AnalyticsIQ is seeking a highly qualified individual for the following position: Senior Analytical Consultant.
This is highly visible position reporting directly to AIQ's COO. The role requires providing technical support to AIQ's sales team and technical management of existing accounts.
This job requires 5+ years of hands-on analytics experience and 5+ years of sales support.
Salary and title commensurate with experience.
Please respond with resume to gailm@analytics-iq.com
Thursday, February 22, 2018
Wednesday, February 21, 2018
The Case for Automated Modeling
I have been a vocal opponent to all forms of (truly) automated modeling for years. When I started my first analytics firm (Sigma Analytics) in 1998, there were some seemingly valid players out there. Before we went out and hired what turned out to be 20 humans to build models, we did take a look at these automated modeling platforms. None could touch what a well trained human could do.
Just to be clear, I am not innately against automated modeling - it is just that we have tested most newcomers to this field and have universally found that an experienced modeler can outperform automated platforms. By outperform, I mean in predicting the future (as opposed to what automated platforms have traditionally done well - predict the past).
My view has recently changed - we have tested a platform that equals the power of an experienced analyst, and does it far more efficiently. Of course, the ideal solution is for an analyst to intervene into a 'near final' solution to add in the art of modeling.
What does this mean going forward? I think it will allow experienced analysts to cover far more ground (maybe 3X), while still producing really solid algorithms that actually predict the future. I am not under any illusions that this platform will remain a competitive advantage for my firm - but it does mean that we need to continue to adapt new technologies as they come along.
Just to be clear, I am not innately against automated modeling - it is just that we have tested most newcomers to this field and have universally found that an experienced modeler can outperform automated platforms. By outperform, I mean in predicting the future (as opposed to what automated platforms have traditionally done well - predict the past).
My view has recently changed - we have tested a platform that equals the power of an experienced analyst, and does it far more efficiently. Of course, the ideal solution is for an analyst to intervene into a 'near final' solution to add in the art of modeling.
What does this mean going forward? I think it will allow experienced analysts to cover far more ground (maybe 3X), while still producing really solid algorithms that actually predict the future. I am not under any illusions that this platform will remain a competitive advantage for my firm - but it does mean that we need to continue to adapt new technologies as they come along.
Tuesday, February 13, 2018
Predicting What Cannot Be Known
We here at AIQ have spent the prior ten years building marketing tools. We often focus our analytics skills towards taking things that you 'know' on a relatively small universe and projecting this information to the entire US adult population. An example would be household income: we start with a 'truth' file and use modeling techniques (both standard and proprietary) to extrapolate to every household in the US. Interestingly, this tool sometimes outperforms 'known' income files in accuracy (of course, the reason is that any widely available 'known' data is probably not really known).
Recently, my firm has turned its attention to items that cannot be known at scale. Examples would be whether someone has extrinsic or intrinsic motivations around philanthropy - so, basically the spectrum from 'why would I donate money to a charity if no one knows about it?' to 'I would never donate if people found out!'. Another example is predicting the spectrum of 'impulsivity'. On one extreme are people who are essentially impulse driven (and thus responsive to digital advertising) or, on the other hand, extremely conscientious. Individuals who are conscientious require a lot of information and multiple touches. We have found that digital advertising is not the best way to reach these individuals. The exciting thing about this is that we have jumped from predicting what people 'are' (meaning, typically, things that cannot be easily changed - such as gender, age, ethnicity) to what really motivates people.
Of course, the challenge in producing these tools is that it requires some deep psychological knowledge. About four years ago we began discussions with a leading cognitive psychologist, Holly Adams, about this concept. The ultimate result truly marries purchase/donation drivers with actionable marketing data.
We have repeatedly seen these scores validate in the real world.
Recently, my firm has turned its attention to items that cannot be known at scale. Examples would be whether someone has extrinsic or intrinsic motivations around philanthropy - so, basically the spectrum from 'why would I donate money to a charity if no one knows about it?' to 'I would never donate if people found out!'. Another example is predicting the spectrum of 'impulsivity'. On one extreme are people who are essentially impulse driven (and thus responsive to digital advertising) or, on the other hand, extremely conscientious. Individuals who are conscientious require a lot of information and multiple touches. We have found that digital advertising is not the best way to reach these individuals. The exciting thing about this is that we have jumped from predicting what people 'are' (meaning, typically, things that cannot be easily changed - such as gender, age, ethnicity) to what really motivates people.
