Jessica MannaFormer SVP Growth Marketing, Better

Selected work · Jan–Sept 2026

AI work that shows up
in the numbers.

I led growth marketing at Better. This is a selection of my team’s AI work there from January to September 2026, across acquisition, lifecycle, content, design and marketing operations: what changed, what it’s worth, and where people still make the call.

Total value

$4.49M+

annual savings, costs caught, projected revenue and released capacity priced at $48 an hour, combined

Most of it is the $3.56M revenue projection, not revenue booked

Annual savings

$405K+

a year in agency and tool spend, eliminated

Content agency $400K+, email tool $5K

Costs caught

$383K

caught through invoice reconciliation and financial QC

Not recurring, and not the same as recovered

$788K+a year in broader cost impact: savings plus costs caught.

Projected revenue

$3.56M

incremental annual revenue projected from rolling out lifecycle personalization

A rollout model, not revenue booked

Capacity released

3,078

hours a year back to the team, upper-end estimate

$147,744 at $48 an hour; capacity, not headcount

Four kinds of value, and how they add up.

Savings, costs caught, projected revenue and released capacity are measured differently and over different periods. The $4.49M+ total adds them together, with capacity priced at $48 an hour. The charts below show each part on its own basis.

  • Annual savings: solid
  • Costs caught: hatched
  • Projection: dashed
  • Capacity: segmented by workflow

$405K+ a year eliminated, $383K caught

Annual savings recur. Costs caught do not, and they aren’t the same as amounts recovered. The $788K+ line is a broader view of cost impact over a year. The $383K was caught within 12 months.

Content agency eliminatedReplaced by Claude agents
$400K+a year
Email tool subscription eliminatedReplaced by a custom editor
$5Ka year
Annual cost savingsRecurring
$405K+a year
Costs caught in invoice QCNot recurring
$383Kcaught
Broader cost impactSavings plus costs caught
$788K+a year

How $3.56M is projected

A rollout model built on a vendor case study. It projects revenue; it doesn’t report revenue booked.

  1. 12,356baseline applications a year× 30% lift
  2. 3,707incremental applications× 12% funding rate
  3. 445incremental funded transactions× $8,000 revenue per funded transaction
  4. $3.56Mprojected incremental annual revenue

Calculated on unrounded counts (3,706.8 applications, 444.816 transactions): $3,558,528.

The evidence behind the 30%

Before 9.25%After 12.08%+2.83 pts, about 30.6% relative

Application-start rate in JustAI’s public case study of a shopping-journey test. The model applies 30% and assumes it carries across the annual baseline. Creative and cadence changed too, so not all of the lift is attributable to AI.

Where 3,078 hours come from

Upper-end estimates across ten recurring workflows. Capacity released, not headcount or dollars.

3,078hours a year

  • Content marketing
  • Affiliates
  • Marketing operations
  • Design
  • Lifecycle

Known overlapping components are excluded and some remaining overlap is unresolved. One-time and per-unit results, such as the SOP batch, kickoff pages and the email pilot, aren’t included.

Thirty-five projects, from assisted tasks to monitored automation.

Each square is a project, colored by function. Seventeen are repeatable workflows the team reuses, six feed working systems directly, and two run as monitored automations with people approving the decisions that matter.

Projects by function and operating model. Select a project to open it, or a function or operating model to filter the full list.
Messaging matched to market eventsAssistants connected across the rate-drop workflow
Email subscription replaced by a custom editorNewsletter rate charts without the design queueBrand-voice first drafts for lifecycle emailRate-drop alert decision support
Weekly newsletter without the weekly rebuild, featuredEmail templates built from approved designs, featuredLifecycle campaign guides in one shared format
Lifecycle conversion through always-on personalization, featured
Affiliate analysis in code, within access limits
Affiliate marketer performance insight engine, featuredAffiliate budget pacing made visible
Paid-media spend planning from unit economics
Acquisition reporting built around the decisionLead sources grouped for clean reportingAcquisition pacing checked against target
Content agency replaced by Claude agents, featuredCustomer personas grounded in first-party evidenceOne workflow for rate, news and search content
Sales licensing data kept currentForward-looking budget and forecast planningEngineering-request quality reviewLaunch reporting for a new direct-mail programEngineering sprints read for marketing impact
Daily commercial performance viewStandard operating procedures in one formatEngineering requests with the business case built inDaily direct-mail performance refresh
Acquisition spend reconciled to the invoice, featuredCampaign tracking audited end to end before launch, featuredSales kickoff pages for direct-mail campaigns
Autonomous direct-mail campaign management, featured
AI creative exploration
On-brand presentations without manual formatting
ProjectFeatured belowSelect a square to open the project, or a row or column to filter the full list.

All 35 projects.

Filter by function, application, tool or operating model, then open any project for its full record: objective, starting point, workflow, tools, impact, oversight and evidence.

Showing all 35 projects

Filtered project list
Content marketingContent agency replaced by Claude agentsFeatured projectAn external content agency is gone, replaced by three Claude agents: Brief Builder, Draft Writer and New Generator. People still own content strategy, editorial standards and what gets published.$400K+a year saved, plus 936 hoursRepeatable workflowClaudeBuilder.io
Objective
Eliminate an external content agency through a series of Claude agents: Brief Builder, Draft Writer and New Generator.
Starting point
Content production relied on external agency spend, with briefing and drafting work recurring every cycle.
How it works
  1. Brief Builder develops the brief.
  2. Draft Writer and New Generator work within the content-production system.
  3. People set content strategy and editorial standards and decide final publication.
Before and after

