of systems I have built as a product marketer, see my projects page.

How to present your portfolio

The portfolio page: five blocks, top to bottom

  1. One-line positioning. For example: "Product marketer who builds AI systems for lead response, sales enablement and reporting."
  2. Three to five build cards. Each card has the build name, its headline result with a number, and the tools as small tags. Put your highest PROOF score first.
  3. One case study page per card, using the template above.
  4. How I work. Three lines on your guardrails: human review, data privacy and how you retire builds that do not move a metric.
  5. Contact. LinkedIn and email.

Keep it to three to five builds. GTME Pulse notes that ten projects leave each one 25 to 40 seconds of a reviewer's time (GTME Pulse). Notion is fine to start, and so is a page on your own site. Spend an afternoon on the hosting, not a month.

The 3-minute interview walkthrough

When an interviewer says "walk me through something you built", use this script:

TimeSayShow
0:00 to 0:30The problem and the baseline numberOne sentence and one number
0:30 to 1:30How it works, step by step, and the key decision you madeThe workflow diagram
1:30 to 2:15The results, with the time windowThe before-and-after table
2:15 to 2:45What broke and how you found outThe alert or log
2:45 to 3:00What you would build nextNothing; just say it

The "what broke" step is the one most candidates skip, and it is the one that builds the most trust. It answers the interview signals The Keyword recommends: live systems, who maintains them and recent breakages (The Keyword).

The interview question bank

These are the questions your portfolio should already answer. Prepare a two-sentence answer for each, using one of your builds.

QuestionWhat they are really testingA strong answer includes
Tell me about the last process you built from scratch.Systems thinkingThe trigger, the logic, the output and who uses it
Which of your systems are live right now, and who maintains them?Real users and ownershipNamed systems, an owner, how often they run
What broke most recently, and how did you find out?OperationsThe failure, the alert or log that caught it, the fix
How do you know it worked?OutcomeBaseline, after, time window and what else changed
How did you stop the AI from making things up?JudgementContext documents, structured outputs, human review
What would you not automate?Commercial sensePricing talks, sensitive customer moments, anything without a clear owner
If I gave you our stack tomorrow, what would you build first?PrioritisationA bottleneck tied to a metric, and a 30-day plan
What did you teach yourself recently?CuriosityA specific skill, why, and what you built with it

Several of these come straight from published hiring guides: Clay suggests asking about the last process a candidate built and what they taught themselves recently (Clay), and The Keyword suggests asking about live systems, owners and recent breakages (The Keyword).

Expect a take-home task as well. Clay's examples include building an outbound campaign from sample data within 60 minutes, and analysing inbound leads to recommend better follow-up. Your speed-to-lead build is good practice for both.

A 90-day plan to ship three builds

Three phases, one build each, with a gate you must pass before moving on. The gates matter more than the dates: a build nobody uses is practice, not proof.

PhaseBuildGate before you move on
Days 1 to 30Build 5: weekly reporting agentBaseline measured, someone other than you reads it every week, PROOF score of 14 or more
Days 31 to 60Build 2 or 3: competitor agent or content pipelineA second team uses it, case study page written, PROOF score of 14 or more
Days 61 to 90Build 1 or 4: speed-to-lead system or ROI calculatorGuardrails live, first before-and-after numbers in, 3-minute walkthrough recorded

Budget about 2 hours a week just for documenting: screenshots, metrics and notes on what broke. Writing the case study at the end, from memory, is where most portfolios lose their numbers.

No company data? Where to practise

Not having an employer is not a blocker. Every build can run on data you can legally use. Keep anything private out of your screenshots either way.

BuildPractice data you can useHow to make it real
1. Speed-to-leadA form on your own website or a friend's small business, with HubSpot's free CRMRoute real enquiries; disclose the AI and only call people who asked
2. Competitor agentPublic pricing, product update and careers pages of any software categorySend the digest to a sales or founder friend and ask whether it is useful
3. Content pipelineYour own talks, podcast appearances or LinkedIn LivesPublish the outputs on your own channels and track engagement
4. ROI calculatorPublic pricing and the public case studies of a product you knowAsk a salesperson to use it in one real conversation
5. Reporting agentThe Google Analytics demo accountRun it weekly for a month and publish the reports in your portfolio

The Google Analytics demo account is free for anyone with a Google account. It holds real data from the Google Merchandise Store, an e-commerce site, and from Flood-It!, a mobile game, with view-only access.

