2. Kaveh
PhD in Econ
Janel
B.S. Computer
Science
Michael
Creative Dir. at SF
Business Times
Titus
Investigative
journalist
Michelle
B.S. Computer
Science
The Team
3. News organizations haven’t been doing so well.
Advertising & subscription revenue
for U.S. newspapers (in billions)
60
50
40
30
20
10
4. Advertising & subscription revenue
for U.S. newspapers (in billions)
We wanted to do something about this.
CS 206: Computational
Journalism
60
50
40
30
20
10
6. News organizations
(e.g., NYT, WaPo)
Software
development
Reader-side platform
User data
Writer-side platform
Help news
organizations
attract new readers
and better engage
existing ones
Make the news more
informative and
enjoyable for
readers
KEEP: Customer
support
GET: Free testing
Direct / Sales Team
Software development Customer acquisition
Customer support
Annual licensing fee
Tech companies
with large amounts
of demographic data
(e.g., facebook)
Customer support
Help readers
understand different
perspectives
Direct / Internet
GROW: Upsell to
higher-end license
Original Business Model Canvas
News organizations
Readers
7. News organizations
(e.g., McClatchy,
NYT, WaPo)
UX design, learning
archetype engagement
User data collection
Writer-side platform
KEEP: Customer
support
Direct / Sales Team
News organizations
Annual licensing
fee structure
News agencies
Integrating with existing
news CMS
Direct / Internet
GROW: Upsell to
higher-end license
Predict archetypes
(branches) who
should get a different
version of the story
Provide framework
and structure to make
archetyping easy for
every story
Identify archetypes by
leveraging/clustering
user data already
collected by news orgs
Advertising, PR, and
marketing agencies
Parsing data, collecting
engagement stats
Analytics platforms
Ad server platforms
Publisher onboarding
Suite:
Premium features
Versioning for
other industries
Advertisers, PR, and
Marketing agencies
143 Interviews Later... Current Business Model Canvas
Software development Customer acquisition
Customer support
14. Experiment:
Do readers enjoy personalized news texts? Nope.
“That detail doesn’t
belong in this
article.”
“All this one
has is more
numbers.”
“Meh. This one
doesn’t flow well.”
“That statistic
doesn’t fit in.”
18. Did we
personalize the
wrong way?
Is personalized
text the wrong
solution?
Let’s understand one
customer segment really
well and personalize the
right way this time.
19. Who would be our early evangelists?
People who are...
1. Hungry for news
1. Dissatisfied with mainstream news
1. Making the extra effort to fill appetite
20. Who would be our early evangelists?
People who are...
1. Hungry for news
1. Dissatisfied with mainstream news
1. Making the extra effort to fill appetite
31. We went back to our vision, but with a twist.
This was from our Week 1 Presentation.
32. Then we took our MVP to news agencies, who said...
33. Irene Chang
Product Manager,
Text & Multimedia
It’s not a super huge need.
Maybe it’ll be more useful to
publishers?
Then we took our MVP to news agencies, who said...
“Meh.”
34. A solution for news agencies is a small market.
$600 millionSubscriptions cost
$10k to $1-2 million
We estimated our annual revenue to be $15 million.
Revenue of Top 3 in U.S.
35. With nothing to lose, we asked news executives
for thoughts on all our previous MVP iterations.
VS
MVP 1:
Personalize news texts
MVP 3:
Augment wire text
VS
MVP 2:
Automate style & standards
36. MVP 1:
Personalize news texts
…and they jumped out of their seats
for our very first iteration.
37. …and they jumped out of their seats
for our very first iteration.
Wait, what?
MVP 1:
Personalize news texts
38. Localization is really valuable, but at
the community level.
Daniel Schaub
Corporate Director of
Audience Development
We want our journalists to write
with a specific person in mind.
But there’s no structure in place
to suggest
that right now.
Tim Grieve
VP of News
What got them so excited?