Of course, the challenge in producing these tools is that it requires some deep psychological knowledge. About four years ago we began discussions with a leading cognitive psychologist, Holly Adams, about this concept. The ultimate result truly marries purchase/donation drivers with actionable marketing data.
We have repeatedly seen these scores validate in the real world.
Tuesday, December 19, 2017
The Truth About Marketing Data
I sold an analytics firm to Merkle in 2004. While spending two years at Merkle, I learned a lot about the state of the art when it comes to marketing data (and Merkle was and is one of the better data providers).
Here is what I learned:
Here is what I learned:
- There is very little that is known about individuals
- There is a delicate balance between privacy and accuracy
These days, there are two paths in the data biz: scraped data from the internet or compilation of public data.
While making a small attempt not to sound judgmental, I am not a fan of 'scraped' data related to marketing to consumers. I believe there is an ever larger backlash coming for using data that individual consumers consider to be 'private' - such as posting on social networks.
AIQ operates in another world altogether - the world of predicted data. To do this, we acquire everything you can know (via public data) about an individual or the geographic area where an individual resides. In total, AIQ now pulls in > 120 data sources. No single source is anything close to a 'silver bullet'. The only way to optimize this all of this disparate data is via heavy-duty analytics.
The analogy I frequently use is a chef making soup: no one would find any single ingredient in the soup very attractive as a meal, but when combined correctly something special can be created.
While making a small attempt not to sound judgmental, I am not a fan of 'scraped' data related to marketing to consumers. I believe there is an ever larger backlash coming for using data that individual consumers consider to be 'private' - such as posting on social networks.
AIQ operates in another world altogether - the world of predicted data. To do this, we acquire everything you can know (via public data) about an individual or the geographic area where an individual resides. In total, AIQ now pulls in > 120 data sources. No single source is anything close to a 'silver bullet'. The only way to optimize this all of this disparate data is via heavy-duty analytics.
The analogy I frequently use is a chef making soup: no one would find any single ingredient in the soup very attractive as a meal, but when combined correctly something special can be created.
Wednesday, November 29, 2017
Analytics and Digital Marketing
It is funny how history repeats itself. Digital marketing has followed (in many ways) the path of 'traditional' marketing by starting with simplistic judgmental criteria and steadily advancing towards more complex multivariate solutions.
When AIQ first began selling our data via DMPs in 2012, the state of the art was to use segments (sometimes) created from multiple data points. For example, the segment 'affluent traveller' might have been created by selecting the universe of individuals who had a net worth > $2MM and scoring high on our 'in the market for international travel' model. This reminds me of the early days of direct marketing (way back in the 20th century) when campaigns were often list driven. Basically, knowing that someone was a member of the NRA or a subscriber to Vanity Fair was all you needed to send a solicitation.
Just as things did evolve in the analog market, we have seen progression in the digital world. It is refreshing to see that measurement is now standard. We view this as wonderful - good data rises to the top.
We are also starting to see modeling as a more common element (albeit at mis-90's levels of sophistication). This is long overdue.
Where next? It would be awesome to see data-driven optimization become prevalent. For example, a segment of people who are 'targets' for a particular online campaign are not likely to be a truly homogeneous group. Designing data-driven content that optimizes response over and above targeting is the panacea.
When AIQ first began selling our data via DMPs in 2012, the state of the art was to use segments (sometimes) created from multiple data points. For example, the segment 'affluent traveller' might have been created by selecting the universe of individuals who had a net worth > $2MM and scoring high on our 'in the market for international travel' model. This reminds me of the early days of direct marketing (way back in the 20th century) when campaigns were often list driven. Basically, knowing that someone was a member of the NRA or a subscriber to Vanity Fair was all you needed to send a solicitation.
Just as things did evolve in the analog market, we have seen progression in the digital world. It is refreshing to see that measurement is now standard. We view this as wonderful - good data rises to the top.
We are also starting to see modeling as a more common element (albeit at mis-90's levels of sophistication). This is long overdue.
Where next? It would be awesome to see data-driven optimization become prevalent. For example, a segment of people who are 'targets' for a particular online campaign are not likely to be a truly homogeneous group. Designing data-driven content that optimizes response over and above targeting is the panacea.
Wednesday, November 15, 2017
Analytics Entreprenuerism
I have started three companies thus far in my life. Two were successful and one was a dismal failure. I am here to tell you that my latest success (AnalyticsIQ) would not have happened if I had not failed at my prior venture. The reality is that I had a lot to learn - and I am one of those extremely not-rare people who learns best by experiencing pain.
