Archive for October, 2018

Customer Feedback

October 14, 2018

I’m old, ancient, pre-internet. The thing the internet does with server logs was something we did without server logs. It was a new practice at the time to track every interaction with a touchpoint in a database, and associate that interaction with a person. But, there were CEOs and others that would tell their marketer that they didn’t need a database. Geez. Then, the internet came along and much was forgotten. Everybody’s careers depended on getting on the internet and keeping up with the rapidly moving practices in the marketer’s silo.

Eventually, SEO got away from most of us and ended up in the hands of SEO experts. And, in the race, print marketing communications went out of style, and many printers went out of business. Paper was weight, weight was expensive, and cost justifications that server logs gave the SEO crowd were nonexistent for print. It wasn’t the internet that disrupted print, it was server logs. Fixing this disruption should have been easy, but the print industry wasn’t listening. We already had the technology. Oh, well.

These days, product managers talk about talking to the customer. Really? What customer? Are we talking prospects, users, managers of users, managers of functional units, or managers of business units? The technology adoption lifecycle defines the person we are selling to differently in different phases. Alternate monetizations drive another set of customer definitions. So who the hell are we supposed to talk to?

With SAAS-based applications, every click can be analyzed via SEO methods to generate a long tail of feature use. We can associate these tails with users and customers across all of our definitions. We could know what the heck as soon as we had a normal distribution for a particular click. Sorry, Agilist, but you don’t have enough data. Much would be seen in the changes to those long tails.

In the earlier, every touchpoint captured data, we can watch prospects mature as we and they crossed the normal distribution of our technology adoption lifecycle. We can watch their onboarding, their use, their learning, their development of their expertise, and their loyalty effects.

Today, I checked out a book that was mentioned on a blog that was mentioned in my tweeter stream. Amazon showed the customer reviews in an interesting manner that I’d Amazon Customer Reviews Breakdownnot seen before. It broke down the average score into the contributing levels of satisfaction. As it is, this is great for retention efforts, and social networking efforts. It would also be useful in our presale market communications efforts. Prospects are ready to buy only when they reach a 5-star rating across all the marketing communications they’ve touched across the entire buying team.

It would be great in our efforts to develop new features, aka use cases, user stories, and other such efforts. We could push this further by capturing negative reviews, which when tied to the application’s long tail and the individual customer would tell us what we needed to do to retain the customer across all definitions of the customer. If a customer that gave us rave reviews suddenly isn’t, it wouldn’t be sudden if we were paying attention, and it wouldn’t have to end with a retention effort. There is a long tail of customers, not just SKUs. In a software application, every feature is an SKU.

All of this would require an infrastructure that more widely defined what we captured in our server log and what analytic equivalence would look like in all these uses beyond SEM.

In the adoption lifecycle, we could break down the clicks from every pragmatism slice. That would tell us how soon a given pragmatism slice would be ready to buy, and that would inform the marketing communications effort and the feature development effort. We’d know what that pragmatism slice wanted and when. We’d know how well our marketing communications is working for that slice. It would greatly inform our tradeoffs.

One last thing, customers don’t know what we need them to know, so they can’t tell us about the future. Without good definitions of the generic customer, we could be talking to the wrong customer and addressing the wrong issues. We could be taking value to a particular “customer” that would never care about the delivery of that value.



We are never far from the exception

October 9, 2018

I’ve played around with a new compass this weekend. Out on Twitter, Antonio Gutierrez posted is Geometry Problem 1392, Two Cyclic Quadrilateral, Cyclic Octagon, Circle. The figure looked interesting enough, so I drew a few and found them impossible. Hint, draw the circles first. I drew the quadrilaterals first. A funny thing happened on the way to the barn.

Yes, there is a process for inscribing a quadrilateral. The process works. But, more often than not, it does not work.  But, just for the fun of it. Draw a quadrilateral and only the quadrilateral. Draw it on a blank sheet of paper.

The procedure is to find the midpoints of the segments. Then, draw lines from the opposite midpoints. The intersection of those two lines would be the center of the circle inscribing the quadrilateral. Try it!

Theoretically, of course.

Because you tried it and failed. Keep trying. Keep failing. Call it risky. Conclude that inscribing a quadrilateral is risky. Then, go forward with something simpler, something more achievable, something more amenable to process, something simpler to invest in.

But there is a cheat. Draw the circle and then draw the quadrilateral with the vertexes on the circle. Yes, you end up with a perfect inscribed quadrilateral. You are in Euclidean space.