39. Journalists & editors felt the same way.
Andre Taylor
Desk Editor
“If a story gets clicks or blows up, we
want to know how to do that again.
When something spikes,
it’s like Christmas.”
“People care about things you may not
suspect they’d click on.
Most of the time, the editor or reporter
thinks they know what people like.”
Arlene Washington
Digital Editor
40. Key Insight:
There’s a gap between the way journalists write
and the way publishers wish they would write.
To journalists and publishers, our original MVP
looked like a bridge between this gap.
41. Back to our Origins,
But Wiser this time
116
interviews
42. A Hunch
The original MVP tweaked sentences.
What really moves people is stories.
We need to inform writers about readers earlier
in the story writing process.
43. We changed how we pictured our readers
Demographic data → Archetypes, Occupations, Interests
(Consider the reader as a whole)
We changed how we personalized to our readers
Statistics & Detail → Tone, Structure, Anecdotes
(Consider the narrative as a whole)
47. Experiment: Do readers segmented by archetype enjoy
personalized news narratives? YES! “It helped me imagine the
actual product and made it
sound cool.”
(Tech)
“I like that it sounds like it is written
in a viewpoint of not needing to
advertise but to just write the facts
about the researchers' work.”
(Biology)
“I liked that it gave some
background on the condition
and some facts/figures...this
could very well be an issue of
interest though.”
(Biology) “It provided me the
opportunity to learn
more.”
(Tech)
“I liked it more because it had a
quote from Apple.”
(Tech)
50. Our idea:
We help journalists find
product-market fit for their stories.
51. Our idea:
We help journalists find
product-market fit for their stories.
How?
A writing tool that informs the writer
about their target audience during the
writing process.
52.
53. Advertising & subscription revenue
for U.S. newspapers (in billions)
We wanted to do something about this.
60
50
40
30
20
10
54. Advertising & subscription revenue
for U.S. newspapers (in billions)
We will do something about this.
60
50
40
30
20
10
55. Three of us will be continuing!
Janel
B.S. Computer
Science
Michael
Creative Dir. at SF
Business Times
Michelle
B.S. Computer
Science
59. Now our customers are news organizations.
$28 billionannual revenue (ads + subscription)
Pains: Subscriber model in a digital age.
Gains: Segment readers to drive subs and better target ads.
If we can address even 3% of this pool, that’s a $1B market.
60. Lesson 1: Good startups run like efficient code
while (product_market_fit==0) {
# define experiment;
# define threshold for
success;
if (result < threshold) {
next_step =
next_step_fail;
}
else {
next_step = next_step_pass;
}
}
Week Experiment Result
1 The Value of
Personalized News
Fail (18%)
3 Is There Unmet Need
for Crypto News?
Fail (67%)
4 The Value of
Personalized Crypto
News
Fail (44%)
... ... …
8 The Value of
Personalized News,
Part II
PASS!
(80%)
KD
61. Lesson 2: Like journalists, startups need key sources
Michelle Park
Interview Request from Stanford University Team
To: Schaub, Daniel (McClatchy)
-----------------------------------------------------------------
Hi Mr. Schaub,
I’m an engineering student at Stanford
working on a class project. I was hoping to
talk with you about how news agencies
could better serve readers.
...
Titus Plattner
Interview Request from Tamedia News
To: Johansen, Ed (MU)
-----------------------------------------------------------------
Hi Mr. Johansen,
I’m a journalist at Tamedia writing a story
on offshore tax havens. I would like to
talk with you about your company based in
the Cayman Islands.
...
KD
63. Tone: Excitement about new product
Structure: Surface tech-related information to the top
Anecdote: Incorporate quotes from leading technologists
Experiment: Do readers segmented by archetype enjoy
personalized news narratives?
