Archive for the english category

Joni

Managing business development of an ad platform

english

Here’s a great example of a business development program of an ad platform:

Google provides similar service through its AdWords Partner program. Facebook and Google are offering the free 1-on-1 help for one simple reason:

It improves the quality of ads.

Because of this, two positive effects take place:

a) the users are happier. As two-sided markets, FB and Google need to constantly monitor and improve the experience for both sides, users and advertisers. Particularly, they need to curb the potential negative indirect network effect resulting from bad ads.

b) the results are better. Most of FB’s +2M advertisers are small businesses and lack expertise – with expert guidance, they will use the funtionalities of the ad platform better and will see better results. This prompts an increased investment in the ads, which increases the platform’s revenues.

Thus, this program is an example of a win-win-win business development program of a platform. The users are shown better ads, the advertiser gets better results and the platform increases its revenue. Given that FB and Google conduct some “lead scoring” to choose the advertisers with the most growth potential, the ROI of these efforts is almost certainly positive.

Conclusion

With these programs, FB and Google are once again beating the traditional media industry that has very weak support in managing online advertising. Basically, no interest in the client after getting the money. To do better in competition, traditional publishers need to help their clients optimize and increase the quality of their ads, as well as improve their core technology to close the gap between them and FB and Google.

Joni

How to reach B2B audiences online? That’s the question.

english

Introduction

It *is* indeed the question – most often focus of B2B digital marketing is on lead generation and marketing automation. Much less has been written about targeting.

Yet, B2B targeting is far from having been solved. The typical conversation between decision makers and agencies goes something like this.

Client: “All this online advertising is great – but I don’t want to reach regular people on Facebook. How can I reach managers who are interested in buying my products?”

So far the agency’s answer has been something like: “Well Sir, even the managers are using Facebook!”

Which, although true, remains as a quite shallow answer to a serious concern. Especially because the chance of finding the managers is in proportion to their prevalence in the general Facebook audience – that is, very tiny. Unless we solve the problem of targeting.

Solving the B2B targeting problem

Now, I’ve been interested in this question for a while. So far, I can see four tactics for reaching B2B audiences with digital marketing:

  1. Keyword targeting
  2. Manual display placements
  3. Advanced targeting criteria
  4. Audience looping process

Let’s explore each of these.

First, obviously we can opt for intent-based search marketing. When there is search volume, search advertising tends to outperform other digital marketing channels in terms of cost-efficiency. That’s because people are actively seeking for the products and services that show in the ads. However, the problem is that the industrial search volumes tend to be small in many verticals, leaving the impact on revenue very minimal. As such, the B2B marketer needs to look elsewhere to increase the number of leads.

Second, to get extra traffic we can go and buy media from known business press (e.g., Forbes, Wall Street Journal; you get the picture). The downsides of this approach are at least three:

a) CPM prices are often high in these venues, resulting typically in poor ROI – this is partly because the way ad space is priced relies on archaic methods, such as selling impressions, instead of modern approaches like click auctions

b) display advertising generally suffers from multiple problems, including banner blindness, ad clutter, and ad blockers. If you haven’t noticed, people don’t really like banner ads, or at least they seem to like them a lot less than text ads and Facebook Ads which are often better targeted and less intrusive

c) uncertainty – how can we know that the people of our industry are reading this publication? There may be other publications that they read, but those might not always be available to display advertisers.

Third, we can use various targeting parameters available in online platforms. For example, in Facebook and GDN you can find people by income level (e.g., top 10% of income earners) and net worth. However, here we assume that more rich people have better jobs, which is true in statistical sense but does not narrow the audience down enough to reach the B2B decision makers in our industry. Thus, it is better if we can get into direct targeting criteria, such as job position and company. And, we can.

For example, here is Facebook targeting for people who for Kone, the Finnish elevator company.

Figure 1 Targeting company employees on Facebook

In a similar vein, we can target a) industries and b) job titles on Facebook. Below are examples of both.

