Business Ideas II: HelpGen

Idea: HelpGen – Auto-Generate Help for Apps that doesn’t Suck

Let’s face it – online help sucks. Online knowledge bases usually suck too. This is a problem I’ve faced both with my own companies, but also in using a wide array of web and mobile applications. Why does help always suck? Because it’s an afterthought, usually written by a harried human who has 5 other jobs and for whom this is not job 1. Taking the opposite approach, while the machines are getting smarter, they are not yet able to write context-sensitive help that could actually tell you how to use a product. Or could they…?

MVP (Minimum Viable Product):

Use machine learning combined with assisted workflows to help organizations create instant help and knowledge base materials for their products. HelpGen would first traverse the site or mobile app in question, making note of terminology used, available buttons, and inputs, and other interactions possible on each part of the application. HelpGen would then guide the user to write content for each term, control, and step on each screen of the application – using machine learning to suggest content whenever possible. By monitoring the application for updates, HelpGen would also flag changes requiring help content update. By removing the friction from keeping help up-to-date, HelpGen could dramatically lower the complexity and cost of maintaining quality context-sensitive help.

Market: 5M+ applications deployed globally

The number of web and mobile applications has exploded over time, so app stores alone (this doesn’t count millions of web applications) count roughly 5M applications globally. The top 10% of all applications might see enough use to require ongoing maintenance – but that is still likely around 1 million applications globally. SMB SAAS solutions in related spaces like support / chat / feedback capture provide both a likely business model and comparison on market size. Various reviews of applications in the help/support space seem to indicate that $50-100/month per customer is rapidly attainable, with broader feature sets required to expand beyond that. This implies a total addressable market size in the $5B range, which is quite reasonable as a starting point.

 

Scoring (0-10 scale, up to 2 points per question): 7 points

1. Is it Transformative? 1 point

The HelpGen concept would be truly transformative if AI or ML could be used to automate the entire process – but as with most use cases, that’s not possible just yet. This is more likely to be a guided process for some time, which is still far better than the status quo.

2. Revenue Market Size or Eyeballs: 1.5 points

As discussed above, comparisons with other leaders in the customer support space indicate a thriving and rapidly growing market. I consider a $10B addressable market to be the gold standard, and this idea isn’t far off that.

3. In a Growing Market? 1.5 points

This market is growing as rapidly as cloud and mobile apps themselves, since all require some form of help and support.

4. Difficulty, Barriers to Entry, and Competition: 1 point

Many existing players in the help/support space may choose to attack this idea if it begins to gain traction. This is also not a winner-take-all market, providing some room for a well-built product to grow without being sidelined immediately.

5. Riding Hype or a Trend? 2 points

AI and machine-learning are arguably THE trends of the moment, and while arguably overhyped, HelpGen could definitely ride this marketing wave.

Business Ideas I: Juggler, Never Let A Message Drop

Over the years I have kept a running spreadsheet of business ideas – my current business, HiddenLevers, was once a denizen of the same spreadsheet. But ideas have expiration dates [1], and my idea list has grown while my available time has shrunk. Over the next few months I will be sharing my ideas – I’d love to hear feedback and to inspire others to take the next step or gain inspiration. To provide structure, for each idea I’ll share my thoughts on what I thought the MVP might be, and a scoring of the idea using my own 10 point scale. Here goes!

Idea: Juggler – Never Let a Message Drop

Juggler would watch your firm’s emails, LinkedIn, and other messaging platforms to ensure that every inbound request is tracked and gets a response. The challenge today is that inbound business communication arrives across channels, and often comes in to many different personnel at your firm. Using AI, Juggler would determine which messages actually require response, and monitor these across all firm users, alerting managing when prospects and clients are awaiting response.

There are a ton of AI-based email solutions and also support email solutions from firms like Zendesk – but none of these seem to focus on this specific use case – firm-wide monitoring and taking a global look at all communications to a particular client.