64. Lesson 3: Passion is necessary (but not sufficient)
Crypto News
(MVP 2)
Social News
Personalized News Texts
(MVP 1)
KD
67. Kaveh Danesh Jihyeon Janel Lee Michael Grant Titus Plattner Michelle Park
MG
Econ PhD,
former Duke
Trustee, writer for
Obama, RA for Raj
Chetty, NCAA
Division I soccer
coach
CS Major,
Section Leader,
KPCB Engineering
Fellow, RA at
Stanford AI Lab,
Theater gal who
likes corgis
Knight Fellow,
Creative director
at the SF Business
Times, Lead SF
Chronicle, digital
incubator training
ground
Knight Fellow,
Investigative
reporter, created a
tool to share
terabytes of data
in newsrooms,
Swiss
CS Major,
CS Course Asst,
created platform
to crowdsource
education content,
intern at Apple,
Google, and NASA
68. Small market:
The three big wire agencies:
$600M per year for text in the US
(Subscriptions based on feeds + total readers, news orgs pay
btw. a few 10k to a 1-2 M. per year)
Content Cube max. revenue: 15M per year
=> “This is a hobby” said Steve W.
69. Interesting market:
Revenue of US newspapers ($28B in 2017)
If we even can address 3% of this market by driving more
subscriptions or better targeting ads, we would have a
$1B market.
1. For news orgs, getting more online subscriptions is key.
2. Better targeting online ads would allow news orgs to
compete again with the internet giants.
70. TAM:
Revenue of US newspapers in 2017:
Ads: $18B
Subscriptions: $10B
Total: $28B
SAM:
Bigger US newspapers revenue in 2017: $14B
Target:
Address about one fifth of this market: >$2B
Content Cube
helps on both
markets
}
Editor's Notes
add a hero’s journey slide
Once upon a time...
And every day...
Until one day...
Because of that...
Because of that...
Because of that...
Until, finally...
And, ever since then…
Put learnings on a timeline
What parts should be well illustrated
Overlay the storytelling
The plummeting curve
Journalism class
Naive but hopeful
Carried out with us a shining MVP
The plummeting curve
Journalism class
Naive but hopeful
Carried out with us a shining MVP
We created a writing tool that allows journalists to personalize sentences to different readers based on their demographics.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
And so our journey began. The first thing we did was test our assumptions about our original idea. Would readers even enjoy personalized news texts?
So we ran an experiment. We took the breast cancer article shown in our tool and personalized it to two different demographics.
YAF and OWM. We asked them to read both the standard and personalized versions.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
Not only did they not like it, they really hated it. They did
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
This is what we asked ourselves. Saying yes to the question on the right broke our hearts, so we decided to take a guess at the question on the left. Maybe we didn’t know how people wanted their articles personalized.
This is what we asked ourselves. Saying yes to the question on the right broke our hearts, so we decided to take a guess at the question on the left. Maybe we didn’t know how people wanted their articles personalized.
This is what we asked ourselves. Saying yes to the question on the right broke our hearts, so we decided to take a guess at the question on the left. Maybe we didn’t know how people wanted their articles personalized.
People who have a high information diet but aren’t getting satisfied from mainstream news.
This is what we asked ourselves. Saying yes to the question on the right broke our hearts, so we decided to take a guess at the question on the left. Maybe we didn’t know how people wanted their articles personalized.
This time, we wanted to really understand our customers. We talked to many cryptocurrency investors and found a natural grouping amongst them
We had the coin founders and the early investors
Then we had the experienced investors, who had hopped over from the finance space.
And finally, the finance and crypto novices who started investing because of the hype.
They were having a really hard time finding credible news sources, and told us the learning curve for crypto was uncomfortably high.
We spent a lot of time listening to their pain points and started mocking up solutions for the biggest pains.
But in the midst of this, something felt wrong.
This unease came together during one of our weekly presentations.
Steve Blank asked us how much we would charge for our product, and we nervously said “ummm, $20 a month?”
He immediately asked us: “Do you even believe in your product?” “Do any of you care enough about this space to buy a coin?”
And he was completely right. We sat down afterwards and realized that none of us were passionate about crypto, and it was eating us alive.