A. The current industries on Facebook Ads

  • Work > Industries > Management
  • Work > Industries > Administrative
  • Work > Industries > Sales
  • Work > Industries > Production
  • Work > Industries > Personal Care
  • Work > Industries > Education and Library
  • Work > Industries > Arts, Entertainment, Sports and Media
  • Work > Industries > Healthcare and Medical
  • Work > Industries > Transportation and Moving

B. Job Titles with search query ‘manager’:

  • Manager
  • Manager Employers
  • Talent manager
  • Manager (baseball)
  • Manager (association football)
  • Hotel manager
  • Sales Manager
  • Business Manager
  • Branch Manager
  • Relationship Manager
  • Marketing Manager
  • Finance Manager
  • Store Manager

For example, we could target people who work for ‘Kone’ and are ‘managers’. While this might work for corporations, targeting smaller companies is not possible because they are missing from Facebook’s database. We could also try and find the physical addresses of the corporations, and use geo-targeting to reach them – although without IP targeting (not available e.g. in Google and FB but can be purchased with big money elsewhere) it’s a shotgun approach, unless the company is located in the middle of nowhere. Overall, the problems of using targeting criteria include validation, sparsity and availability. I discuss these in the following.

a) validation – the accuracy of this data is not guaranteed, and we have no way of knowing that the ad platforms correctly classifies people. For example, I’ve seen people who most certainly don’t work for Facebook write Facebook as their employer.

b) sparsity – secondly, not that many people declare their workplace in Facebook, so the data is sparse and we don’t end up making a great number of qualified matches.

c) availability – all data is not available in all locations – e.g., Finland is missing the income level targeting.

Therein comes LinkedIn. As you can see from the following figure, LinkedIn provides several different targeting options that make it the most potential B2B marketing platform in the world.

Figure 2 LinkedIn targeting options

For example, we are likely to reach the proper Kone employees in different sub-companies, as visible from the following figure.

Figure 3 Company targeting on LinkedIn

While people may be careless about providing the correct job information on Facebook, on LinkedIn it is very rare that people would fake their job positions. Such behavior is easily captured and reported.

The downsides of LinkedIn are that a) the ad platform is less developed than those of Google and Facebook. If you’d like to rank them, ‘Google > Facebook > LinkedIn’ is the order from best to worst in terms of functionalities, although LinkedIn is rapidly catching up. Addtionally, b) the prices are considerably higher on LinkedIn than on Facebook or Google.

In the attempt to combine some of the good parts of each technique, I’ve created a simple process whereby we obtain leads through effective LinkedIn advertising and build segmented audiences for retargeting. This process is illustrated in the following figure.

Figure 4 Audience looping process

In particular, the idea is to create special landing pages that have a unique pixel configuration that corresponds to that audience — for example, you can “Managers from Mid-West US” who are targeted to a specific landing page that has unique Google and Facebook pixels with the proper description installed.

Managers from Mid-West -> Landing page A <–> Pixel A

Managers from upstate NY -> Landing page B <–> Pixel B

This way, you can segment your B2B audience further, and in retargeting through GDN, Facebook and potential other networks such as AdRoll, address them with highly tailored communication. In a similar vein, you want to cross-target to visitors you reach from those channels also on LinkedIn – see the figure.

Figure 5 Cross-retargeting

If you already have an email database (CRM, newsletter lists), you should obviously use that to build custom audiences, and then use lookalike audiences to maximize reach (called ‘Audience expansion’ on LinkedIn).

Since all the platforms provide the same metrics (CPC, clicks, conversions), the allocation of your budget can be elastically applied to where it provides the best return. This is aligned with optimization best practices. To identify the proper companies and job roles, you can do investigative lead research work by using tools such as Leadfeeder, LinkedIn Sales Navigator, Ghostery, and Vainu.io.

Conclusion

There are many targeting options for B2B digital marketing. For example,

  • income
  • net worth
  • IP address
  • industry
  • company
  • job title

Each has some strength and weaknesses. Ideally, the B2B digital marketing process captures the best parts of each platforms. Because we use LinkedIn, we can be sure that the seed audience is of high quality and accuracy. Then, we use the other platform’s superior reach and cheaper prices to re-address this audience with what I call ‘continued information’ (=not the same we told them already, but something more). We can also use GDN to narrow down the placement, thereby including only specific venues in retargeting.

Targeting or discovery?

