Austin Hay (Reforge, Ramp, Runway)13 August 2023

The ultimate guide to Martech

7Frameworks

14Insights

Episode overview

Austin Hay defines marketing technology as a cross-functional discipline connecting marketing, growth, product, engineering, people, processes, and first- or third-party systems. He explains when growing organizations need dedicated martech ownership, how B2C and B2B stacks differ, and why practitioners should design data collection and attribution infrastructure for future needs. The discussion also covers probabilistic attribution, hiring signals, organizational design, recommended technology stacks, and the problem-people-system, build-and-buy, and thinking-gray frameworks.

Key ideas

Marketing technology is best understood as product management for the systems and platforms that support marketing and growth.
Dedicated martech ownership commonly becomes necessary around 100–150 employees, when shared responsibility for tools, data flows, contracts, and permissions stops scaling.
B2C organizations can choose between a conventional CDP-centered stack and a warehouse-centered architecture using reverse ETL, depending on engineering maturity and resources.
Attribution infrastructure should capture referring URLs, advertising parameters, and first- and last-touch UTM values from the beginning so later multi-touch analysis has usable history.
As privacy restrictions, cookie blocking, and mobile tracking limits weaken deterministic attribution, marketers increasingly need probabilistic models and carefully controlled experiments.
Strong martech hires combine intellectual curiosity, engineering resourcefulness, cross-functional persuasion, and a tool-agnostic focus on solving problems.
Austin's PPS framework evaluates the problem and people before choosing a system, while build-and-buy replaces a rigid build-versus-buy choice.
Austin's suggested B2C stack includes Amplitude, Customer.io or Braze, Snowflake, Hightouch, and AppsFlyer or Branch.

Transcript available · source text is retained privately and is not published

Frameworks in this episode

Strategy4 steps

Build AND Buy

Reframe build-vs-buy as build-and-buy: buy 90% off the shelf, build the 10% that differentiates.

Read →

Strategy4 steps

CDP vs Reverse-ETL Stack Decision

Choose a centralized CDP stack or a warehouse-plus-reverse-ETL stack based on your engineering resources.

Read →

Marketing4 steps

Future-Proof MTA Data Collection

Instrument first and last UTMs on day one so you can move to multi-touch attribution without waiting.

Read →

Productivity3 steps

Problem-People-System (PPS)

Diagnose a systems request by problem and people first, never jump straight to the tool.

Read →

Mindset3 steps

Thinking Gray

Defer black-or-white judgments about decisions and people until the last responsible moment.

Read →

Leadership4 steps

Tool-Agnostic Hiring Test

Screen martech candidates for curiosity, engineering scrappiness, and freedom from tool bias.

Read →

Leadership4 steps

When to Centralize a Martech Function

Use the 100-150 person threshold and b2c/b2b axis to decide when and where to own martech.

Read →

People & resources mentioned

Attributed to the moment in the episode. Timestamps are approximate.

People · 11

Lenny RachitskyMentionstoday we've got another very special compilation episode something I've been pulling on more and more with the podcast and the newsletter
“welcome to Lenny's podcast where I interview world-class product leaders and growth experts”
▶ Watch at ~0:30approx.
Austin HayMentionstoday my guest is Austin Haye Austin is one of the smartest people in the world on the field of martech
“today my guest is Austin Haye Austin is one of the smartest people in the world on the field of martech”
▶ Watch at ~1:00approx.
Seiki ChanMentionsSeiki Chan who's the CEO of Runway
“Seiki Chan who's the CEO of Runway”
▶ Watch at ~9:00approx.
Cody MorganMentionshe hired a guy named Cody Morgan at ramp who has a user acquisition team
“he hired a guy named Cody Morgan at ramp who has a user acquisition team”
▶ Watch at ~27:00approx.
Sri BatchuMentionsShri was VP of opposite open door then head of growth at instacart and currently he's the head of growth at Ram
“I know SRI bachu is on here a while back he hired a guy named Cody Morgan at ramp”
▶ Watch at ~27:00approx.
Mike MolinetMentionsI remember working at Branch I was working for our CEO Mike Mulaney
“I remember working at Branch I was working for our CEO Mike Mulaney”
▶ Watch at ~37:00approx.
Mike TaylorMentionsit was somebody else I can't remember the name it was another mic Mike Taylor
“it was somebody else I can't remember the name it was another mic Mike Taylor”
▶ Watch at ~56:00approx.
Steven B. SampleCoinedSteven B sample is a professor at USC he wrote a book called The the contrarians guide to leadership
“Steven B sample is a professor at USC he wrote a book called The the contrarians guide to leadership”
▶ Watch at ~1:11:00approx.
Warren BennisCoinedthe art and adventure of leadership by the name Warren Bennis
“the art and adventure of leadership by the name Warren Bennis”
▶ Watch at ~1:14:00approx.
Adam FishmanCoinedI've been a chief product officer and VP of growth and product at a bunch of different companies imperfect Foods patreon Lyft
“I actually was recently listening to another podcast by Adam Fishman and he had Brian Balfour on”
▶ Watch at ~1:16:30approx.
Brian BalfourMentionsI actually was recently listening to another podcast by Adam Fishman and he had Brian Balfour on
“I actually was recently listening to another podcast by Adam Fishman and he had Brian Balfour on”
▶ Watch at ~1:16:30approx.