After all, as our instructors say, “If we aren’t excited about our idea, how will we convince other people to love it?”
And thus ended our brilliant but short-lived dive into cryptocurrency.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
It takes a long time for publishers to take content from news agencies and customize it to their own style and standards. After talking to a lot of wire editors, we found a lot of this could be automated, and it would enable news agencies to offer more value to their clients.
Because of that, thought deeply about it
Because of that, made pivot
Because of that, thought deeply about it
Because of that, made pivot
Because of that, thought deeply about it
Because of that, made pivot
Because of that, thought deeply about it
Because of that, made pivot
We were confused because we thought readers hated it. But somehow, news organizations seemed to love it.
For example, executives at McClatchy were already thinking about personalizationBut it wasn’t tweaking statistics. It was about creating stories that resonate with specific people.Tax plan change, you would tell the entire narrative differently to blah
So we talked to journalists and editors too, and they also wanted to know more about who they should write too. But like Tim said, they were only taking a guess at
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
tried occupation because thought of biggest things that shape people as a whole
We took an article about a collaboration between Apple watch and Stanford medicine
& personalized it two ways.
And again 3 people liked it
This time, they were in the majority.
They mentioned they strongly preferred the personalized version because of tone and anecdotes, confirming our hunch.
We also learned not to make broad assumptions about our experiment results.
Because of that, thought deeply about it
Because of that, made pivot
Because of that, thought deeply about it
Because of that, made pivot
Because of that, thought deeply about it
Because of that, made pivot
The plummeting curve
Journalism class
Naive but hopeful
Carried out with us a shining MVP
The plummeting curve
Journalism class
Naive but hopeful
Carried out with us a shining MVP
Because of that, thought deeply about it
Because of that, made pivot
add a hero’s journey slide
This is what we asked ourselves. Saying yes to the question on the right broke our hearts, so we decided to take a guess at the question on the left. Maybe we didn’t know how people wanted their articles personalized.
One key lesson from the course has been the importance of a scientific mindset to finding product-market fit. It took us many tries to pass an experiment, which was difficult and frustrating. But we stuck to the script and continued working in an almost code-like way: define the experiment that tests the key assumption, define the threshold for success, define next steps conditional on success and failure. And eventually we got results that did not reject our hypotheses.
Another key lesson was the importance of finding good sources. In the beginning of the semester, we were mostly interviewing friends or friends of friends, but by the end we were sending cold emails or LinkedIn messages to people we never would have thought to contact. What’s funny is that this is exactly what journalists do. They don’t settle for interviews with people who don’t understand the subject of the story. They make bold calls and send bold emails in order to dig up the truth. And that is what we started learning to do over the course of the semester.
This is what we asked ourselves. Saying yes to the question on the right broke our hearts, so we decided to take a guess at the question on the left. Maybe we didn’t know how people wanted their articles personalized.
key learnings included x, y, and z
added a lot more yellow boxes
-completely revised value prop, customer segment and customer relationship
-detail
-clarity of what you’ve learned
-we started by not knowing much, now we know a lot
-put green around stuff you were right about, yellow around stuff that changed after 143 interviews.
Finally, we learned that while passion, ability to contribute, and economic viability are all important components for a successful startup, none of them is sufficient on its own. We started the semester with a tool that personalized news texts, and we were pretty confident that we execute that as well as anyone. But when we heard that people didn’t seem to want that, we thought about making news more social—something we were passionate about—before pivoting to where we thought the money was (and you all know how that went). We feel our current idea strikes a nice balance between all three elements, in a way that sets us up to take this curve -- *next slide* -- and try to flip it upside-down -- *next slide.*
Once upon a time...
Every day...
But, one day...
Because of that...
Because of that...
Because of that...
Until, finally...
And, ever since then...
Once upon a time...
Every day...
But, one day...
Because of that...
Because of that...
Because of that...
Until, finally...
And, ever since then...
http://www.journalism.org/fact-sheet/newspapers/
Ads in 2017: 18B + circulation 10B