Finally, I wanted to discuss an important matter. That is, some proponents of digital marketing suggest to foresake the notion of targeting altogether and focus on ‘inbound marketing’. The theory goes so that being present in social media venues where the industry folks participate, e.g. by answering their questions, one can build a reputation of opinion leader and therefore gain organic leads. Moreover, the inbound tactics entail the publication of free knowledge resources, such as ebooks, webinar and blog posts, all intended to attract organic traffic from social media and search engines to the company’s website. I’ve previously described inbound marketing as a paradigm that defines people as rational agents actively engaged in information retrieval activities, a view which contrasts seeing them as passive “targets” of advertising. Depending on which paradigm you subscribe to as a marketer, you might want to either maximize your targeting or your discoverability.

To many, it makes sense to view people as active information seekers. However, in the flipside I’ve observed inbound marketing is often uncertain and time-consuming process. In addition, companies engaged in struggle to measure their efforts effectively and actually deliver a credible ROI figure for inbound. Finally, the capabilities needed for that correspond to those needed for running a magazine publication, i.e. are not readily available in most organizations. The content game is fiercely competitive, and when you normalize the cost, e.g. CPM prices can be higher than for advertising. Or, the reach is very low, meaning that you really don’t get the impact you’re after. In my opinion, it makes no sense to foresake advertising – advertising should be the primary element in the marketing mix, and should you want to try out inbound marketing, its results should be transformed to conmensurable metrics that enable comparison between advertising and inbound marketing.

Please share if you have further ideas!

Further reading

Using Facebook Ads for B2B Targeting: http://www.practicalecommerce.com/Using-Facebook-Ads-for-B2B-Targeting

Google AdWords for B2B Organizations: 8 Questions Leadership Should Ask: https://komarketing.com/blog/google-adwords-for-b2b-organizations/

15 Audiences You Should Be Targeting with B2B Facebook Ads: https://komarketing.com/blog/15-audiences-you-should-be-targeting-with-b2b-facebook-ads/

Joni

How to achieve a good score from Google Pagespeed Insights

english

Speed is one of the most important ranking factors in technical search-engine optimization (SEO). Besides search rankings, pagespeed directly or indirectly influences Quality Score in AdWords (and thus to click prices), and, perhaps most importantly, to the usability of website. Better usability, more conversions.

PageSpeed Insights is a free service by Google to score the speed of a website. 100/100 is the maximum score.

WordPress, then again, is the world’s most popular content management system. According to some estimates, more than 50% of the Web is powered by WordPress. Also my blog is built on WP. I spent some time to optimize its pagespeed score, and managed to get it to pretty decent numbers (see the picture).

Fig. 1 Performance of jonisalminen.com (GTmetrix)

That is, starting from 80-something and ending up with 90-something. I’d like to have the perfecto score of 100/100, but I’ll settle for this for now since it already took many hours of work to get here. At some point when I have more time, I might continue the optimization process.

Anyway, here are some lessons I learned while making speed improvements:

  • Choose a lean theme – basically theme makes a huge difference. Re-test pagespeed with different themes, for example Schema or the one I use in my blog.
  • Use a dedicated WP hosting – I’m using GoDaddy, but e.g. WP Engine, Studiopress (Genesis Framework) and SiteGround are apparently better
  • Use Clouflare CDN (the free version) – other CDNs are probably good too, but this one is amazing
  • Load everything possible locally instead of calling externally (fonts, scripts, avatars…)
  • Use EWWW Image Optimizer (or similar, like Smush) to optimize image size; also, avoid scaling dimensions in the browser
  • Use GTMetrix (or similar) to check the waterfall of the pageload, and work back to reduce requests
  • Use WP User Avatar plugin to avoid retrieving profile pics from Gravatar
  • Use Autoptimize plugin to concatenate CSS and JS code
  • Place JavaScrip to footer to avoid render-blocking

…and, truth be told, *avoid* using plugins too many plugins. Here’s an example picture showing their impact on pageload. Plugins are easy when you can’t code or want quick fixes, but they introduce unnecessary overhead to page complexity.

Fig. 2 Impact of plugins on site performance (P3 Profiler)

In this example, plugin impact is 82.6% of page load time. In particular, I should stop using WP User Avatar, and instead change the code manually to achieve the same functionality.