BeforeAgency-dependent content production

AfterClaude agents inside the content-production workflow

Human oversight
People own content strategy, editorial standards and final publication.
Impact
$400K+ a year in external agency spend eliminated. Team records also attribute an estimated 936 hours a year to the brief-to-draft workflow; those hours are part of the 3,078-hour capacity estimate. The agency figure is outsourced production expense and is not counted again as headcount savings.
Inputs, roles and dependencies
Content marketing, led by the content strategist and editor role. The agents are Brief Builder, Draft Writer and New Generator.
Tools and roles
Claude agents Brief Builder, Draft Writer and New Generator made up the content-production system. Builder.io is part of this workflow's toolkit.
Takeaway
Replace recurring production expense with a maintained capability, and keep editorial standards explicit.
Date
February 2026
Classification
Content marketingProductionRepeatable workflow
Sources
Agency savings confirmed Oct 14, 2026; team tracker (Aug 6, 2026).
LifecycleLifecycle conversion through always-on personalizationFeatured projectJustAI personalizes and tests lifecycle journeys inside rules the team’s marketers set. In the vendor's case study, the shopping-journey application-start rate rose from 9.25% to 12.08%. Rolled out at a 30% lift, the model projects $3.56M in incremental annual revenue.$3.56Mprojected a yearMonitored automationJustAIIterableRudderStackClaude
Objective
Increase application rates through always-on AI testing and personalization.
Starting point
With many campaigns running at once, it was hard to find the higher-converting email variants, and fixed messages and send sequences limited relevance across very different homebuyer situations.
How it works
  1. Refresh journey creative for the customer's context.
  2. JustAI personalizes and experiments within agreed rules.
  3. Marketers review conversion performance and choose the next test.
Before and after

BeforeFixed creative and send sequence

AfterPersonalized journeys with ongoing experimentation

Human oversight
Marketers define the message, audience rules and success metric, then review the results and choose the next test.
Impact
Rollout projection, not booked revenue. 12,356 baseline applications × 30% lift = 3,706.8 incremental applications (shown as 3,707); × 12% funding rate = 444.816 incremental funded transactions (shown as 445); × $8,000 revenue per funded transaction = $3,558,528 in projected incremental annual revenue, shown as $3.56M. The calculation uses unrounded counts. The lift comes from JustAI's case study of a shopping-journey test: application-start rate moved from 9.25% to 12.08%, +2.83 percentage points or about 30.6% relative lift. The model applies 30% and assumes that lift carries across the annual baseline. Creative and cadence changed alongside personalization, so the full result can't be attributed to AI alone.
Inputs, roles and dependencies
Lifecycle marketing, led by the journey owner. Inputs: approved creative, customer signals and conversion events. Dependency: JustAI integrated with Iterable.
Tools and roles
JustAI runs ongoing personalization and experimentation. Iterable handles email delivery and journey integration. RudderStack and Claude are also part of this workflow's toolkit.
Evidence
Actual journey and results visuals are in JustAI's public case study. Review before reuse.
Takeaway
Optimize for the customer action that creates value, then follow it through to funding.
Classification
LifecycleExperimentation and personalizationMonitored automation
Sources
JustAI public case study; vendor comparison (Sep 8, 2026). Baseline, rollout lift, funding rate and revenue per funded transaction are internal planning assumptions.
Marketing operationsAcquisition spend reconciled to the invoiceFeatured projectInvoice reconciliation and financial QC moved to a Claude project that matches acquisition spend to Ramp invoices using vendor knowledge it keeps between cycles. Marketing works the exceptions, Finance receives a reconciled workbook, and the process caught $383K in costs.$383Kcosts caughtIntegrated workflowClaudeRampQuickSightLookerSharePoint
Objective
Make acquisition spend accountable down to the invoice, with a clear view of exceptions for Marketing and Finance.
Starting point
The monthly reconciliation of acquisition costs against vendor invoices was slow, fragile and undocumented, leaving the team chasing invoices, interpreting billing differences and rebuilding vendor context every month.
How it works
  1. Compare acquisition costs with Ramp invoices using retained vendor mappings.
  2. Review gaps, rejected bills and material variances, and resolve missing invoice information.
  3. Deliver the QC workbook to Finance and keep confirmed billing knowledge for the next cycle.
Before and after

BeforeManual invoice search and exception handling

AfterInvoice-level QC, reviewed exceptions and a Finance handoff

Human oversight
Marketing investigates exceptions. Finance receives the reconciled workbook and any open items. Confirmed vendor knowledge carries into the next run.
Impact
$383,000 in costs caught through acquisition-cost reconciliation. This is not a recurring annual figure, it is shown separately from annual savings, and costs caught are not the same as amounts recovered. Time saved: 78–102 hours a year, about an 85% reduction; the capacity estimate uses the upper end.
Inputs, roles and dependencies
Marketing operations, Acquisition and Finance. Inputs: spend exports, invoices and vendor mappings. Dependencies: QuickSight, Ramp and the reconciliation project.
Tools and roles
Claude Project holds the vendor mappings, billing history and reconciliation instructions. QuickSight supplies the cost CSV. The Ramp connector looks up invoices. SharePoint stores the QA workbook. Looker is also part of this workflow's toolkit.
Build and upkeep
Maintain vendor names, billing patterns and unresolved exceptions after each reconciliation.
Takeaway
Financial control improves when every exception has evidence, an owner and a resolution path.
Date
May 2026
Classification
Marketing operationsQuality and controlIntegrated workflow
LifecycleEmail subscription replaced by a custom editorA Claude-built Chrome editor with live preview, reusable snippets and click-to-code navigation replaced a $5,000-a-year Litmus subscription with a tool made for how the team actually builds email.$5Ka year savedRepeatable workflowClaudeChrome
Objective
Replace recurring tool expense with an editor designed for the team's email workflow.
Starting point
Building and testing email required a Litmus subscription, and the team still lacked a lightweight editor tailored to its workflow.
How it works
  1. Build with live preview and reusable snippets.
  2. Jump from the preview to the matching code with click-to-code navigation.
  3. Save HTML emails straight to shared folders.
Before and after