MVP:

The MVP is simple – do the machine learning work to simply determine whether a particular email requires response. Emails asking questions clearly come to mind – but taking a true machine-learning approach, can we approach 99% accuracy here? This can then be married to a simple UI showing individuals (not messages) requiring attention – this sort of dashboard data could ideally then be integrated into Salesforce or other CRM platforms.

Scoring (0-10 scale): 6 points

1. Is it Transformative? 1 point

This is a fairly standard use of machine learning in 2018 – but the accuracy level required to make this viable is not. Also, many businesses still don’t take real advantage of CRM systems, and this idea automates some of the key value concepts from CRM for a small business (ie don’t let any leads slip through the cracks – I’m looking at you, contractors).

2. Revenue Market Size or Eyeballs: 1 point

This is a broad market – virtually every business could use this capability, so volume pricing of even a few dollars a month in a SaaS solution could scale quickly. Presuming that this capability is worth $5/user/month – the US market alone is greater than a billion per year.

3. In a Growing Market? 1 point

While email utilization is stable, multi-channel communication is growing – think chat, social media, VOIP (phone) – in theory the same approach could be applied to all of these.

4. Difficulty, Barriers to Entry, and Competition: 1 point

It may prove difficult to achieve the level of accuracy with machine-learning to inspire user confidence. If businesses suspect that even a few important messages might be slipping through, they will lose confidence and not use the product. Ideally the system ought to learn based on each user and firm’s data – posing a bit more complexity.

5. Riding Hype or a Trend? 2 points

AI and machine-learning are arguably THE trend of the moment, and while arguable overhyped – the Juggler idea definitely is riding this trend.

 

[1] James Watt’s steam engine was an excellent invention, and applying it to pumping water out of mines an excellent business idea – for the 1770s. The concept of hailing a car via smartphone was likewise a great idea – in 2009. It’s also possible to be too early – Yahoo Briefcase shutdown the same year DropBox was founded (although the latter was also a vastly superior implementation).

 

P.S. Investors out there, feel free to reach out if any ideas in this series are of interest – while my core business continues to grow rapidly, I’m open to discussions on how to seed fund and launch against many of these ideas.

The Great GOP Stimulus

The 2018 Trump stimulus exceeds the Obama-era stimulus package in size – will it pay off at the top of the economic cycle?

In 2010, when Barack Obama pushed for a stimulus package to help boost the American economy, it was decided by many in the GOP as wasteful spending. While there are more productive (infrastructure) and less productive (tax rebates) ways to stimulate the economy, any form of spending (or tax cut) is a form of economic stimulus – this is a point agreed by both economists and businessmen like Warren Buffet. In fact, any form of budget deficit is a form of stimulus, as the government borrows (or prints) money that it doesn’t have to spend it into the economy.

The past year has seen the GOP enact not one but two stimulus measures – first a budget which ended Obama-era budget caps and boosted spending by roughly $150B per year, and second the tax cut which reduces taxes by another $150B per year. Taken together these measures are adding roughly $300B per year in stimulus to the US economy, potentially adding 1.5% to GDP for each of the next few years. Adding this stimulus to a core GDP growth rate of 2-2.5% might thus make 4% possible in the near term, with the bill due much later. The total federal (non-central bank) stimulus under President Trump’s first will hit at least $1.2 Trillion, exceeding President Obama’s 2010 stimulus package by $350 Billion [1], but this time at the top of the economic cycle!

What does this tell us? A few key takeaways emerge:
  • While most economists agree that it’s better to do fiscal stimulus when the economy is at or near recession, democracies don’t work this way, and there’s little correlation between economic need and actual governance.
  • When either party has complete control of government, they take the opportunity to spend on favored initiatives – in Trump’s case the DoD received most of the benefit, while in Obama’s case a variety of energy efficiency, infrastructure, and other initiatives were funded.
  • Budget deficits haven’t been a major issue over the last decade, but the tax cuts in particular will layer on top of Social Security and healthcare spending trends to drive debt-to-gdp well past 100% [2].
  • The best stabilizers in the US economy (unemployment insurance) are effectively automated – extending this sort of stabilizer to infrastructure spending (spending more on transportation funding etc as unemployment rises) would not just help buffer downturns – it would also get taxpayers a better deal.