Finally, here’s a list of great resources for optimizing pagespeed for WordPress sites:

Conclusion

I recommend every online marketer to spend some hands-on time with speed optimization. Gets you thinking differently about “nice layouts” and all the other marketing things. Since many websites are actually ignoring speed optimization, you can get a competitive advantage by spending time on speed optimization. And as we know, even a small edge in Google results can mean serious revenue gains because of the overwhelming search volumes.

Joni

Argument: Personas lose to ‘audience of one’

english

Introduction. In this post, I’m exploring the usefulness of personas in digital analytics. At Qatar Computing Research Institute (QCRI), we have developed a system for automatic persona generation (APG) – see the demo. Under the leadership of Professor Jim Jansen, we’re constantly working to position this system toward the intersection of customer profiles, personas, and analytics.

Three levels of data. Imagine three levels of data:

  •  customer profiles (individual)
  • personas (aggregated individual)
  • statistics (aggregated numbers: tables and charts)

Which one is the best? The answer: it depends.

Case of advertising. For advertising, usually the more individual the data, the better. The worst case is the mass advertising, where there is one message for everyone: it fails to capture the variation of preferences and tastes of the underlying audience, and is therefore inefficient and expensive. Group-based targeting, i.e. market segmentation (“women 25-34”) performs better because it is aligning the product features with the audience features. Here, the communalities of the target group allow marketers to create more tailored and effective messages, which results in less wasted ad impressions.

Case of design and development. In a similar vein, design is moving towards experimentation. You have certain conventions, first of all, that are adopted industry-wide in the long run (e.g., Amazon adopts a practice and small e-commerce sites follow suit). Many prefer being followers, and it works for the most part. But multivariate testing etc. can reveal optimal designs better than “imagining a user” or simply following conventions. Of course, personas, just like any other immersive technique, can be used as a source of inspiration and ideas. But they are just one technique, not the technique.

For example, in the case of mobile startups I would recommend experimentation over personas. A classic example is Instagram that found from data that filters were a killer feature. For such applications, it makes sense to define an experimental feature set, and adjust if based on behavioral feedback from the users.

Unfortunately, startup founders often ignore systematic testing because they have a pre-defined idea of the user (à la persona) and are not ready to get their ideas challenged. The more work is done to satisfy the imaginary user, the more harder it becomes to make out-of-the-box design choices. Yet, those kind of changes are required to improve not by small margins but by orders of magnitude. Eric Ries call this ‘sunk code fallacy’.

In my opinion, two symptoms predating such a condition can be seen when the features are not

  1. connected to analytics, so that tracking of contribution of each is possible (in isolation & to the whole)
  2. iteratively analyzed with goal metrics, so that there is an ‘action->response->action’ feedback loop.

In contrast, iterative (=repetitive) analysis of performance of each feature is a modern way to design mobile apps and websites. Avoiding the two symptoms is required for systematic optimization. Moreover, testing the features does not need to take place in parallel, but it can be one by one as sequential testing. This can in fact be preferable to avoid ‘feature creep’ (clutter) that hinders the user experience. However, for sequential testing it is preferable to create a testing roadmap with a clear schedule – otherwise, it is too easy to forget about testing.

Strategic use cases show promise. So, what is left for personas? In the end, I would say strategic decision making is very promising. Tactical and operational tasks are often better achieved by using either completely individual or completely aggregated data. But individual data is practically useless at strategic decision making. Aggregated data is useful, e.g. sales by region or customer segment, and it is hard to see anything replace that. However, personas are in between the two – they can provide more understanding on the needs and wants of the market, and act as anchor points for decision making.

Strategic decision aid is also a lucrative space; companies care less about the cost, because the decisions they make are of great importance. To correctly steer the ship, executives need need accurate information about customer preferences and have clear anchor points to align their strategic decision with (see the HubSpot case study).

In addition, aggregated analytics systems have one key weakness. They cannot describe the users very well. Numbers do not include information such as psychographics or needs, because they need to be interpreted from the data. Customer profiles are a different thing — in CRM systems, enrichment might be available but again the number of individual profiles is prohibitive for efficient decision making.

Conclusion. The more we are moving towards real-time optimization, the less useful a priori conceptualizations like target groups and personas become for marketing and design. However, they are likely to remain useful for strategic decision making and as “aggregated people analytics” that combine the coverage of numbers and the details of customer profiles. The question is: can we build personas that include the information of customer profiles, while retaining the efficiency of using large numbers? At QCRI, we’re working everyday toward that goal.