Resources · 47

Lenny's PodcastCoinedpodcast · Lenny Rachitsky
“welcome to Lenny's podcast where I interview world-class product leaders and growth experts”
▶ Watch at ~0:30approx.
RobinhoodMentionscompany
“he's advised companies like notion Airbnb Walmart Postmates Robin Hood even Pete's Coffee and Mars”
▶ Watch at ~1:00approx.
RampMentionscompany · Eric Glyman and Karim Atiyeh
“he's currently head of marketing technology at ramp”
▶ Watch at ~1:00approx.
RunwayMentionscompany · Runway
“before that his VP of Business operations at Runway”
▶ Watch at ~1:00approx.
mParticleMentionscompany
“before that he has VP of growth at M particle”
▶ Watch at ~1:00approx.
BranchMentionscompany
“the fourth employee at the Unicorn Branch metrics”
▶ Watch at ~1:00approx.
NotionMentionssoftware · Notion Labs
“he's advised companies like notion Airbnb Walmart Postmates Robin Hood even Pete's Coffee and Mars”
▶ Watch at ~1:00approx.
AirbnbMentionscompany · Brian Chesky, Joe Gebbia, and Nathan Blecharczyk
“he's advised companies like notion Airbnb Walmart Postmates Robin Hood even Pete's Coffee and Mars”
▶ Watch at ~1:00approx.
WalmartMentionscompany
“he's advised companies like notion Airbnb Walmart Postmates Robin Hood even Pete's Coffee and Mars”
▶ Watch at ~1:00approx.
PostmatesMentionscompany
“he's advised companies like notion Airbnb Walmart Postmates Robin Hood even Pete's Coffee and Mars”
▶ Watch at ~1:00approx.
ReforgeMentionscompany · Brian Balfour
“he's also a teacher at reforge on this very topic of martech”
▶ Watch at ~1:00approx.
OneSchemaMentionssoftware · Christina Gilbert
“today's episode is brought to you by one schema the embeddable CSV importer for SAS”
▶ Watch at ~2:00approx.
MixpanelMentionssoftware · Mixpanel
“this episode is brought to you by mixpanel get deep insights into what your users are doing at every stage of the funnel”
▶ Watch at ~3:00approx.
Google AnalyticsMentionssoftware · Google
“mixed panels built to be more powerful and easier to use than Google analytics”
▶ Watch at ~3:30approx.
SalesforceMentionscompany · Salesforce
“that's also where you typically see tools like Salesforce coming into play and more advanced crms”
▶ Watch at ~8:30approx.
SegmentMentionssoftware · Twilio
“you'll take a third party tool call it like a segment or an amplitude”
▶ Watch at ~11:00approx.
MarketoMentionssoftware · Adobe
“they need to move to like a brace or a Marketo”
▶ Watch at ~30:30approx.
dbtMentionssoftware · dbt Labs
“you are Cutting Edge and you're going to do a bunch of modeling and DBT and you already have snowflake”
▶ Watch at ~34:00approx.
RudderStackMentionssoftware
“rudderstack which is a CDP has always had a reverse ETL function”
▶ Watch at ~36:00approx.
HightouchUsessoftware
“I'm an investor in high touch love their work we use them at ramp”
▶ Watch at ~36:00approx.
CensusMentionscompany
“census which was Back Bay 16z and high touch are the two Standalone reverse etls”
▶ Watch at ~36:00approx.
OutreachUsessoftware
“how we would email them with a little tool called Outreach at the time”
▶ Watch at ~37:30approx.
GongMentionssoftware · Gong
“you have Salesforce and then the whole world and the universe revolves around Salesforce you just have more advanced tools you have like gong”
▶ Watch at ~38:00approx.
HubSpotMentionscompany · HubSpot
“having both HubSpot and Salesforce both are crms both have the ability to track users and companies neither are cdps”