BeforeSubscription-dependent email tooling

AfterTailored Chrome editor with preview and HTML storage

Human oversight
The builder edits and previews each email.
Impact
$5,000 a year in Litmus subscription expense eliminated. Together with the content agency, this makes up the $405K+ in annual cost savings.
Inputs, roles and dependencies
Lifecycle, used by the email production specialist. Inputs: email HTML and reusable snippets. Dependencies: Chrome and folder access.
Tools and roles
Claude was the build tool. Chrome is the editor environment. HTML folders store the emails.
Takeaway
Build a focused tool when its ongoing value exceeds its operating cost.
Date
July 2026
Classification
LifecycleProductionRepeatable workflow
AffiliatesAffiliate marketer performance insight engineFeatured projectOne Claude-built dashboard shows affiliate lead volume by purpose, performance and trends at a glance, replacing analysis that used to be rebuilt by hand every day.520hours a yearRepeatable workflowClaudeQuickSightLooker
Objective
Give the acquisition team a usable view of affiliate lead mix, performance and trends without rebuilding the analysis every day.
Starting point
Understanding affiliate performance meant assembling the picture by hand, again and again.
How it works
  1. Open the affiliate dashboard.
  2. Review lead purpose, performance and trends.
  3. The channel owner decides what to do next.
Before and after

BeforeAssemble reports to understand performance

AfterReview the channel in one dashboard

Human oversight
The channel owner interprets the view and decides the actions.
Impact
1–2 hours a day, or 260–520 hours a year at 260 working days. The capacity estimate uses the upper end, 520.
Inputs, roles and dependencies
Affiliates, used by the channel analyst or manager. Dependency: QuickSight access.
Tools and roles
A Claude artifact provides the dashboard interface. QuickSight is the documented analytics source. Looker is also part of this workflow's toolkit.
Takeaway
Design reporting around the decisions a channel owner actually has to make.
Date
June 2026
Classification
AffiliatesReporting and decision supportRepeatable workflow
Marketing operationsAutonomous direct-mail campaign managementFeatured projectEvery direct-mail batch runs from one source-of-truth campaign record. When it changes, Claude and Cowork fan the update out to the calendar, status tracker, budget tracker, change log and sales kickoff materials, while people sign off every legal, creative and financial decision.360hours a yearMonitored automationClaude, Claude CoworkConfluenceJiraRampAsanaSharePointDataGrip
Objective
Automate campaign coordination from one shared record, with people approving legal, creative and financial decisions.
Starting point
Each batch runs a long lifecycle: dates and rates, creative and legal sign-off, vendor invoices, data files, sales materials and post-send reconciliation. Its details had to be kept in sync across multiple files and owners.
How it works
  1. Update the single campaign record first.
  2. Fan the change out to an interactive calendar, a status tracker with pre- and post-send checklists, a budget tracker by batch and payment month, an append-only change log, and sales kickoff decks and Confluence pages.
  3. People handle approvals and exceptions; performance data has its own refresh.
Before and after

BeforeBatch details managed across documents

AfterOne campaign record feeding coordinated operating views

Human oversight
The campaign owner approves status, compliance and send confirmation. The performance operator owns daily input files and refresh checks.
Impact
About 360 hours a year saved, 95% faster campaign administration, and sync errors down from about 30% to near zero, per the team log. Performance updates and kickoff-page benefits are counted here, not added again.
Inputs, roles and dependencies
Marketing operations, with campaign and performance owners. Inputs: batch details, approval status, invoice information and aggregate performance files.
Tools and roles
Claude and Cowork hold campaign state, generate the artifacts and run scheduled tasks. Confluence and Jira carry reviewed pages and legal tickets. The Ramp connector looks up invoices. SharePoint holds shared files. DataGrip runs the current manual matchback query. Asana carries data tickets, as documented in the SOP.
Build and upkeep
The SOP documents path, schema and HTML-shell fragilities, and owners check digests and output. Performance ingestion can silently produce zeros after column changes, so it is watched.
Takeaway
Scale coordination through shared state and clear ownership.
Classification
Marketing operationsKnowledge and coordinationMonitored automation
Marketing operationsDaily commercial performance viewChannel exports become a daily funnel report in Claude, putting prior-day and month-to-date performance, pacing and talking points in one place for the team's standup.292hours a yearRepeatable workflowClaudeQuickSightLookerGitHub
Objective
Bring channel performance, pacing and talking points into the daily marketing discussion.
Starting point
Seeing prior-day and month-to-date performance meant checking several places before the funnel could even be discussed.
How it works
  1. Export channel data from QuickSight.
  2. Generate the standup report in Claude.
  3. Review the numbers and insights before the meeting.
Before and after

BeforeMultiple sources and recurring report assembly

AfterChannel exports into one shared performance report

Human oversight
The operator exports and uploads the source files, checks the report and prepares the discussion.
Impact
About 292 hours a year saved in report preparation.
Inputs, roles and dependencies
Marketing operations, run by the daily-report operator. Inputs: channel CSVs and an optional direct-mail workbook. QuickSight and project access required.
Tools and roles
QuickSight supplies the source exports. A Claude Project produces the standup analysis and HTML report. An optional direct-mail workbook adds campaign performance. Looker and GitHub are also part of this workflow's toolkit.
Takeaway
Make the operating discussion easier without treating generated analysis as a substitute for review.
Date
June 2026
Classification
Marketing operationsReporting and decision supportRepeatable workflow
Marketing operationsSales licensing data kept currentA Claude in Chrome prompt updates sales licensing information live, replacing a manual, repetitive update process.288hours a yearAssisted taskClaude in ChromeChrome
Objective
Keep sales licensing information current with less manual administration.
Starting point
Licensing data required repetitive updates in the browser.
How it works
  1. Use the Claude browser prompt to update licensing information.
Before and after

BeforeManual licensing updates

AfterBrowser-assisted updates

Impact
288 hours a year saved, per the team tracker.
Tools and roles
Claude in Chrome running in Chrome.
Takeaway
Use AI for repetitive administration where accuracy can be checked.
Date
April 2026
Classification
Marketing operationsQuality and controlAssisted task
DesignOn-brand presentations without manual formattingA Claude brand-deck skill does three jobs to the brand standard: rebrands off-brand decks, builds new presentations from scratch and outlines new decks.192hours a yearRepeatable workflowClaudePowerPoint
Objective
Give the team a faster way to rebrand decks, create new presentations and develop outlines.
Starting point
Decks arrived off brand and had to be reworked by hand to the brand colors and standard.
How it works
  1. Choose rebranding, a new deck or an outline.
  2. Use the brand-deck skill to build the output.
Before and after