Build on your own job search. A system that finds relevant roles, researches each company and drafts a tailored note is a GTM build in its own right, and hiring managers remember it. Measure it like any other build: roles processed, replies, interviews.

7 mistakes that sink a portfolio

MistakeWhy it hurtsFix
Ten small projectsEach gets seconds of attention, and none shows depthThree to five builds, ranked by PROOF score
Tool lists instead of resultsHiring managers want outcomes, not your stackLead every card with a number
No time window on metrics"Saved 10 hours" means nothing without "per week, over 8 weeks"Always state before, after and the period
Demos built on fake data onlyIt is unclear whether the build survives real inputsRun it on your own work, or on public data, for at least four weeks
No failure storyIt looks like it has never broken, or that you never checkedAdd the "When it breaks" section
Confidential data in screenshotsIt shows poor judgement, and it can cost you the jobBlur names, emails and revenue, or use a sample dataset
Claims you cannot defendOne weak answer in an interview undoes the whole pagePublish only numbers you measured and can explain

The first five are echoed by GTME Pulse's list of common portfolio mistakes, which also warns against showing tools or features that no longer exist (GTME Pulse).

Frequently asked questions

What should an AI marketing engineer portfolio include?

Three to five live builds, each with a one-page case study: the problem, how it works, 4 to 6 before-and-after metrics, how it handles failure, and a short video walkthrough.

How many projects should be in my portfolio?

Three to five. Fewer than three looks thin. More than five dilutes the reviewer's attention.

Can I build a portfolio without a job or without company data?

Yes. Run the builds on your own work: your job search, your LinkedIn content, a friend's small business, or public data. GOAT Group's entry-level marketing engineer listing accepts personal projects, internships and open-source work.

Do I need to code to build these?

No. All five builds can be done with automation platforms, app builders and an AI assistant. Basic SQL and API skills help you handle errors and data cleanly.

Where should I host my portfolio?

Notion, your personal website or a GitHub page all work. Hiring managers care about the content, not the platform.

Which build should I do first?

The weekly reporting agent. It uses data you already have, has no customer-facing risk and shows results within a few weeks.

The bottom line

In this role, the portfolio is the resume. Pick three builds from different areas, score them with PROOF, document each one with the template, and practise the 3-minute walkthrough. Ninety days from now, you could have something no certificate gives you: proof that you can turn an idea into a system that people use and that moves a number.

Start with one build this week. Measure the baseline before you change anything.

References

  1. CXL, "What is a marketing engineer, and why are companies hiring them?" (2026). Source for the Figma, GOAT Group and Wonolo listings and the share of US marketing job descriptions asking for AI skills.
  2. GTME Pulse, "GTM Engineer Portfolio Examples: What to Showcase". Source for review times, the recommended number of projects, case study lengths and common portfolio mistakes.
  3. Clay, "How to hire a GTM Engineer" (2025). Source for hiring traits, red flags, interview questions and take-home tasks.
  4. The Keyword, "AI marketing engineering". Source for interview signals about live systems and breakages.
  5. Salesforce, "State of Marketing 2026" (10th edition, 4,450 marketers). Source for the hours high performers save through automation.
  6. PPC Land on Basis, "AI and the Future of Marketing" (2026, 184 respondents). Source for AI tool usage, AI agent adoption and ROI measurement rates.
  7. Oldroyd, McElheran and Elkington, "The Short Life of Online Sales Leads", Harvard Business Review (March 2011). Figures as summarised by Fronetics.
  8. Wiley, "FCC Extends Regulatory Reach Over AI" (February 2024). Source for the TCPA rules on AI-generated voice calls.
  9. Google, "Demo account", Analytics Help. Source for the free practice data.
  10. Deepak Ruchandani, "The AI Marketing Engineer: What the Role Is, What It Does, and How to Become One".
  11. Deepak Ruchandani, projects.

Tool links in the toolkit table go to each vendor's own site. Features and free tiers change often, so check them before you build.

Data checked as of November 16, 2026.