Time will tell whether the GOP’s late-cycle spending will extend the business cycle substantially, but in the long run US policy will improve if more of these decisions are put on auto-pilot, removing the uncertainty of the political winds and the desire to spend at the least opportune times.

 

[1] The Obama administration stimulus plan cost around $850B in the end, including only the 2010 Stimulus measure and its implementation. Extension of Bush-era tax cuts and similar are not counted here, as these were extensions of existing measures, rather than new tax cuts or new spending as in the Trump administration’s recent moves.

[2] Many charts and news reports on the debt refer only to the publicly-held portion of the US debt, but when debts to the Social Security trust fund are included as in this data from the Federal Reserve, the US debt-to-gdp ratio already exceeds 100%.

The Hidden Factor Lowering the Homicide Rate

Improvements in gunshot trauma care have lowered the homicide rate substantially, even as the gunshot injury rate remained constant

A few years ago I read an article about the war in Afghanistan, and improvements made to battlefield treatment of gunshot wounds. Improvements in treatment at battlefield hospitals there led to a substantial decline in death rates relative to prior conflicts. Fast forward to the present, and many of these techniques have made their way to hospitals throughout the country, greatly improving gunshot trauma care. The rash of mass shootings in recent years has actually resulted in less deaths than would be expected – this New York Times article describes in detail improvements in gunshot trauma care. This led me to a larger question – have improvements in trauma care lowered the death rate from gunshot wounds? If so, this could be lowering the US homicide rate by rendering gunshots less effective in killing victims.

I then looked at CDC injury and fatality data – how much have gunshot injuries risen in the US in recent decades, and how has the rise in deaths compared? If there were no change in healthcare quality, one would expect the ratio of gunshot injuries to deaths to remain constant – barring any major societal change in how or why gunshot injuries were occurring. The primary change in the nature of gunshot deaths over the last few decades has been the rise in suicides, which would raise the ratio of deaths to injuries. Improving care for gunshot wounds would reduce the ratio of deaths to injuries, as more injured patients would survive their wounds. [1]

While gunshot injuries rose 85% between 2001 and 2016, fatal gunshot wounds rose only 31%! On a per-capita basis, gunshot injuries have risen much faster than population growth, while gun-related fatalities have remained constant. This finding has a few implications: [2]

  • Progress in healthcare makes it seem like violence in the US has dropped more than it actually has – Americans are being injured by firearms at a higher rate than in 2000, but they simply die less often. Either perpetrators’ marksmanship has declined rapidly, or we have some great surgical advances to thank. 
  • US healthcare has in fact made substantial progress in reducing the death rate associated with firearms, as the survival rate for severe abdominal gunshot wounds has risen from 15% to 77%.
  • While the decline in violent crime from its peak in the early 90’s is real, it has been augmented by improvements in healthcare. With suicide now accounting for two-thirds of firearm deaths, further gains through healthcare will be hard to attain.

 

[1] If the rate of suicide by firearm rose relative to homicide by firearm (as appears to be the case in recent decades), we would actually expect the number of deaths to rise relative to the number of injuries – because most suicide attempts by gun are successful. Despite this trend, we see injuries rising faster than deaths – meaning ER docs and trauma surgeons are making a real difference!

[2] The raw statistics can be looked up on the CDC’s (primitive) website for the years 2001 and 2016: Fatal firearm injuries were 38,658 in 2016 (11.96 per 100,000) and 29,573 in 2001 (10.38 per 100,000), for an increase of 31% over the period. Non-fatal firearm injuries were 116,414 in 2016 (36.03 per 100,000) and 63,012 in 2001 (22.11 per 100,000), for an increase of 85% over the period.