BeforeManual formatting and deck assembly

AfterReusable brand instructions producing the presentation

Impact
192 hours a year: 96 for rebranding, 48 for new decks and 48 for outlines. A frequency-based estimate that assumes four uses a month, not a timed annual observation.
Tools and roles
A Claude skill holds the brand instructions. PowerPoint is the output.
Takeaway
Keep the brand system reusable so presentation work starts closer to finished.
Classification
DesignReporting and decision supportRepeatable workflow
Marketing operationsCampaign tracking audited end to end before launchFeatured projectClaude in Chrome works through every campaign call-to-action in staging, confirms each one routes to the right application journey with tracking intact, and logs pass/fail evidence for the team to act on.6 hr to 15 minper tracking auditIntegrated workflowClaude, Claude in ChromeChromeExcelGitHub
Objective
Check that campaign clicks reach the right application journey and keep the right attribution.
Starting point
Checking every landing-page call-to-action by hand took 20–30 minutes a page and invited human error. A page could look correct while its links misrouted a lead or lost its attribution.
How it works
  1. Check each campaign call-to-action URL.
  2. Verify routing and persisted tracking in staging.
  3. Review the audit workbook, route fixes and commit the run log.
Before and after

BeforeManual page-by-page checks

AfterA repeatable end-to-end audit with clear exceptions

Human oversight
The operator confirms access, reviews failures, assigns fixes and commits the log by hand. The SOP describes a human-gated workflow.
Impact
Audit runs went from 4–6 hours to 15–20 minutes. The 6 hr → 15 min headline pairs the longest original time with the fastest new time, so these are range endpoints, not a paired measurement. At two runs a month, annual audit time falls from 96–144 hours to 6–8: 88–138 hours saved, a 90–95% reduction. The capacity estimate uses the upper end, 138.
Inputs, roles and dependencies
Marketing operations, run by the campaign QA owner. Inputs: campaign name and staging page URLs. Dependencies: a connected browser, VPN, staging admin access and GitHub access.
Tools and roles
A Claude Project and Claude in Chrome run the repeatable browser audit. The staging admin confirms persisted fields. Excel holds the audit output. GitHub keeps the cumulative run log, committed manually.
Build and upkeep
Maintain routing and tracking rules, browser access, VPN and the manual log. Review false positives and missed defects.
Takeaway
Treat attribution integrity as part of launch quality.
Date
July 2026
Classification
Marketing operationsQuality and controlIntegrated workflow
AffiliatesAffiliate budget pacing made visibleA Claude budget-pacing project shows affiliate spend against budget, so the team can act before spend gets ahead of the plan.130hours a yearRepeatable workflowClaudeQuickSightLooker
Objective
Make budget pacing visible before spend gets ahead of the plan.
Starting point
Affiliate pacing had no clear, accessible view.
How it works
  1. Use the budget-pacing project.
  2. Review current spend against budget.
Before and after

BeforePull pacing information together by hand

AfterA recurring pacing view

Impact
30 minutes a day saved, or 130 hours a year at 260 working days.
Tools and roles
Claude and QuickSight are the documented stack. Looker is also part of this workflow's toolkit.
Takeaway
Spend decisions improve when pacing is easy to inspect.
Date
July 2026
Classification
AffiliatesReporting and decision supportRepeatable workflow
LifecycleWeekly newsletter without the weekly rebuildFeatured projectA Claude project finds new content on Better.com, drafts the newsletter in its recurring format and populates the Iterable template, so each issue starts from a built draft instead of a blank one.120hours a yearIntegrated workflowClaudeIterableRudderStackFigma
Objective
Find new content, draft the newsletter and move it into Iterable without rebuilding the weekly process.
Starting point
Content discovery, drafting and template assembly repeated every week.
How it works
  1. Find relevant new Better.com content through the Claude project.
  2. Draft the newsletter in the recurring format.
  3. Populate the Iterable template.
Before and after

BeforeSource content, draft, then manual template assembly

AfterContent discovery, draft and a populated template

Human oversight
The newsletter owner operates the workflow and sets the editorial direction.
Impact
10 hours a month saved, annualized to 120 hours a year. It also reduced the newsletter's dependency on the content team.
Inputs, roles and dependencies
Lifecycle marketing, run by the newsletter owner. Inputs: new content, editorial direction and an Iterable template.
Tools and roles
A Claude Project handles content discovery and drafting. Better.com is the source content. Iterable is the template destination. RudderStack and Figma are also part of this workflow's toolkit.
Takeaway
Use the recurring format to cut assembly time and protect the publishing cadence.
Date
May 2026
Classification
LifecycleProductionIntegrated workflow
LifecycleEmail templates built from approved designsFeatured projectA Claude production skill turns an approved Figma design into a published Iterable template, cutting a 30-minute build to 2–3 minutes.30 min to 2 minper email buildIntegrated workflowClaudeFigmaIterableRudderStack
Objective
Turn approved creative into launch-ready email templates without rebuilding the same design by hand.
Starting point
Manual template production slowed the handoff from design to delivery.
How it works
  1. Start with the approved Figma design.
  2. Run the production skill to build the Iterable template.
Before and after

