Science-Based Business Campaigns
What makes something a science? Hypothesis testing, challenging theories to be correct, acceptance of new data and theories, numeric methods. And perhaps most important: rigor with a purpose.
Our projects tend to be a cross product of business transactions [customer acquisition, upsales, new products] and math modeling [data science, pattern matching, optimizations]. Our team is very numerate, and we have considerable envy of consumer marketing. Unlike B2B marketing, consumer marketing benefits from continuous rigorous analysis, field research and large sample sizes. B2B marketing has none of these.
Experts on our team began seriously mapping out B2B marketing responses in 1993 when direct mail (the paper kind) and fax responses were de rigueur. They began collecting not only response rates and lead conversion metrics, but also correlations with daily news, response timing trends, frequency effects, and anti-patterning.
The Lever10 team has a very large data set against which we can regression test new theories and challenges. We have developed numeric methods and tools to parse and organize these data for new uses. In 2018, our B2B database will exceed a trillion correlated responses – all with program related meta tags.
B2B campaign results tend towards weak reproducibility. A campaign with good results may have a poor yield when repeated. When we try to analyze these campaigns with statistical tools, we usually find that our sample sizes are far too small. (Ever run a Chi-square test on B2B campaign results? Then you know what we mean.)
Tools: A Workbench, not a Lab
Our team uses several commercial, open source and home grown tool sets. But our research is 100% applied – it is workbench oriented, not a laboratory.
In order to improve any process, we start with baselining prior results. Then we try to determine mean results to serve as a benchmark. This goes pear shaped almost immediately in B2B marketing, especially if there is a moving front of technology involved. (None of our customers make shoes – it is high tech all the way to the bottom.)
Well that makes for a mess now doesn’t it? How can we tell if we are improving, or even keeping up, if we have an uncertain mean yield? No problem: most B2B marketing teams turn over the CMO long before there is any long-term trend analysis anyway. So almost every CMO in the B2B space moves on before they know how effective their programs really are. (We apologize to the few large companies, such as Oracle, who seem to run numeric-based programs quite well. But they are -not- telling their secrets to you, now are they? And their scale of operations comes close to B2C populations anyway.)
What does the Lever10 team do to overcome the UM (Uncertainty of Mean) problem? We use a proprietary implementation of Bayesian methods that we apply to both prior campaigns and progressive A|B testing. Our system has been refined for more than a decade to help us deliver reproducible results and program improvements. Are we the only people using Bayes rule to adjust the mean? Heavens No. It is an obvious approach, but we do it very well. And we believe that Lever10 has a unique ability to apply our methods to very small programs: Yes even startup companies. Our current sandbox project is using our models to improve Social Media program results.
Related to UM is a need for non-destructive Data Normalization. For example, we need to find the relative density in a CRM of Vice Presidents versus Directors. But both titles are often spelled in non uniform ways (including simple typographical errors). This trivial example can be extended to normalization across Industry Sectors, Revenue, Geo-Location, and many more filters in addition to Job Title.
You run a campaign against a random sample in week one. You get results. Then you run the same campaign two weeks later. You get very different results. Why? You don’t know, and that is a problem. Everyone has an opinion, but no one really knows. Is it the way you ran the two (apparently identical) programs? Is it some external influence? Are program results really that variant?
Lever10 uses polynomial matrices (a table of curves) to remove the differences caused by time of month, time of year, and even the effects of external news. (Ever send out a big campaign right when some major news broke?) Our exact system is a trade secret of course, but can openly tell clients that we use our large data set of results and adjust for CP with Eigenvalue normalization. Again, using Eigenvalues is not unique to Lever10 – economists do this every day to adjust labor statistics seasonally. But we have a powerful toolset to map small scale B2B campaigns and determine CP versus other factors.
Branding is all about recognition and patterning. Direct response is the opposite. Humans have a genius for detecting ‘marketing’ versus actionable inputs. B2B marketing is often very confused about the effect of entropy in direct response. Lever10 uses a set of metrics to correct the various stages of target interaction. (Please forgive a baseball analogy – baseball is a mathematician’s sport). The most important pitch in baseball is arguably the fastball; the most important pitch in marketing is the change up.
Solutions: Programs That Work
Enough about the problems and the fancy equations. Our clients need us to help them execute effective programs, not measure the universe. Our engagements are usually between 4 months and 24 months. Our process begins with a few simple programs wherein we discover (and invent) what will work. Then we take the working programs to scale.
Once we have a working model for our clients, we normally try to ‘teach them to fish’ on their own. We can come back for tune-ups, but unlike advertising firms, we do not camp out forever.
. . .
Notice the omission?
There is no call-to-action button here. No invitation to ‘Contact Us Now.’
Why? Well first, we are pretty busy. This is not an anti-sale; it is a fact. Our dance card stays pretty full. And we are specialists. Not everyone needs us. But we do accept new clients. The door is open. You can ask for a no cost evaluation.