A Better Estate Tax Reform

Replacing the estate tax with fair (no step-up) capital gains taxation at death could raise revenue for tax reform, and get rid of complex tax avoidance schemes

Among the many changes proposed among the Trump and GOP tax plans is the end of the estate tax – long a cherished Republican goal. Today’s Republicans decry the estate tax as a form of double taxation, while proponents (including Republican President Teddy Roosevelt) view it as a means to prevent an aristocracy formed through inter-generational wealth transfer.

What’s overlooked in the estate tax debate is that there’s a simple solution at hand, if we just look north, to Canada. This may be surprising to many Americans, but in the early 1970s Canada repealed its estate tax, replacing it with a simple application of capital gains taxes.

Canada applies its capital gains tax to an estate by assuming that the assets have been sold on the date of the owner’s death. Instead of taxing an estate in a special way, a consistent application of the existing capital gains tax serves to eliminate loopholes (in particular by eliminating step-up basis) and raise revenue while also substantially lowering the top rate of tax on estates. If transfers of ownership are treated as taxable for capital gains purposes, this eliminates the use of trusts and step-up basis as a multi-generational tax avoidance scheme, since tax would be paid on any change of ownership, including when assets are transferred into the trust.

Instead of exempting substantially all estates (as with current law), a capital gains tax-based approach could simply apply current capital gains brackets. The top rate of 23.8% would represent a reduction of over 50% from current rates. This change could generate substantial revenue to enable other aspects of tax reform – in the year 2000, when the estate tax exemption was $1.3M for a couple, it generated $25B per year in revenue (after substantial exclusions, credits, and deductions). With the economy today 90% larger than in 2000, it’s likely that a similar tax would generate nearly $50B today. Elimination of step-up basis could double this figure by adding another $50B – and $100B per year would pay for a huge chunk of current Republican plans on business tax reform, without penalizing most individuals.

Unfortunately, Republicans are fixated on ending the “death” tax and ramming through their current plan, while Democrats are interested in keeping top estate tax rates in place – when a broader capital-gains based approach would be fairer and would generate more revenue. Hardly the last time a good moderate approach is left to die in our polarized political climate!

The Only Gun Control That Would Have Helped in Vegas

In the wake of major mass shootings like that in Las Vegas, gun control supporters have brought attention back to a wide range of gun control measures. Universal background checks, assault weapons bans, and numerous other measures are floated, and end up going nowhere given the heated opposition of gun rights supporters.

But while many of these measures would help with day-to-day gun violence (which, along with suicide, is responsible for 99% of all gun deaths) – none of them would have helped in Las Vegas. What might have helped?

A high capacity magazine ban, banning magazines with a capacity over 10 rounds, would have reduced casualties in Las Vegas by over half, and perhaps by as much as 90%.

After the Sandy Hook shooting, I analyzed every mass shooting in the US between 1980 and 2012 – and in each case I reviewed what forms of gun control, if any, might help. One finding stood out – the majority of mass shooters commit suicide, and they tend to commit suicide after running out of bullets and being forced to reload.

The Las Vegas shooter committed suicide as well. If mass shooters commit suicide as a result of the brief thoughts or impulses that occur during the reload cycle, then how many lives would have been saved by forcing the shooter to reload 10 times more often? In studying previous mass shootings, it turned out that a high capacity magazine ban could have saved over 50% of all lives lost.

Many Americans have absolutely no interest in gun control – Bill O’Reilly recently posted that “This is the price of freedom.” But for those with an interest in improving upon the status quo, there’s a lesson here: rather than wasting energy on a wide range of proposals, those interested in curtailing mass shootings should focus on EFFECTIVE measures, not feel-good measures. A high capacity magazine ban is unlikely to pass anytime soon – but placing the focus on a single effective policy instead of broad, sweeping measures helps set the stage for eventual success.