In those earlier attempts, you were venturing into an unknown geometric space. But, as a manager who knows their numbers, a manager that cranks out numbers and decisions in L2, in the geometric space of the spreadsheet, you only travel in L2. A safe assumption is that L2 is Euclidean although I’ve seen the Euclidean metric being dropped into a spreadsheet. This was done to promote the Euclidean space from an assumption to an explicit assertion.

I’ll assume that I cannot inscribe a quadrilateral in L2.

I was doing this on paper. I’d need a vector graphics-based drawing application to show you. I’ll give it a shot in MS Paint. We need to see it. And, you probably don’t have a compass in that pocket protector you don’t wear outside the house.

First, I’ll draw the cheat, the circle, the root of all things Euclidean–Well, 00 Circlea lot of Euclidean things. Circles were ideal. Mathematicians love their ideals because they make the calculations simple.

Back in our high-school geometry class, we were told that there are 360 degrees around a circle. If you didn’t believe this, oh, well, you flunked, and you didn’t end up in b-school. Maybe you became an artist instead.

Then we were told that any inscribed triangle with a base passing through the center, 01 Circleaka on the diameter would have a right angle opposite that base. Starting with a circle, this is easy. Starting with an arbitrary right triangle, rather than a circle, means we have to know a lot more before we can inscribe that right triangle. It’s still simple, but it’s more complicated to start with the triangle. And, if it isn’t a right triangle, we can still inscribe it, but being on the diagonal is another matter.

So we have a situation that lets us talk about 360 degrees and another that lets us talk about 90 degrees.

Then, we ditched the circle with the triangle postulate, which told us 02 Trianglethat a triangle’s angles add up to 180 degrees. We, again, were forced to believe this or face grave consequences.

03 TriangleWe went so far as to ditch the right angle as well. But, get 04 Trianglethis, then, they forced trigonometry on us. Oh, we also got rid of that diagonal as well. So we end up with two more kinds of angles: acute and obtuse. Still, the process to draw a circle around a triangle worked, circumscribing a triangle. But, we then faced an explosion in the kinds of centers.

Did you catch that terminology change? When we start with a circle we inscribe the 05 Inscribe vs Circumscribetriangle. When we start with the triangle we circumscribe that triangle. Two different situations. Maybe AutoCAD makes that clear, I don’t know. It matters. It just doesn’t matter yet, so expect an Agilist to jump on the elementary, near one and object to the implementing the further one. Oh, well, we can refactor later.

That explosion of centers illustrates a concept that translates well to the base of a normal distribution. The centers: the centroid, circumcenter, and incenter show up in different orderings for different triangles. For equilateral triangles, the centers are one and the same point. When the normal distribution is indeed normal, all three statistical centers, the mean, median, and mode, show up at the same point.

Moving up to quadrilaterals, we add a point and a line and start to run through the geometries of four-sided things. We can circumscribe squares, rectangles, and all the “pure” forms, the ideals again. But, the defects have us skewered on the fork at the question of the circle first or later.

So we’ll walk through that argument next starting with an inscribed cyclical quadrilateral, we’ll draw the circle first, aka using the Euclidean cheat.

“If you’re given a convex quadrilateral, a circle can be circumscribed about it if and only if the quadrilateral is cyclic.” That quote from Stack Exchange in a Google search. Yes, a cyclic quadrilateral is where started us off with trying to circumscribe06 Inscribed Quadrilateral a quadrilateral. By assuming it was cyclic, we got it done, but we set ourselves up with the needle from the haystack, so we didn’t have to find the needle. Always bring your own needle to a haystack. That needle is Euclidean space. We assume that if and only if rule when we start with a circle. That center is the center of the circle. That center is not the center of the circle.

We’ll look at the intersection of the diagonals. But, before we get there, 07 Inscribed Quadrilateral Center NOT.pngnotice that none of the angles are right angles, and none of the lines are on a diagonal. They would have to include the center of the circle, that black dot. We’ve already departed Euclidean space.

The intersection of the diagonals is not the correct procedure for finding 08 Inscribed Quadrilateral Center NOTthe center of the circumscribing circle. The process I referred to earlier was one of finding the midpoints of the line and joining the midpoints of the opposites side. I did it in MS Paint. That all by itself introduces errors. But, by in large, the error is larger than the error of MS Paint’s bitmap resolution, aka the error of quantification. We got closer, but still wrong and obviously so.