▶ Watch at ~39:30approx.
Brave Search APIMentionstool · Brave Software
“today's episode is brought to you by Brave search and their newest product the brave search API”
▶ Watch at ~40:30approx.
Lenny's NewsletterMentionsnewsletter · Lenny Rachitsky
“if I went to Lenny's newsletter from a Google ad and that's all he has that would be my first touch”
▶ Watch at ~43:30approx.
SnowplowMentionssoftware
“there's only one CDP I know that does that that's snowplow”
▶ Watch at ~49:30approx.
RedditMentionswebsite · Steve Huffman, Alexis Ohanian, and Aaron Swartz
“Reddit is also a very interesting place to convert now they're opening up their conversions API”
▶ Watch at ~51:30approx.
RecastMentionscompany
“there's a company called recast maybe you're I think you're an investor”
▶ Watch at ~55:30approx.
SplitMentionssoftware · Split Software
“they were evaluating it to build the entire system themselves or buy a third party I think it was split.ir or something like that”
▶ Watch at ~1:09:00approx.
The Art and Adventure of LeadershipRecommendsbook · Warren Bennis
“second book that would be really good is the art and adventure of leadership by the name Warren Bennis”
▶ Watch at ~1:13:30approx.
The Contrarian's Guide to LeadershipRecommendsbook · Steven B. Sample
“first book I already mentioned uh but say it again it's the contrarian's guide to leadership awesome book”
▶ Watch at ~1:13:30approx.
SuitsRecommendstv show · Aaron Korsh
“I'm currently watching for the first time suits which I'd never seen before which I think is like pretty good”
▶ Watch at ~1:14:00approx.
SiloRecommendstv show · Hugh Howey
“if that's not your jam and you like excitement also watching Silo”
▶ Watch at ~1:14:30approx.
What We Do in the ShadowsRecommendstv show
“for just like really dumb comedy what we do in the shadows is hilarious”
▶ Watch at ~1:14:30approx.
Our Flag Means DeathRecommendstv show
“there's our flag means death which is hilarious have you seen that show”
▶ Watch at ~1:14:30approx.
CalendlyUsessoftware · Tope Awotona
“I've been a big calendly user for a long time but countley is pretty expensive”
▶ Watch at ~1:15:30approx.
Cal.comRecommendssoftware
“cal.com has not failed me it has been awesome so if people are looking for new calendly tools strongly recommend”
▶ Watch at ~1:16:00approx.
SnowflakeRecommendscompany · Snowflake Inc.
“I put everything in Snowflake I'd buy High test reverse CT all that data out to my ad networks”
▶ Watch at ~1:18:30approx.
BrazeRecommendssoftware
“I'd buy customer i o and maybe I'd upgrade to braze in the future”
▶ Watch at ~1:18:30approx.
AmplitudeRecommendssoftware · Amplitude
“if I was a b2c business I'd buy amplitude for my CDP”
▶ Watch at ~1:18:30approx.
Customer.ioRecommendssoftware
“I'd buy customer i o and maybe I'd upgrade to braze in the future”
▶ Watch at ~1:18:30approx.
AppsFlyerRecommendssoftware · AppsFlyer
“probably apps flyer from a mobile app if not Branch but it'd probably be apps flyer first”
▶ Watch at ~1:18:30approx.
HightouchRecommendssoftware
“then reverse ETL is high touch”
▶ Watch at ~1:19:30approx.
SkydioUsescompany
“I have a mavic air 2 and a skydio Enterprise”
▶ Watch at ~1:22:00approx.
DJI Mavic Air 2Usesproduct · DJI
“I have two drones I have a mavic air 2 and a skydio Enterprise”
▶ Watch at ~1:22:00approx.
ReforgeRecommendscompany · Brian Balfour
“I have a marketing technology course coming out with reforge in the fall”
▶ Watch at ~1:23:30approx.