BeforeApproved design, then a manual build

AfterApproved design, production skill, Iterable

Human oversight
Every build starts from a design that has already been approved.
Impact
Build time went from 30 minutes to 2–3 minutes per email. The 2-minute headline uses the fastest new time, so it is a range endpoint, not a paired measurement. A 32-email pilot saved 14–15 hours; pilot hours are not annualized.
Inputs, roles and dependencies
Lifecycle, run by the email production specialist, who also built the skill. Input: an approved Figma design. Dependencies: Figma and Iterable access.
Tools and roles
A Claude skill holds the production instructions. Figma supplies the approved design. Iterable is where the template is published. RudderStack is also part of this workflow's toolkit.
Takeaway
Automate a repeatable handoff once the creative decision has been made.
Date
August 2026
Classification
LifecycleProductionIntegrated workflow
LifecycleNewsletter rate charts without the design queueA Claude-built artifact turns the approved daily-rate file into a publication-ready newsletter chart in minutes.30 min to 3 minper chartRepeatable workflowClaudeExcel
Objective
Turn approved rate data into a publication-ready chart in minutes.
Starting point
The recurring chart needed manual design effort every week.
How it works
  1. Generate the chart from the daily-rates file.
Before and after

BeforeManual weekly chart production

AfterStructured rates into reusable chart generation

Impact
Chart production fell from 30 minutes to 3 minutes, 90% less task time, and it removes the need for design resources.
Inputs, roles and dependencies
Lifecycle marketing, run by the newsletter owner. Input: the approved daily-rate file.
Tools and roles
A Claude-built artifact generates the chart. Excel holds the daily-rate input.
Takeaway
Start with a recurring output whose accuracy is easy to check.
Date
April 2026
Classification
LifecycleProductionRepeatable workflow
Marketing operationsStandard operating procedures in one formatA Claude skill drafts standard operating procedures in one shared Confluence format, so procedures are easier to write, follow and hand over.19SOPs standardizedRepeatable workflowClaudeConfluence
Objective
Make the team's SOPs easier to create, follow and hand over.
Starting point
Procedure quality and formatting varied by author.
How it works
  1. Use the SOP-building skill to draft the procedure in the shared format.
Before and after

BeforeAuthor-specific procedure documents

AfterA reusable SOP format and consistent documentation

Impact
About 6.5 hours saved across 19 SOPs, a one-time batch excluded from annual hours. The 19 now share one house format, 63% of all SOPs and about 80% of those authored, and defects such as dead links and format misses fell to near zero.
Tools and roles
A Claude skill handles SOP drafting and formatting. Confluence holds the procedures.
Takeaway
Make recurring work teachable before trying to scale it.
Classification
Marketing operationsReporting and decision supportRepeatable workflow
Marketing operationsSales kickoff pages for direct-mail campaignsA Claude skill turns campaign details into a consistent sales kickoff page in Confluence, making the sales handoff a standard part of every launch.60 minsaved per kickoff pageIntegrated workflowClaudeConfluence
Objective
Turn campaign details into a consistent sales kickoff page.
Starting point
Kickoff documentation was assembled by hand for each campaign.
How it works
  1. Use the kickoff-page skill.
  2. Create the Confluence page from campaign information.
Before and after

BeforeManual kickoff-page creation

AfterReusable campaign-to-sales documentation

Impact
60 minutes saved per page, a source-estimated 90% reduction. No annual volume is recorded, so this is not annualized or added to direct-mail savings.
Tools and roles
A Claude skill creates the kickoff page. Confluence carries the sales handoff.
Takeaway
Make the sales handoff a repeatable part of launching a campaign.
Classification
Marketing operationsProductionIntegrated workflow
Paid mediaAcquisition reporting built around the decisionChannel owners use Claude to build exactly the performance view they need, then hand the analyst a working example instead of a request for a comprehensive report.The team's primary way to review performanceRepeatable workflowClaude
Objective
Give the channel owner a focused performance view and a concrete brief for the analyst.
Starting point
Analyst capacity was stretched across platforms and reporting requests.
How it works
  1. Use Claude to build the specific view needed.
  2. Share the working example with the analyst.
Before and after

BeforeAsk for a comprehensive report

AfterBuild the needed view and share an actionable example

Impact
The team tracker identifies these focused views as the team's primary way of reviewing performance. Decision speed and savings are not yet measured.
Tools and roles
Claude creates the focused performance views.
Takeaway
A precise working example improves the reporting handoff.
Classification
Paid mediaReporting and decision supportRepeatable workflow
Marketing operationsEngineering requests with the business case built inA four-step guided intake built in Claude turns a marketing need into a formatted engineering ticket with the business problem, quantified impact, KPI and scope, reviewed and approved before it reaches Jira.Structured, reviewed requestsRepeatable workflowClaudeJiraConfluence
Objective
Turn a marketing need into a business-grounded engineering brief before it reaches the development queue.
Starting point
Ticket quality depended on who wrote the request: formats varied, business objectives were often missing and scope was uneven.
How it works
  1. Capture the business problem, impact, KPI and requirements.
  2. Generate the brief and resolve scorecard issues.
  3. Review and approve the request before an authorized Jira submission.
Before and after

BeforeUneven requests and inconsistent scope

AfterStructured intake and a reviewed business case

Human oversight
The requester reviews the brief, marketing operations reviews it, the SVP approves it, and an authorized submitter creates the Jira ticket.
Impact
Structured intake and review are documented in the SOP. Request-quality gains and time saved are not yet measured.
Inputs, roles and dependencies
Marketing operations and engineering requesters. Inputs: the business problem, quantified impact, KPI, scope and dependencies.
Tools and roles
A Claude artifact handles intake and brief generation. Confluence is used for review. Jira receives the authorized manual submission.
Takeaway
Improve the handoff by making the commercial case and acceptance criteria explicit.
Date
February 2026
Classification
Marketing operationsKnowledge and coordinationRepeatable workflow
Content marketingCustomer personas grounded in first-party evidenceA Claude project analyzed funded-loan data and customer transcripts to build refinance-segment profiles and personas, giving marketing testable hypotheses about who to target and how to speak to them.Refinance personas from a completed pilotRepeatable workflowClaudeQuickSightLooker
Objective
Use customer evidence to sharpen who marketing targets and how it speaks to them.
Starting point
Better's highest-value customer segments weren't defined, and neither was how to speak to them.
How it works
  1. Analyze funded-loan data and customer transcripts in the Claude project.
  2. Develop customer profiles and personas.
Before and after