The math of the thing assumes that the 09 Inscribed Quadrilateral Center NOTcircumscribed circle would be in Euclidean space, so opposing angles, those connected by diagonals in the earlier figure, would add up to 180 degrees. I don’t know the number of degrees, but we can still add them up visually.

The area in question is a matter of is it equal to the area marked c. If we stick with a Euclidean space, the two have to add up to 360 degrees.

Cyclic means that opposing angles add up to 180 degrees. Not being cyclic means we cannot circumscribe the quadrilateral in Euclidean space. Not adding up to 180 degrees means we are short a few degrees, aka we are in hyperbolic space, or we are over a few degrees, aka we are in spherical space. We’ve seen this before with our evolution to the normal distribution.

Keep in mind that we are just drawing. We didn’t measure. We didn’t pull out a protractor. We are not running around trying to add up degrees and seconds. We are not navigators trapped in Mercator geometry. Oh, projection!

As product managers, we project profits in L2, and that is why we never try to innovate discontinuously. We are projecting in L2, but the underlying space is not Euclidean, so those nice numbers don’t tell us the truth. The underlying space is almost never Euclidean.

But, the point of this post is to find out if space is hyperbolic or spherical. If the sum of the opposite angles on B where more, then we are hyperbolic in one dimension and spherical in the other. Then, we have to know where the transition happens. Then, we would like to know what the rates of transition would be.

We see the same thing in our data collection towards a normal. We are asking the same questions. There we can see our skew and kurtosis and our rates of data collection. I’m not p-hunting. I’m hunting for my normal. I’m hunting for my short tail so I can invest with some expectations about risk, about stability. Be warned there are plenty of short tails in sense of belief functions in fuzzy logic, in the swamped by surface sense. There is a geologic like structure to probability densities, but we hide those from ourselves with our dataset practice, a practice about preventing p-hunting. We are not p-hunting. We are looking for investments hiding in “bad” numbers, numbers that appear bad because we insist on L2, thus we insist on the 10th-grade Euclidean space of high school. Nowadays, even that space is not strictly Euclidean.

I’ve hinted at heterogeneous spaces. Trying to circumscribe a freely drawn quadrilateral reveals how space transitions to and from geometries generating homogeneous spaces.

“Because in reality there does not exist isolated homogeneous spaces, but a mixture of them, interconnected, and each having a …”

a quote in the google search results citing  An Introduction to the Smarandche Geometries, by L. Kuciuk and M. Antholy.

We do business in those spaces, spaces where the ideal, the generic, are fictions. Discontinuous innovation happens in hyperbolic space. Continuous innovation happens in Euclidean and spherical spaces with the spherical being the safest bet. And, that hyperbolic space being the riskiest. We no longer invest in discontinuous innovation because we believe it is risky. It appears that such investments would offer little return because in hyperbolic space the future looks smaller than it will be.

And, in the sense of running a business, “One geometry cannot be more valid than another; it can only be more convenient.”Henri Poincaré (1854 – 1912), Science and Hypothesis (1901).

10 Non-cyclic QuadrilateralSo at last, we will encircle an indigenous quadrilateral in the wild. This is probably not a good example since it is concave. The circles generate a conic, but the points of the quadrilateral that are on the circles hint at a more complex shape. The circles hint that the geometry changes when we move from one circle to another. The smallest circle give us three spherical geometries; the largest circle, only one. The smallest circle gives us no hyperbolic geometries; the largest circle, two hyperbolic geometries.

Given that most companies work in late mainstreet adoption phase selling commodities from within a spherical space, we rarely dip into the hyperbolic space, except when we undertake a sampling effort towards the normal. In that effort, we might jump the gun and infer before we are back in Euclidean or our more native spherical space. So much for that inference. It will be fleeting. Know where you are Euclidean. Is it a normal with three sigma, or a normal with six sigma? It isn’t a normal with twenty sigmas. Your choice. It is not Euclidean when it is not yet normal, aka when it is skewed and kurtotic.

It makes me wonder if the reproduction crisis is an artifact of inferring too early. A statistician out on Stack Exchange insisted that you can infer with kurtotic and skewed distributions. Not me. Is the inference fleeting, a throwdown, or a classroom exercise?

Anyway, back to those rates, those differentials, which 11 Becoming Euclideantakes us to two more diagrams. The first one shows us what happens as 12 Becoming Sphericalwe achieve the Euclidean, aka the cyclic; the next one, we achieve the spherical.

Enough already. Thanks. And, enjoy!