Joni

First experiences with startup user studies (ca. 2010)

english
First experiences with startup user studies (ca. 2010)

I was reading through old emails while backing up my Gmail inbox with Gmvault, an open source command-line tool. Among other interesting trips to memory lane, one message was about the first startup user study I was involved with. It was for a life-coaching startup that eventually failed. In retrospect, it’s interesting to reflect on what went wrong and how we could have improved. In that spirit, I’m sharing these notes on the user study.

***

Okay, here’s my analysis based on user feedback.

We have three big issues to tackle at the moment:

  1. value
  2. mobilizing users
  3. clarity and purpose of the service

Value refers to a) providing such benefits customers want to pay for and b) setting the right price point.

Mobilizing refers to practical issue of getting customers to use the service on daily basis – i.e. facilitating the data input process as much as possible.

Clarity refers to layout and functionality of the site.

1. VALUE

a) Problem:

Are users willing to pay for the service?

“Sivusto ei tuonut mitään uutta tässä vaiheessa muihin vastaavanlaisiin sivustoihin verrattuna, paitsi jos sivustonne pysyy ilmaisena.” (“The site doesn’t bring anything new compared to other similar sites, except if it stays free.”)

-> There’s a need to introduce features people are willing to pay for.

I’ll doubt they’d pay at the moment, although we should have asked that in the feedback form.

Solutions:

-> Creating a mobile version for paying customers (cf. Spotify)

“Ehdottomasti mobiiliversiota tarvitaan.” (“A mobile version is absolutely needed.”)

I agree that an optimized version for mobile phones is needed. Technically this would require mobile device detection and loading appropriate html+css files to fit the smaller resolution. Later on, if the service succeeds, an iPhone/Android app would be great 🙂

-> Other additional features for paying users such as meal (1) and training recommendations (2) (integrated in the app)

(1) “You could include an suggestion for the diet. You now have something like 50% proteins, 30% Carbo., 20% fat, but would be great to see already a diet suggestion based on that.

“It could include in the categories some traditional dishes like: Fish and potatos, Pasta and tomato, Kebak, etc…”

This was a constant concern by many respondents. As Valtteri suggested earlier, building meals is a great way of providing value. We could for example divide them to three categories based on nutritional values: snacks (välipalat), light meal (kevyt ateria) and heavy meal (raskas ateria). By combining these three, the service could propose a personalised diet based on consumption and need of calories calculated from person’s weight and other factors.

Also users could be given the possibility to add own meals and make them public for other users, as suggested in this comment:

“Yksittäisten ruokalajien sijaan voisi lisätä malliannoksia ja tarvittaessa muokata niitä samalla tavalla, miten annos pitää nyt koota täysin itse.”

(2) Another idea is creating a mentoring/peer supporting system that enables users to get support

b) Problem: “How much you’re thinking to charge the subscription?”

Solutions:

  1. Yearly fee of 25€
  2. Monthly fee of 2.95€ (“less than coffee cup in Starbucks” 🙂

Free of charge -> revenue coming from advertisements/affiliate marketing

So, I think we’ll need to make a decision now of whether to offer the service for free and acquire sponsors, or make it paid and introduce such features that would increase likelihood of paying.

Also, we can think of providing a free version for free and premium with extra features for a small fee.

2. MOBILIZING USERS

Problem: Users are too lazy/busy to add daily information, ergo the app is not used actively.

“In general I think that it is really great, but the problem is to add the information on a daily basis…. :(“

“Jos palvelua alkaa käyttää niin on riskinä, että sitä vain kokeilee, eikä käytä jatkuvasti”

Solutions:

-> Creating a mobile version that allows on-the-go editing of profile

-> Creating a scheduled email system that allows easy updating (“Did you do the assigned training? Answer ‘yes’ to this message and information is updated automatically to your Muscler profile”)

-> Making it possible to update via sms? (requires an sms gateway + might be complex to use)

-> Creating community pressure for using the service (assigning personal mentors who have access to a person’s progress data — while conserving anonymity)

-> Offering rewards after completing milestones (e.g. reductions of sponsor products such as proteins)