Spot an error or want something removed? Request a correction or removal.

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster56:30

Most Companies Chasing MMM Actually Just Need MTA

Austin pushes back on the rush toward media mix modeling (MMM). When he evaluates whether a business is ready, he asks if they have the data and whether it will actually change their marketing decisions. Most of the time the answer is no: they don't really need MMM, they need better multi-touch attribution and better probabilistic modeling.

Many companies gravitate to MMM without understanding when it's useful
Key questions: do you have the data, and will it change decisions?
Most businesses aren't ready for MMM; they actually mean MTA
Better probabilistic modeling is usually the real need

“most of the time businesses are not ready for mmm they actually just mean MTA and they need better probabilistic modeling”
Austin Hay · 56:30

#attribution#mmm#mta#measurement

Hot Take· 2

Hot Take19:30

Why the Martech Person Is the Real Quarterback

Austin argues the most important trait in a martech hire isn't how many people they've managed, but how well they manage upward, laterally, and downward. Because the role sits between so many departments, it survives on persuasion and salesmanship rather than headcount. He calls the martech person a true cross-functional quarterback.

Martech teams are small, so cross-functional skill beats management experience
They must persuade rev-ops, product, and data leaders to make changes
The job is a game of persuasion and salesmanship, not authority
They call plays across departments like a quarterback

“I almost view them like a true quarterback every like says people are quarterbacks but really marketing technology because it lives between so many departments…”
Austin Hay · 19:30

#leadership#cross-functional#martech#influence

Hot Take1:23:00

Austin's Hot Take: Twitter Ruins Careers

Asked where to find him online, Austin says he's on LinkedIn and Threads but deliberately not on Twitter. His reasoning is a blunt hot take: he's seen multiple careers ruined by Twitter because some people don't know when to stay quiet.

Austin is on LinkedIn and Threads but not Twitter
He's seen multiple careers ruined by Twitter
The core issue: people who don't know how to shut their mouth

“I think Twitter like ruins people's careers I've already seen multiple careers run by Twitter some people just don't know how to shut their mouth”
Austin Hay · 1:23:00

#social-media#twitter#careers#opinion

Explainer· 5

Explainer04:30

What Martech Actually Is (and How It's a PM Role)

Austin defines marketing technology as an amorphous, cross-functional discipline sitting between product, growth, engineering, and marketing. He frames the martech person as essentially a product manager whose product is the marketing systems and platforms themselves. Crucially, martech isn't only third-party tools; as companies scale it includes first-party homegrown systems built on top of them.

Martech lives at the crossroads of product, growth, engineering, and marketing
Think of a martech person as a PM whose product is the system/platform
It spans both third-party tools AND first-party homegrown solutions
The role's shape is a function of company size and stage

“marketing technology is like this very amorphous cross-functional discipline that lives at the crossroads of product and growth and engineering and marketing”
Austin Hay · 04:30

#martech#definitions#product-management#growth

Explainer25:30

Martech vs Marketing Ops: The Key Distinction

Austin distinguishes marketing technology from marketing operations. Martech has 'tech' in it, so it's usually an engineer or someone with an engineering background architecting and building systems. Marketing ops is semi-technical work like setting up campaigns, sending email blasts, debugging, and running SQL queries, and you typically see the mar-ops function in B2B.

Martech is an engineering-based role: architecting and building systems
Marketing ops is semi-technical: campaigns, email blasts, SQL, analytics
Mar-ops people may be systems/business analysts, not engineers
The mar-ops function shows up most often in B2B

“marketing technology has Tech in it so it's usually an engineer or somebody with an engineering background doing that function marketing operations is usually not…”
Austin Hay · 25:30

#martech#marketing-ops#roles#b2b

Explainer31:30

How Cheap Warehousing Killed the Classic CDP Stack

Austin walks through the b2c stack 'then and now.' From ~2016-2020 the CDP sat in the middle collecting user data and piping it to ad networks, email, and analytics tools. Around 2020 the cost of warehousing dropped sharply, making it cheap to store everything in something like Snowflake, which gave rise to reverse ETL and let companies effectively build their own CDP.