BeforeBroad assumptions about the audience

AfterData-grounded profiles and communication hypotheses

Human oversight
The output is treated as a hypothesis for marketing to test, not a validated performance lift.
Impact
A completed refinance-segment pilot produced customer profiles and personas. Marketing adoption and commercial lift are not yet measured.
Tools and roles
A Claude Project builds the customer profiles and personas. QuickSight and first-party records are the documented data inputs; customer transcripts add language and motivation. Looker is also part of this workflow's toolkit.
Takeaway
Use first-party evidence to inform messaging, then test the hypotheses it produces.
Date
February 2026
Classification
Content marketingAudience intelligenceRepeatable workflow
Content marketingOne workflow for rate, news and search contentOne Claude project supports daily rate commentary, timely story pitches and keyword-led articles.Three content modes, one workflowRepeatable workflowClaude
Objective
Support rate commentary, timely story ideas and search-led articles through one content workflow.
Starting point
Content was going stale, and researching and writing new pieces took time across several recurring formats.
How it works
  1. Choose rate commentary, trending ideas or keyword-led content.
  2. Generate the output in the Claude project.
Before and after

BeforeResearch and draft each format separately

AfterReusable content modes producing drafts and pitches

Impact
Three content modes for rate posts, timely pitches and search articles. No separate savings are added beyond the content agency result.
Tools and roles
A Claude Project produces daily-rate posts, trending pitches and keyword-led drafts.
Takeaway
Package repeatable content formats while keeping sources and editorial judgment visible.
Classification
Content marketingProductionRepeatable workflow
LifecycleBrand-voice first drafts for lifecycle emailA Claude email-writing skill turns a campaign brief into a first draft in Better's brand voice, so no send starts from a blank page.A brand-voice draft for every sendRepeatable workflowClaude
Objective
Give every campaign a stronger starting draft in Better's brand voice.
Starting point
Each send began as another blank-page copy exercise.
How it works
  1. Provide the campaign brief.
  2. Use the email-writing skill to draft copy.
Before and after

BeforeDraft each email from scratch

AfterReusable writing instructions producing a campaign draft

Impact
A reusable starting point for lifecycle copy. Time, quality and conversion impact are not yet quantified.
Tools and roles
A Claude skill holds the email-copy instructions in Better's brand voice.
Takeaway
Encode the brand rules so the team can focus on the message.
Classification
LifecycleProductionRepeatable workflow
LifecycleLifecycle campaign guides in one shared formatA Claude campaign-guide skill creates a structured operating brief for each lifecycle campaign in Confluence, so knowledge carries from one campaign to the next.Standard campaign briefs in ConfluenceIntegrated workflowClaudeConfluence
Objective
Give lifecycle campaigns a consistent operating brief in Confluence.
Starting point
Campaign guides varied in format and had to be assembled again each time.
How it works
  1. Use the campaign-guide skill.
  2. Create the structured guide in Confluence.
Before and after

BeforeIndividually assembled campaign guides

AfterA reusable guide format and a shared campaign record

Impact
Standard campaign guides in Confluence. Handoff time and consistency are not yet measured.
Tools and roles
A Claude skill generates the campaign guide. Confluence holds the shared campaign documentation.
Takeaway
Standard documentation makes work easier to hand over.
Classification
LifecycleKnowledge and coordinationIntegrated workflow
LifecycleRate-drop alert decision supportA configured GPT assesses week-over-week rate movement to help marketers decide when a rate drop is a timely reason to contact customers.A configured GPT behind each alert callRepeatable workflowChatGPT
Objective
Help marketers decide when a rate move creates a timely reason to contact customers.
Starting point
Week-over-week rate monitoring had to be translated into an alert decision.
How it works
  1. Use the Rate Drop Checker GPT to assess the rate movement.
  2. A marketer decides whether to issue an alert.
Before and after

BeforeMonitor rates, then decide on an alert

AfterConfigured decision support, then a marketer's decision

Human oversight
A marketer decides whether the rate movement warrants an alert.
Impact
Supports the decision to issue a rate-drop alert. Accuracy, time savings and commercial lift are not yet quantified.
Tools and roles
ChatGPT, through a configured GPT, provides rate-alert decision support.
Takeaway
Define the commercial trigger before automating the response.
Classification
LifecycleReporting and decision supportRepeatable workflow
LifecycleMessaging matched to market eventsEvent-aware copy and matching logic are designed to send each contact the right email variant when a market event fires, instead of one generic email to everyone, built from reviewed snippets and defined audience rules.Event-aware outreach designAssisted taskClaudeIterableJustAIRudderStack
Objective
Connect market changes to timely customer communication.
Starting point
Market events triggered one generic email to everyone, and overlapping triggers could hit the same contact in the same week.
How it works
  1. Develop event-aware copy and matching logic.
  2. Use reviewed snippets and defined audience rules.
Before and after

BeforeGeneric market-event outreach

AfterEvent context, approved message, defined audience

Human oversight
Messages draw on reviewed snippets and defined audience rules.
Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Tools and roles
Claude, Iterable and JustAI are the documented stack. RudderStack is also part of this workflow's toolkit.
Takeaway
Define the event, the audience and the approval boundary together.
Date
July 2026
Classification
LifecycleExperimentation and personalizationAssisted task
Paid mediaLead sources grouped for clean reportingClaude's pattern recognition groups a high volume of incoming lead sources into a consistent reporting structure.A reviewed source mapRepeatable workflowClaude
Objective
Group incoming lead sources into a usable reporting structure.
Starting point
High source-lead volume made classification difficult.
How it works
  1. Provide the source data to Claude.
  2. Use pattern recognition to group sources.
Before and after