3. CLARITY AND PURPOSE

“On vähän epäselvä. Ohjeet voisi olla vaikka tyyliä ‘ranskalaiset viivat’ settiä” (“A bit unclear. Bulleted instructions needed”)

“Parantakaa/tiivistäkää putkea, joka alkaa tarvoitteiden asetannasta ja etenee seurannan aloittamiseen ja raportteihin, tällä hetkellä minun tulee itse välillä muistaa edetä seuraavaan kohtaan ilman kehhoitusta/vinkkiä” (“Improve the process starting from setting goals to tracking and reports – atm, i have to remember to move myself”)

“Tutoriaali tai step-by-step-ohjeet olisivat hyvät :)” (“tutorial or step-by-step instructions would be great”)

-> Inserting tooltips

-> Creating “Proceed to [next]” buttons (Proceed to Tracking/Reports)

-> Creating a “How to use?” page

Besides the video, a separate “How to use?” page is needed. a bullet-type list would do ok, like suggested by the user. it could contain steps of using the service along with links, being short and simple.

“Mitä tarkoittaa tarkalleen Vahvistus tai ‘Lisää kestävyyttä’. Näistä olisi hyvä olla tietoa, sillä eri lajien ihmiset saattavat nähdä nuo eri tavoin.”

(“What does strenghten or more endurance mean?”)

-> Maybe these could be removed and focus on mass increase?

4. Bugs & Language

“Suosittelisitko palvelua ystävillesi?” kohdassa sanasta “Kyllä” puuttuu toinen l-kirjain (“Kylä”). :)”

-> Kylä should be Kyllä on feedback page

“Tavoite laatikon perässä saisi olla mitä yksikköä siinä halutaan”

-> Add units after goal form field to clarify what is asked.

“Kun lisäsin ruokia omalle “tililleni”, ei missään tuntunut olevan hiilihydraatteja. Se valikko, mistä ruuat valitaan, voisi olla selkeämpi (esim. isompi ja ADD-nappula aina samassa kohtaa). Haku-toiminto ruoka-aineille olisi myös kätevä.”

-> search function for nutrition, bigger add button with fixed position

“Itse tykkään etsiä aina lisätietoja varsinkin tällaisista urheiluun liittyvistä aiheista, joista löytyy aina monia mielipiteitä. Siksi olisin kaivannut esim. linkkejä lisätietoon “The Harris Benedict Equation” -menetelmästä”

-> Insert a link to “The Harris Benedict Equation”

“-Report is showing for today: 364% ACHIEVED (!!??)”

This is an issue I spotted myself, too. There must be something bizarre in the calculation formula, or can you give me example of a proper diet of 100% per day for let’s say a person weighing 60kg?

“- Many times a orange button option is dimmed, and little confusing (do I need to enter more infp? Does it work?)”

-> The button should be yellow in normal state and changed in mouseover, not other way around.

In reports/diet, “viikottain” should be “viikoittain”

Traning advice -> Training advice

Also, server location still shows Ukraine. There’s still some timelag which can be annoying.

There’s a logic problem -> when i insert a weightlifting training of 5 x 35kg in goals and then go into tracking, i shouldn’t reinsert the same stuff. i should be able to select the activity and then check that it’s done. if the weight was different, it should be changed in goals and not here.

***

Joni

Thoughts on Remora’s Curse

english

Remora’s curse takes place when startup attaches itself to a large platform in the attempt to solve the chicken-and-egg problem of getting users. The large platform then exercises its greater power to void the investments made by the startup into the platform, essentially causing more or less deadly delays and needs for re-design. The idea originates from Don Dodge who wrote about the Remora Business Model.

Examples:

  • Facebook stopping “friends of friends” access
  • Twitter killing ecosystem players (cf. Meerkat)
  • LinkedIn killing Developer program
  • Google’s Panda update dropping sites

The popular platform is not your friend. If their interests collide with yours, they will walk over you. Period.

Solutions:

  • diversify – don’t be dependent on only one platform
  • limit the overall dependence on platforms; i.e. do not make integration your secret sauce (aka “never build your house on rented land”)
  • capture the users (envelopment): when you get them to visit for the first time, make them yours; e.g. email subscription, registration

Purposefully limit the role of platform to user acquisition as opposed to being core value prop. Platforms, seen this way, are just like other marketing channels – if they work, scale. if not, kill. The benefit of platform integration is that it may partially solve the cold-start problem: get faster traction and accelerate user growth.

