Recent talk I gave at the The Institute of Direct and Digital Marketing. How we use Marketing Automation to help sell our equipment to scientists.
At Owlstone we manufacture complex instrumentation, with lots of parts, and lots of people involved at each step in the process. We have a file of information called a ‘manufacturing traveller’ that starts empty and at the end of the manufacturing line, is full of documentation and test results. At each step in the build process, different members of the team will complete some tests, add the results and pass the traveller on to the next team member. Each has their own responsibility and is trying to answer a specific question, but the integrated output answers a whole set of questions that no one person can answer individually. I think there is a parallel between the manufacture of physical goods and the manufacture of sales.
A sales person in the field, tends to have the clearest understanding of the customer problem and value of the solution. However, this doesn’t always make it downstream, with high fidelity, to those who generate marketing collateral. Equally the marketing team will be generating significant insight into the sales messages that work through testing of ad copy, landing pages, email copy , offers etc. A sales and marketing ‘traveller file’ that gets passed between the team, could help connect different parts of the cycle. For a new product or application the infield sales team will be trying to understand the customer pain and decide if there is a ‘rough fit’ with your offering. They are populating the traveller with information on the problem, the value of a solution, hot buttons, prospective customers, client interviews etc. This is the analogous to the exploration stage in the lean start-up methodology. When it comes to validation, other members of the team will start to get involved; the traveller gets passed on to marketing who start to put together some collateral and run test campaigns. The goal is to achieve product-market fit and as a by product the traveller is now populated with solid empirical data – the inklings of working campaigns and channels, persuasion assets that resonate. The collected traveller is to able to answer questions that no one person can answer on their own.
With internet based marketing you generally don’t want to ask for too much information on a webform. One common practice is to take the minimum amount of information on the form and use a ‘data appending’ service e.g. Jigsaw, InsideView or LinkedIn to build out the details for a new lead e.g address and company information.
A quick and dirty, but effective technique is to send an email to your lead database on a public holiday – a high percentage of people have their Outlook ‘Out of Office’ switched on, which means you get a return email, often with a lot of useful detailed information, e.g. job title, direct phone number that can be used to append the lead record in your CRM system.
There are many ways to come at a problem ranging from thorough analysis through to use of simple heuristics and rules of thumb. I always like it when people can get to an answer by looking at it slightly differently. I came across this one in Baeyer’s book, Information.
Samuel Morse, of Morse code fame, wanted to develop the most efficient way to code letters so they could be transmitted quickly. The principle of achieving this is pretty obvious; the most efficient code assigns short symbols to common letters, and long symbols to rare ones. He then had to answer the question what is common and what is rare? What is the order of the frequency with which letters appear in English? One way to gather such statistics is to select a text, and count the number of times each letter appears. This method works well for the three or four most common letters but it becomes less reliable for the more uncommon ones, such as Q, X, Z, unless the reference text is very long. Besides, who wants to count letters from a 1000 page book. Morse’s pragmatic solution was a lot quicker; he walked into a newspaper office and counted the number of letters in each compartment of the printers box. Presumably decades of experience had reduced its contents to an efficient compromise between supply and demand. Since he found more Es than any other letter, E is represented by a single dot, followed by T with a dash. X,Y and Z, on the other hand, whose compartments in the type box where relatively empty, drew four symbols each.
If you ever want to get people of a mathematical bent shouting at each other you should try to get them to agree the solution to the Month Hall problem.
“Suppose you’re on a game show, and you’re given the choice of three doors: Behind one door is a car; behind the others, goats. You pick a door, say No. 1, and the host, who knows what’s behind the doors, opens another door, say No. 3, which has a goat. He then says to you, “Do you want to pick door No. 2?” Is it to your advantage to switch your choice?”
There are lots of ways to approach the problem; I’ve just heard a new one from Hans Christian von Baeyer’s book, Information, which should convince any die hard “it makes no difference if you stick or switch” people.
“Imagine there are not three but, but a thousand curtains, and one car. Initially you pick, say, number 815 with a resigned shrug – realising that your chances of success are one in a thousand. The host (who knows precisely where the car is) now opens 998 empty cubicles. Not the one you have picked and not cubicle number 137. Now he asks politely: ‘Do you want to stick with your first guess, curtain number 815, or switch to curtain number 137?'”. What should you do? By changing the degree it makes it a lot more intuitive.
A while ago we noticed that our company has a surprising number of people with matching initials – whenever we were writing meeting minutes we would have to use an initial for the person’s middle name to distinguish them. Out of 18 people (as it was then) there we four pairs of matching initials e.g. there were two BB’s, two JS’s etc.
What are the chances of there being exactly 4 sets of matching initials in a population of 18 people?
This seemed to be quite unlikely, however when you look at the problem it is almost the same as the famous Birthday Problem – In a group of 23 (or more) randomly chosen people, there is more than 50% probability that some pair of them will both have been born on the same day. Our initial problem is similar except we have 26×26=676 possible combinations of initials instead of 365 days of the year. The same approach can be used to calculate the odds of there being a match in our company of 18 people. However I wanted to know what the chances of there being exactly 4 sets of matching initials and got stuck, at which point I sent an email around the company (a lot of engineers and scientists) and resorted to a brute force Monte Carlo model.
The approach is outlined below (John Somerville cracked the problem the same way). We generate a random number between 1 and 676, which defines the possible set of two initials, for each the 18 people. We then do a pair wise comparison to see if there is a match between people. In the example below there is a match between Person 10 and Person 3. We can then run a series of iterations and keep track of the number of times a single match, double match etc occurs.
After a run of 10,000 iterations we got the table below. There was about a one in five chance of a single match, but for four matches the probability was very low indeed about 0.01-0.03% (only ran the simulation a couple of times). Not very likely at all!
Another guy, Maccas, came up with an even better simulation that took account of the fact that not all initials are equally likely e.g. John Smith, JS, is more prevalent that the initials ZZ. Alas the file is too big to link to from here. Here is a link on Wikipedia to letter frequencies .
Closed Form Solution
Not happy with just getting the numerical output I waited for one of my more gifted colleagues to come up with a closed form solution. Dave did not disappoint and sent the following MATLAB expression
It is 1 in 52047
This can be written with prettier conventional symbols. The number seems higher than that suggested by the simulations.
If anyone else has a better approach, numerical or closed form, please feel free to suggest……