All three companies have been in the analytics space. The first one (Sigma Analytics) prospered due to luck in that we happened to provide analytics to the mortgage space and rode the wave up and got lucky again by selling the company right before the crash (which was very bad luck for the acquirer).
People occasionally ask me what is behind the success of AIQ. My answer is pretty simple: all business success is tied to two things:
All three companies have been in the analytics space. The first one (Sigma Analytics) prospered due to luck in that we happened to provide analytics to the mortgage space and rode the wave up and got lucky again by selling the company right before the crash (which was very bad luck for the acquirer).
People occasionally ask me what is behind the success of AIQ. My answer is pretty simple: all business success is tied to two things:
- Niche: Unless the advantage behind what you are offering is not clearly visible to your prospective market, do not bother.
- Luck: More precisely, the avoidance of bad luck. Rough economic times or unexpected competition are real possibilities.
- Learning from mistakes: If you are not the type to blame yourself first when bad things happen, I would absolutely never invest in your company.
In my case, after 20 years of running analytics firms I have began to run out of new mistakes to make. The highest praise I can give myself is that I (very) rarely make the same mistake twice.
So, while is is incredibly important to have a distinct competitive advantage in mind before jumping into any business, you will no doubt make mistakes. The question is: do you blame yourself enough?
Monday, November 6, 2017
Election 2020
I woke up in the middle of last night with a burning question: who am I going to vote for in the Atlanta mayoral election tomorrow? I grabbed my iPad and did some 2AM research. The first thing I noticed is that the election is 'non-partisan', meaning I do not have an automatic home base. Refreshing, in a way, not to have anything pre-decided and to make a decision without my normal bias towards one particular party.
It is interesting that the last two US presidential elections have given a lot of credit to the field of analytics for the outcome. The Obama campaign was very data-driven and - as an example - tailored messages for 'door to door' campaigning based on segmentation. The Trump campaign gave much credit to a previously unknown (to me) firm named Cambridge Analytica.
An analysis of the electoral college shows that 39 states are 'locks' for one party or the other. By my count, the Democrats currently have 197 safe electoral votes and the GOP has 195. The other 146 votes come from 12 battleground states which will decide the next election. When you dive into these states, the average margin of victory in 2016 was 4.1%. If FL and MI had gone to Clinton (where Trump had an average margin of 0.8%), she would now be president.
The reality of our currently landscape is that our elections are decided by a remarkably small group of people in these key states. When you look towards the 2020 election, why should the DEM candidate waste valuable resources marketing to the tens of millions or voters in the 'lock ' states? Conversely, why should President Trump takes his otherwise entertaining road show to states like Texas and Alabama? Outside of galvanizing the base or fundraising, they should not.
To actually win the election in 2020, the candidates can probably personally meet the truly undecided voters in these key states. Improving analytics (driven by voting history in each county and congressional district, along with predictors of education and ethnicity) will identify this small group of people. Barring a landslide (and it does appear that Trump's base is rock solid), we should expect a very close election where about 1% of voters in 2 or 3 key states decides the election.
It is interesting that the last two US presidential elections have given a lot of credit to the field of analytics for the outcome. The Obama campaign was very data-driven and - as an example - tailored messages for 'door to door' campaigning based on segmentation. The Trump campaign gave much credit to a previously unknown (to me) firm named Cambridge Analytica.
An analysis of the electoral college shows that 39 states are 'locks' for one party or the other. By my count, the Democrats currently have 197 safe electoral votes and the GOP has 195. The other 146 votes come from 12 battleground states which will decide the next election. When you dive into these states, the average margin of victory in 2016 was 4.1%. If FL and MI had gone to Clinton (where Trump had an average margin of 0.8%), she would now be president.
The reality of our currently landscape is that our elections are decided by a remarkably small group of people in these key states. When you look towards the 2020 election, why should the DEM candidate waste valuable resources marketing to the tens of millions or voters in the 'lock ' states? Conversely, why should President Trump takes his otherwise entertaining road show to states like Texas and Alabama? Outside of galvanizing the base or fundraising, they should not.
To actually win the election in 2020, the candidates can probably personally meet the truly undecided voters in these key states. Improving analytics (driven by voting history in each county and congressional district, along with predictors of education and ethnicity) will identify this small group of people. Barring a landslide (and it does appear that Trump's base is rock solid), we should expect a very close election where about 1% of voters in 2 or 3 key states decides the election.
Subscribe to:
Posts (Atom)