BeforeHigh-volume source classification

AfterPattern-assisted grouping and a reviewed source map

Impact
Source-mapping creation is documented. Classification accuracy and time saved are not yet measured.
Tools and roles
Claude provides pattern-assisted source grouping.
Takeaway
Keep reporting categories consistent as source volume grows.
Classification
Paid mediaReporting and decision supportRepeatable workflow
Paid mediaAcquisition pacing checked against targetClaude reviews 7-, 14- and 30-day averages and flags pacing issues against the monthly target.Recurring checks against the monthly targetRepeatable workflowClaude
Objective
Surface performance trends and pacing issues against the monthly target.
Starting point
Routine reporting and troubleshooting required the same calculations again and again.
How it works
  1. Use Claude to review 7-, 14- and 30-day averages.
  2. Compare pacing with the monthly target and flag issues.
Before and after

BeforeManual averages and pacing checks

AfterRecurring performance analysis and issues to review

Impact
Recurring performance and pacing analysis is documented. Commercial lift and savings are not yet measured.
Tools and roles
Claude handles averages, pacing and troubleshooting analysis.
Takeaway
Build recurring checks around the targets that guide action.
Classification
Paid mediaKnowledge and coordinationRepeatable workflow
Marketing operationsDaily direct-mail performance refreshClaude Cowork refreshes the direct-mail performance dashboard daily: it backs up the dashboard, appends the day's matchback and call data, rebuilds the analysis tables and validates every formula.Matchback and call data refreshed dailyRepeatable workflowClaude CoworkDataGripExcel
Objective
Keep matchback and call reporting moving through a repeatable daily refresh.
Starting point
Performance review depended on assembling several documents and inputs.
How it works
  1. Supply the daily call and matchback files.
  2. Cowork backs up the dashboard, appends the new data, rebuilds the analysis tables and validates formulas.
  3. Review the updated workbook and check for missing inputs or schema changes.
Before and after

BeforeManual aggregation and workbook updates

AfterFile inputs into a repeatable performance refresh

Human oversight
The performance operator supplies the files and checks the refresh. Missing inputs and column changes are investigated.
Impact
Its time benefit is counted within direct-mail campaign management, not added again.
Inputs, roles and dependencies
Marketing operations, run by the performance operator. Inputs: aggregate call and matchback files. Shared storage and query access required.
Tools and roles
Claude Cowork runs the repeatable refresh. DataGrip runs the current matchback query. Shared files and Excel hold the inputs and the report.
Takeaway
Separate campaign administration from performance-data ownership.
Classification
Marketing operationsKnowledge and coordinationRepeatable workflow
Marketing operationsForward-looking budget and forecast planningAn exploration of a Claude and QuickSight workflow to extend backward-looking reporting into forward-looking budget decisions.A forward-looking planning approachAssisted taskClaudeQuickSightLooker
Objective
Extend current reporting into forward-looking budget decisions.
Starting point
The team wanted a planning view beyond backward-looking performance reports.
How it works
  1. Explore a Claude and QuickSight budget workflow.
Before and after

BeforeHistorical performance views

AfterA forward-looking budget and forecast approach

Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Tools and roles
Claude and QuickSight are the documented stack. Looker is also part of this workflow's toolkit.
Takeaway
Agree how the forecast will change a decision before building the model.
Date
July 2026
Classification
Marketing operationsReporting and decision supportAssisted task
Paid mediaPaid-media spend planning from unit economicsClaude ingested historical spend and plotted daily spend within the eCAC threshold after applying diminishing returns. The daily plan proved too granular: the agency needed room to pace, so it kept that control.An economics-informed monthly planAssisted taskClaude
Objective
Use historical economics to inform the monthly spend plan.
Starting point
Monthly planning needed a view of spend within the acquisition-cost threshold.
How it works
  1. Analyze historical spend and diminishing-return patterns in Claude.
  2. Explore a daily spend curve within the target threshold.
  3. Keep the agency's freedom to manage pacing.
Before and after

BeforeHistorical spend review

AfterEconomics-informed planning with flexible pacing

Human oversight
The agency keeps control of day-to-day pacing.
Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Tools and roles
Claude is the documented stack.
Takeaway
The most granular model is not always the best operating plan.
Classification
Paid mediaReporting and decision supportAssisted task
AffiliatesAffiliate analysis in code, within access limitsWe explored running affiliate analysis in code through Claude Code and the AWS CLI. The access request was declined as far broader than the task needed, so analysis stayed in the standard tooling.An access boundary defined up frontAssisted taskClaude CodeAWS CLI, QuickSightLooker
Objective
Make affiliate reporting reproducible and version-controlled by working with QuickSight in code, not only through the interface.
Starting point
QuickSight changes relied on one team member, and the interface had a learning curve.
How it works
  1. Request scoped AWS and QuickSight access.
  2. Assess the permission boundary with the data owner.
  3. The request was declined as too broad an access grant, so analysis stayed in the standard reporting tools.
Before and after

BeforeFront-end analytics interaction

AfterProposed CLI analysis within an access boundary

Human oversight
Access was reviewed against the permission boundary and declined.
Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Tools and roles
Claude Code and the AWS CLI are the documented stack, against QuickSight. Looker is also part of this workflow's toolkit.
Takeaway
Access scope is an implementation constraint, not a detail to discover at the end.
Date
July 2026
Classification
AffiliatesReporting and decision supportAssisted task
Marketing operationsEngineering-request quality reviewA Claude reviewer checked marketing requests for clarity and completeness, then moved into the request-intake tool instead of living as a separate step.Folded into request intakeAssisted taskClaude
Objective
Check marketing requests before they reach the engineering queue.
Starting point
Reviewing individual tickets took recurring effort.
How it works
  1. Review the request for clarity and completeness.
  2. Carry the review capability into the request accelerator.
Before and after