2016-2020: the CDP was the centralized hub of the b2c stack
Around 2020 the cost of ownership of warehousing became much cheaper
Cheap Snowflake-style storage enabled the rise of reverse ETL
Simplicity-first businesses buy a CDP; cutting-edge eng teams use reverse ETL

“the cost of ownership of warehousing became much cheaper”
Austin Hay · 32:30

“now you can actually build your own CDP and lots of businesses already have”
Austin Hay · 33:00

#cdp#reverse-etl#data-warehouse#stack

Explainer52:00

The End of the Deterministic Matching 'Golden Years'

Austin describes how ad attribution has fundamentally shifted. From 2010 to 2020 marketers enjoyed deterministic matching, tying an install to a device's IDFA and PII with precision. That era is over, so ad networks have become more complex while marketers understand their spend less. He now spends most of his time on probabilistic modeling that extrapolates from ~30% of a population to 100%.

2010-2020 were the 'golden years' of deterministic matching
You could tie an app install to an IDFA and PII with precision
That's no longer possible; opt-in rates make it very hard
Probabilistic matching/attribution is now a core skill marketers need

“from 2010 to 2020 we had the golden years of deterministic matching where you know it was very easy to run an ad and understand…”
Austin Hay · 52:00

“so so probabilistic matching and probabilistic attribution I feel like is a skill set that more marketing technologists and marketers should just like get familiar…”
Austin Hay · 53:00

#attribution#privacy#probabilistic#ad-networks

Explainer54:00

Why Your Paid Traffic Now Shows Up as Organic

Austin explains two forces making web attribution harder. Browsers are stripping URL parameters, so a growing share of paid clicks arrive with their tracking truncated and get counted as organic. Separately, cookie blockers keep users anonymous throughout their journey until they log in, so you lose the pre-conversion source data entirely.

Browsers truncate URL parameters, so paid clicks get miscounted as organic
Third-party cookie blockers keep users anonymous until they log in
Anonymous journeys mean you lose pre-conversion source information
Blocked signups 'might as well be organic'

“browsers now are stripping out those URLs we talked about so you're just seeing a bigger and bigger percentage of people being counted as organic…”
Austin Hay · 54:00

#attribution#cookies#browsers#tracking

Story· 1

Story22:00

The Intern Who Emailed a Million People

Austin illustrates why permissioning and access control matter with a real failure mode: give someone HubSpot access and a test email goes out to a million people, embarrassing the company on Twitter. He says this has happened to him. The lesson is to design automated systems that grant access based on role, tenure, and department rather than manually clicking approvals.

Loose tool access can send a fake test to a million real users
This kind of permissioning failure has happened to Austin firsthand
The fix is automating access decisions by role, experience, and tenure
Admin/permissioning work is unglamorous but high leverage at scale

“you give added access to somebody who wants HubSpot and they send a fake email test to like a million people and now you're like…”
Austin Hay · 22:00

#permissions#operations#martech#risk

Q&A· 1

Q&A1:00:30

The Interview Question: 'How Did You Prepare?'

Austin's favorite interview question is simply asking how the candidate prepared for the interview. It reveals how a person thinks, plans, and takes things seriously without asking them directly. If you have to prompt them to list what they did, they're not a systems thinker; a rich, unprompted answer signals a more interesting candidate.

Ask candidates how they prepared for the interview
It surfaces how they think, plan, and take things seriously
Needing to be prompted for details signals a non-systems-thinker
The more interesting the answer, the more interesting the candidate

“how did you prepare you're really asking like how does the person think how do they plan how do they take things seriously or not”
Austin Hay · 1:01:00

“if you have to prompt them to tell you all the things they did then like they're just not a systems thinker”
Austin Hay · 1:01:00

#hiring#interviews#questions#systems-thinking

Tool· 3

Tool36:00

Reverse ETL Tools: Census and Hightouch

Asked for concrete reverse ETL products, Austin explains reverse ETL is a capability (moving data from a warehouse to a tool) found both inside CDPs and as standalone products. Segment, mParticle, and Rudderstack have reverse ETL functions, while Census and Hightouch are the two standalone options. He discloses he's an investor in Hightouch and uses it at Ramp.