 

Cue the rebuttals:

There are millions of high capacity magazines in the US already! A ban would stop further sales, and since magazines are cheap a buyback could reduce the volume available while driving up the price of those remaining (making them harder to obtain)

Bad guys don’t follow the law! – Ending the public sale of new magazines will rapidly diminish their availability to the black market. Magazines are simple, but need to be machined precisely or they will tend to cause jams.

Shooters will just reload more often! – Analysis of over 60 mass shootings shows that the vast majority of shooters commit suicide after reloading a handful of times. Limiting the amount of damage possible per reload cycle thus reduces the overall damage.

You’re depriving individuals of their 2nd Amendment rights – No less a figure than Antonin Scalia held that the 2nd Amendment is subject to limits and regulations. Bump stocks aside, fully automatic weapons have been banned for individual ownership since 1934 (as a result of misuse of Tommy guns by Al Capone and other mobsters). The question has never been about unfettered access to arms – in this case suitcase nukes would be an arm that one could bear. The question has always been one of striking the appropriate balance between gun rights and the dangers they introduce in society.

100-Year Flood, 500-Year Flood: Real Risk Probabilities

When the Army Corp of Engineers and NFIP came up with the 100-Year Flood and 500-Year Flood designations, it’s almost as though they wanted to confuse the public. With Hurricane Harvey, much has been written on the meaning of the terms 100-year flood (it means a 1% chance of flooding in a single year), and the term 500-year flood (a 0.2% chance of flooding in a single year). While these basic definitions are correct, they don’t really help homeowners, whose question is: what’s the chance that my house will flood while I own it?

In the case of the 100-year flood zone, this means that the chance of flooding is at least 1% in a single year. But what if you plan to own your home for 30 years? In this case, you have a 99% of NOT flooding each year, but you’ve got to NOT flood for all 30 years. The probability of NOT flooding over 2 years is 0.99 * 0.99, and thus the probability of not flooding over 30 years is 0.99^30, or 74%. This means that the home has a 26% of flooding over the 30 years in question.

Of course, that doesn’t take into account the change in probabilities resulting from a combination of climate change and reckless development in most American cities. According to Kenneth Trenbeth, a scientist at the National Center for Atmospheric Research, “What used to be a 500-year event has become a 50- or 100-year event.” With this in mind, we can lay out the following table of homeowners’ flood risks:

Flood Probability Over 10 Years Flood Probability Over 30 Years
100-year flood zone  10% 26%
100-year flood zone, climate-change adjusted (to 10-year flood) 65% 96%
500-year flood zone 2% 6%
500-year flood zone, climate-change adjusted (to 50-year flood) 18% 45%

The flood risks over 30 years likely exceed many homeowners’ assumptions even before accounting for climate change – the climate-change adjusted risks make flooding within flood zones a virtual certainty! Homeowners in these areas are well advised to buy flood insurance, which is an incredible value as it is priced by the government below fair value. Homeowners not in, but simply near the 100-year floodplain, should realize that THEY are now likely in the true 100-year flood risk area, while their neighbors in the floodplain are likely at much greater risk.

Think that the probabilities can’t possibly have shifted that much? Consider that parts of Houston have had 3 500-year flood events since 2001 – Harvey, the 2016 Tax Day Flood, and Hurricane Allison. The chance of having one such “500-year” flood in 16 years is around 3.2%, but three such events? It’s less than 0.15% if these are really 500 year floods! [1] The reality is that 500-year floods are likely more than 10 times more common than the old statistics indicate, and homeowners should plan accordingly.

 

[1] The chance of 3 or more 500 year floods is equal to 100% minus the chance of 0, 1, or 2 such floods. The chance of 0 floods is 96.85%, and the chance of exactly 1 flood is roughly 3%, leaving less than 0.15% for the other potential outcomes (which drop off very rapidly because of the 0.2% chance happening multiple times).