BeforeSeparate ticket review

AfterQuality review inside the structured request workflow

Impact
No separate financial or time benefit is assigned. The capability moved into the engineering-request intake.
Tools and roles
Claude is the documented stack.
Takeaway
Build review into the handoff rather than maintaining another disconnected tool.
Classification
Marketing operationsQuality and controlAssisted task
Marketing operationsLaunch reporting for a new direct-mail programClaude and DataGrip produced a performance dashboard for a newly launched FHA Streamline direct-mail program. The daily Cowork performance refresh later took over that job.Folded into the daily refreshAssisted taskClaudeDataGrip
Objective
Give a newly launched direct-mail program a clear performance view.
Starting point
The channel needed a way to track launch performance.
How it works
  1. Develop the performance dashboard with Claude and DataGrip.
  2. Connect the reporting need to the recurring direct-mail performance workflow.
Before and after

BeforeNew-program performance tracking

AfterA dashboard approach feeding broader performance operations

Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Tools and roles
Claude and DataGrip are the documented stack.
Takeaway
Build one reporting capability rather than parallel versions of the same analysis.
Date
March 2026
Classification
Marketing operationsReporting and decision supportAssisted task
Marketing operationsEngineering sprints read for marketing impactClaude Cowork archived engineering sprint updates and compared planned work with marketing needs. It was only as useful as the upstream updates feeding it.Sprint plans tied to marketing needsAssisted taskClaude Cowork
Objective
Connect engineering sprint plans with their implications for Growth Marketing.
Starting point
The team wanted visibility into planned engineering work and its marketing impact.
How it works
  1. Archive the sprint updates in a repeatable workflow.
  2. Compare the planned work with the marketing need.
  3. The approach depends on recurring upstream updates.
Before and after

BeforeRecurring sprint updates

AfterA record of plans and their marketing implications

Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Tools and roles
Claude Cowork is the documented stack.
Takeaway
An operating workflow is only useful while its source inputs are available.
Date
March 2026
Classification
Marketing operationsKnowledge and coordinationAssisted task
LifecycleAssistants connected across the rate-drop workflowWe mapped how Claude, ChatGPT, Zapier and Confluence could cover the full rate-drop workflow. Zapier couldn't reach Teams group chats without Claude API access, and Confluence asked for permission before every edit, so the design had to work within those limits.Integration and approval designAssisted taskClaudeChatGPTZapierConfluence
Objective
Explore how connected assistants can support the full rate-drop workflow.
Starting point
Tool access was split: ChatGPT had Microsoft Teams but not Iterable or JustAI, and Claude had the reverse, so no single assistant could run the whole rate-drop workflow.
How it works
  1. Explore routing through Claude, ChatGPT, Zapier and Confluence.
  2. Check connector, API and approval coverage.
  3. Design within the supported access and human approval steps.
Before and after

BeforeSplit tool access

AfterAn orchestration approach shaped by connector limits

Human oversight
Human approval steps are part of the design.
Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Tools and roles
Claude, ChatGPT, Zapier and Confluence are the documented stack.
Takeaway
Confirm the integration path before promising end-to-end automation.
Date
July 2026
Classification
LifecycleKnowledge and coordinationAssisted task
DesignAI creative explorationThe design team tested where AI can contribute to design and creative production, judging the tools against the brand and quality bar.AI tested in creative productionAssisted task
Objective
Test where AI can contribute to design and creative work.
Starting point
The design team was exploring AI-assisted creative production.
How it works
  1. Explore design prompts and creative outputs.
Before and after

BeforeThe existing creative process

AfterPrompt-based creative exploration

Impact
No separate financial or time benefit is assigned. The record captures the approach and its operating lessons.
Takeaway
Judge creative tools against the quality and brand standard.
Classification
DesignProductionAssisted task

The stack behind the work.

AI sits on top of the systems that already run marketing, from customer data and creative to finance and delivery. Claude’s formats share one mark. Select a tool to see where it’s used.

AI and experimentation

Where the reasoning, drafting, testing and automation happen.

Campaigns, creative and customer data

The systems customers actually experience.

  • TwilioCommunicationsIn the stack

Data and financial control

Where performance and spend are measured and checked.

Documentation and delivery

Where work is recorded, handed off, approved and shipped.

How the work is run.

Four principles held across the portfolio. Four lessons changed the plan, including work that was narrowed or consolidated once its limits were clear.

Principles that held

Lessons that changed the plan

Scope and measurement.

How each number is built, and what it doesn’t claim. JustAI’s public case study has the vendor’s own visuals.

Scope
A selection of AI work from January to September 2026. It isn’t all of the team’s AI work or its full marketing workload. Figures come from different measurement periods, and annual figures and projections are not realized January–September results.
Total value, $4.49M+
$405K+ annual savings + $383K costs caught + $3,558,528 projected revenue + 3,078 hours at $48 an hour ($147,744) = $4,494,272. It combines recurring, one-time and projected value; about 79% is the revenue projection.
Annual savings, $405K+
$400K+ in external content agency spend and $5K in tool subscription spend, both eliminated. No headcount savings are assigned.
Costs caught, $383K
Found through invoice reconciliation and financial QC. Shown as its own breakout because it isn’t recurring, and costs caught aren’t necessarily amounts recovered.
Broader cost impact, $788K+ a year
Annual savings plus costs caught within 12 months. A broader view of cost impact, not a savings total.
Projected revenue, $3.56M
12,356 baseline applications × 30% lift × 12% funding rate × $8,000 revenue per funded transaction = $3,558,528, calculated on unrounded counts. A rollout projection, not revenue generated. The lift draws on JustAI’s case study (9.25% to 12.08% application-start rate, about 30.6% relative), where creative and cadence also changed.
Capacity, 3,078 hours a year
The upper end of team estimates across ten recurring workflows. Capacity released, not headcount eliminated. Priced at $48 an hour ($147,744) only for the total value. Known overlapping components are excluded; some remaining overlap is unresolved.
Timing headlines
6 hr → 15 min and 30 min → 2 min pair the longest original time with the fastest new time reported. They are range endpoints, not paired measurements.
Visuals and evidence
Workflow diagrams are illustrative. The page contains no customer records, internal system links or source files.
Sources
Team tracker (Aug 6, 2026), AI use-cases log (Aug 14, 2026), operating SOPs, confirmation of agency savings (Oct 14, 2026), and JustAI’s public case study.