Reverse ETL = moving data from a warehouse out to a tool
Census and Hightouch are the two standalone reverse ETL products
Segment, mParticle, and Rudderstack also ship reverse ETL functions
Austin is an investor in Hightouch and uses it at Ramp

“census which was Back Bay 16z and high touch are the two Standalone reverse etls and like I said I'm an investor in high touch”
Austin Hay · 36:00

“you should pick tools because they help solve problems not because of like anything else”
Austin Hay · 36:00

#reverse-etl#tools#hightouch#census

Tool1:15:30

Cal.com as a Cheaper, Cleaner Calendly Alternative

Austin's recently discovered favorite product is cal.com. A longtime Calendly user, he found Calendly expensive, clunky at syncing multiple business and personal calendars, and dated in its interface. Cal.com gave him a more Notion-like experience with a command-K interface and integrations that just work.

Austin was a longtime Calendly user but found it expensive
Calendly struggled syncing multiple business/personal calendars
He wanted a more Notion-like command-K interface
Cal.com delivered and he strongly recommends it

“I've been a big calendly user for a long time but countley is pretty expensive”
Austin Hay · 1:15:30

“cal.com has not failed me it has been awesome”
Austin Hay · 1:16:00

#tools#scheduling#cal.com#productivity

Tool1:18:30

Austin's 'Golden Stack' of Martech Tools

At the end of the episode Austin delivers his promised 'golden stack.' For b2c: Amplitude as CDP and product analytics, Customer.io for email (upgrading to Braze later), Snowflake as the data warehouse, Hightouch for reverse ETL, and AppsFlyer (or Branch) for mobile attribution. For B2B it's roughly the same, but Branch for web-only attribution and everything connected into Salesforce.

b2c: Amplitude (CDP + analytics), Customer.io email, Snowflake warehouse
Hightouch for streaming data out; AppsFlyer or Branch for mobile attribution
B2B: similar stack, but Branch for web-only and data connected to Salesforce
Email path: start on Customer.io, move to Braze later

“if I was a b2c business I'd buy amplitude for my CDP I'd buy customer i o and maybe I'd upgrade to braze in the…”
Austin Hay · 1:18:30

#stack#tools#amplitude#snowflake

Takeaway· 1

Takeaway07:30

The 100-150 Person Tipping Point for a Martech Hire

Austin explains that early startups can run tools with a 'village' approach where everyone chips in, but this breaks down at scale. Around 100 to 150 people, someone has to own how data flows through tools, the schema, and the procurement/legal contract exposure. Pain (not just efficiency) is usually what finally forces the hire.

Below ~30-40 people, everyone chips in to manage the CDP and tools
Around 100-150 people is the critical mass where the village approach fails
Unmanaged SaaS contracts create near-infinite liability exposure
Pain drives the decision more than a desire for efficiency

“usually around I would I would call it like 100 to 150 people is the critical mass where you can't just have a village approach…”
Austin Hay · 07:30

#hiring#scaling#martech#org-design

Key ideas

Frameworks in this episode

Build AND Buy

CDP vs Reverse-ETL Stack Decision

Future-Proof MTA Data Collection

Problem-People-System (PPS)

Thinking Gray

Tool-Agnostic Hiring Test

When to Centralize a Martech Function

People & resources mentioned

People · 11

Resources · 47

Insights & moments

⚡Myth Buster· 1

Most Companies Chasing MMM Actually Just Need MTA

◆Hot Take· 2

Why the Martech Person Is the Real Quarterback

Austin's Hot Take: Twitter Ruins Careers

✶Explainer· 5

What Martech Actually Is (and How It's a PM Role)

Martech vs Marketing Ops: The Key Distinction

How Cheap Warehousing Killed the Classic CDP Stack

The End of the Deterministic Matching 'Golden Years'

Why Your Paid Traffic Now Shows Up as Organic

❝Story· 1

The Intern Who Emailed a Million People

?Q&A· 1

The Interview Question: 'How Did You Prepare?'

✦Tool· 3

Reverse ETL Tools: Census and Hightouch

Cal.com as a Cheaper, Cleaner Calendly Alternative

Austin's 'Golden Stack' of Martech Tools

▲Takeaway· 1

The 100-150 Person Tipping Point for a Martech Hire

Myth Buster· 1

Hot Take· 2

Explainer· 5

Story· 1

Q&A· 1

Tool· 3

Takeaway· 1