Read more about Remora’s curse in my dissertation: Startup dilemmas – Strategic problems of early-stage platforms on the internet

Joni

Machine learning and Facebook Ads

english

Introduction

One important thing in machine learning is feature engineering (selection & extraction). This means choosing the right variables that improve the model’s performance, while discarding those reducing it. The more impact your variables have on the performance metric, the better. Because the real world is complex, you may start with dozens or even hundreds of variables (=features), but in the end, you only want to keep the ones that improve the model’s performance.

While there are algorithms, such as information gain, to help, expert judgment can be of help as well. That’s because experts may have prior information on the important inputs. Therefore, one could interview industry insiders prior to creating a machine-learning model. Basically, the expert opinion narrows down the feature space. While this approach has risks, primarily foregoing hidden or non-obvious features, as well as potential expert biases, it also has obvious advantages in terms of distinguishing signal from noise.

So, the premise of narrowing down search space is the motivation for this article. I got to think, and do some rapid research, on what features matter for performance of Facebook advertising. These could be used as a basis for machine learning model e.g. to predict performance of a given ad.

A. Text features

  • topic
  • sentiment [1]
  • includesPrice
  • includesBrandName
  • wordCount [1]
  • wordLength
  • charCount [1]
  • includesEmojis
  • meaningEmojis
  • includesQuestion
  • includesExclamation
  • includesImperative
  • includesBenefits
  • includesNumbers
  • isSimpleLanguage
  • includesShortURL

B. Images

  • includesText [2]
  • includesPrice
  • includesProduct [2]
  • includesLogo
  • imageObjects
  • includesPeople
  • includesFace
  • includesAnimals
  • imageLocation
  • isStockphoto
  • includesCTA
  • isDarkColorTheme [2]

C. Metrics

  • clicksAll
  • clicksWebsite
  • websitePurchases
  • countLikes

D. Demographics

  • gender
  • age
  • location

E. Misc features

  • adPlacement
  • campaignGoal

Application

A simple model could only account for C (=independent and dependent variables) and D (independent variables), while more complex models would run a more complex analysis of text and images using linear or non-linear optimization, such as neural networks (shallow or deep learning). Also, some of these features could be retrieved by using commercial or public APIs. For example,

  • Google Cloud Vision API – for image analysis [3]
  • MonkeyLearn – for text analysis [4]
  • EmojiNet API – for emoji analysis [5]

Limitations

Ideally, each advertiser has his own model, because they may not generalize well (e.g., different advertisers have different target groups). However, feature selection may benefit from learning from earlier experiences. Also, given that there is enough data, it may be possible that the model learns which features apply across different advertisers, achieving a greater degree of generalizability.

References

[1] https://adespresso.com/academy/blog/we-analyzed-37259-facebook-ads-and-heres-what-we-learned/

[2] https://venngage.com/blog/facebook-images/

[3] https://cloud.google.com/vision/

[4] http://monkeylearn.com/

[5] http://emojinet.knoesis.org/home.php

Joni

How to do political marketing on social media? A systematic process leveraging Facebook Ads

english
How to do political marketing on social media? A systematic process leveraging Facebook Ads

This post very briefly explains a process of using and scaling Facebook advertising for political marketing. It might not be clear for all readers, but professional online marketers should be able to follow.

The recipe for political marketing by using Facebook Ads:

  1. Create starting parameters (Age, Gender, Location, Message)
  2. Create total combinations based on the starting parameters
  3. Use prior information to narrow down search space: e.g., identify the 100 most important target groups (e.g., battle-ground states)
  4. Create Facebook Ads campaigns based on narrowed down search space
  5. Run the campaigns (the shortest time is one day, but I would recommend at least 2-3 days to accommodate Facebook’s algorithm)
  6. Analyze the results; combine data to higher level clusters (i.e., aggregate performance stats with matching groups from different campaigns)
  7. Scale up; allocate budget based on performance, iteratively optimize for non-engaged but important groups, and remove the already-converted voters.

The intuition:

You are using Facebook Ads to test how many different target groups respond to your message. You will cluster this data to identify the most engaged target groups. You will then try to maximize voter turnout within those groups (i.e., maximize conversion). In addition, you will create new messages for those groups which are not currently responding well but which you need to capture in order to win the election. You will keep testing these groups by creating new messages, one by one finding the most responsive groups for a given message.

Once a target group shows a high level of engagement, you will scale up your advertising efforts (think 10x or 100x increase). You will keep the test cycle short (a week is more than enough), and the scaling period long. Based on campaign events, you may want to revisit already secured groups to ensure their engagement remains high. Because you are not able to measure the ultimate conversion (=voting directly), you will use proxy metrics that reflect the engagement of different target groups (particularly, clicks, CTR, post-click behavior such as time-on-site, newsletter subscriptions). This enables you to predict likelihood to vote based on social media engagement. Once a person has “converted”, he or she is removed from targeting – this is done to avoid wasting your budget by preaching to the choir.

Here are some additional metrics you can consider, some of them are harder to infer than the basic ones: frequency of activity, sentiment level, interest in a single issue that cause votes, and historical voting records (district level). According to different metrics used, we can set a target level (e.g., time-on-site > 3 mins) or binary event (subscription to campaign newsletter) which represents conversion.

Overall, we try to mimic the best practices of online marketing optimization here by 1) testing with explore-exploit mentality (scaling appropriately), and 2) excluding those who converted from future targeting (in effect, they are moved into a different budget which is direct targeting by email – a form which is more personal and cheaper than ads). In addition, we delimit the search space by using our prior information on the electorate, again to avoid wasteful impressions and maximize ROI-efficiency.

Then, we fill the selected groups with data and observe the performance metrics. Finally, we cluster the results to get a higher-level understanding of each group, as well as find points of agreement between the groups that can be used to refine the communication strategy of the larger political campaign. Therefore, the data we obtain is not solely limited to Facebook Ads but can be used to further enhance messaging in other channels as well.

There. The methodology represents a systematic and effective way to leverage Facebook Ads for political social media marketing.

Also read:

Agile methods for predicting contest outcomes by social media analysis

Analyzing sentiment of topical dimensions in social media

Affinity analysis in political social media marketing – the missing link

Joni

The Role of Assumptions in Startup Pitching

english
The Role of Assumptions in Startup Pitching

More than truthfulness of the numbers, investors evaluate the assumptions underneath a pitch. They are not asking “Are these numbers real?” but “Could they be real?”.

The assumptions reveal the logic of thinking by the founders. When examining them in detail, one should get logical answers to questions like:

  1. How many sales people are needed to hit the sales goals?
  2. How much will it cost to achieve the sales target?
  3. How much is the cost for acquiring a new customer?
  4. How long is the average sales cycle?

(Assuming an enterprise sales case; the questions in a B2C market would be different, so you need to consider the circumstances.)

The investors are looking for “intellectual rigor” and “completeness of thought” from the founders. Therefore, the pitch needs to show that you understand how to run the business, and how those actions are linked with growth within a defined timeframe. Like one investor said, it is better to be roughly right than exactly wrong.

Joni

Startup due diligence: Some considerations

english
Startup due diligence: Some considerations

We had an interesting week with the Qatar Science and Technology Park (QSTP) that had invited several high-profile entrepreneurs from the US to evaluate the technologies of Qatar Computing Research Institute (QCRI).

Unfortunately, I wasn’t able to attend all the sessions, but from what I saw I picked up a few pointers for due diligence work done by investors when evaluating the startups. Here they are:

  • customer references => who are the existing customers and what do they say?
  • investor references => who are the existing investors and what do they say?
  • competitors => feature comparison & position map
  • technology => expert evaluation
  • IPR => defensibility of the core tech
  • key competitive advantage => if not the core tech, then what is the thing preventing others from replicating your success?

Conclusion

It’s worthwhile to mention that the formal due diligence process is something different from an informal one – the latter takes place when the investor does some initial inquiries about the team and the tech, and then decides whether he wants to pursue further discussions. After reaching an adequate level of confidence, formal and detailed due diligence procedures conducted with the help of experts (e.g., tech, legal, science) ensue.