How small businesses can drive value from small data

Small business often means small data, but there are ways to drive value from it.

Chat with MarTechBot

In the world of big data, gigabytes are worth less than pennies. But that loose change can add up to dollars for a small business. Those digital mom-and-pop online stores face the same analytical challenge as their gigantic competitors: figuring out who are the best customers and how to sell to them.

Small businesses simply may not have the quantity of data — or the means — to use big data techniques on their own. But they can always hire a solution that helps drive value from “small data”. Methods vary.

Creating big data from small data

Long known for selling email marketing solutions to small businesses, Mailchimp is also a player in big data. The company has its own corps of data scientists, computer scientists and mathematicians, explained David Dewey, Chief Data Science Officer at Mailchimp. “The objective is to bottle ourselves and put it into the product… You gain the capability without hiring the person.” Dewey said.

On the back end, Mailchimp has petabytes of data from its customers, anonymized to protect their privacy, but available for analysis. Mailchimp uses its machine learning model to act as an input to the client’s small business model. The big data input acts as a center of gravity, keeping the small business model on track, Dewey explained. Small businesses can compare their performance against the averages compiled by Mailchimp’s customer base to see how they rate. Those patterns can also reveal areas where improvements can be made, and best practices that can be used to enhance online sales.

Essentially, Mailchimp is seeking to give small businesses the benefits of big data-driven machine learning by pooling the small data they have available.

Using small data as “Business GPS”

Another approach is to tap into the data stream in real time, recomputing to redraw the business picture, explained Mark Stouse, chairman and CEO of Proof Analytics. “It’s not hard to use data as it’s being collected. It gives you observations, and improved confidence in the results.” he said.

For business leaders, a confidence level around 50, 60 or 70% is enough, Stouse said. “There is a bias in data science towards extreme precision. Ninety-five percent is the benchmark.” he said. The business user never gets to 95% confidence in any situation, nor do they need it, he added.”

Being able to recompute on the fly makes small data much like a GPS. “It allows you to navigate the problem and make changes,” no differently than changing the route on a road trip, Stouse said.

Some familiar small data plays

Working with small data should actually be quite familiar. Small data is always there, even if you are not using a platform or a service to analyze it. “It’s less about the algorithm,” said Scott Brinker, VP Platform Ecosystem at HubSpot. “It’s more about the content. How much of the content is good?”

Big data algorithms can sort through millions of process permutations to predict what might be the best message that produces the best response from a cohort of clients. “The black box is too much. You lose the plot,” Brinker said.

Having a clear idea who the customer is can offer better guidance — a small data model in the head of the marketer rather than a big data algorithm, he explained, Small data techniques are common, varied, and have been around for a while.

Net promoter score (NPS) simply gathers feedback from those little customer survey pop-ups asking the user to rate a service or product on a scale of 1-10. The 9s and 10s are promoters. Below six are detractors. Yet this kind of data can provide an insight with just hundreds or thousands of replies, Brinker noted. “It’s definitely not big data.”

A/B testing is another technique, a conversion optimization tactic now over a decade old, Brinker said. Make two different offers and see which if the two gets a better customer reaction. Again, this is something that only requires hundreds or thousands of participants to yield and insight. Content marketing and search engine optimization are small data plays. “Which content and which keywords are driving organic traffic?” Brinker said. Huge value is derived in a lot of marketing from the effective use of small data, he said.

How much data is enough?

Big data can choke itself, just from sheer size. “There is a fundamental difference between the cult of precision all over data science versus the practical reality of business,” Stouse said. “Data availability is the real problem,” but “small data” is much more readily available.

It is as if the business had parallel data universes — one large, precise universe of data science, and a more approximate but immediate universe of business needs and outcomes. Many large data pools or data warehouses are being taken down as a result, because they are expensive to maintain and secure, while “value extraction” has been less than desired, Stouse said. Analysis can look at an organization’s overall data picture, but actionable data collected in real-time can be more useful, he added.

Small data is vulnerable to outliers

While large data-sets will produce analytic results that are generally unlikely to be swayed by outliers, small sample sizes are vulnerable to outliers, potentially creating false conclusions that can have an outsized influence on outcomes. Mailchimp has a rigorous process to determine if an outlier has an unwanted result for the user, Dewey said. If the outlier has no negative impact, it will simply be noted and reported to the client.

Still, nature has a way of creating its own outliers. The Covid-19 pandemic hit the economy like a freight train out of left field, disrupting even projections based on very large data sets. “It was definitely a period of time when the model drifted in unavoidable ways,” Dewey said. “Pandemic” and “COVID-19” became new keywords in online searches. Yet this was a change in attention, which did not trigger any changes in human behavior, he told us.

Proof Analytics approach to monitoring data in real-time allowed users to see the problem developing and roll with it. “The past as prologue is not true right now,” Stouse pointed out. Still, companies have to project forward with the data they have, “or you are floating freely.” If programs don’t work in the current environment, they need to be adjusted to suit the changing reality.

Choosing the best approach to small data

In the end, Mailchimp provides “outsourced expertise” to the small online business. Such enterprises are run by a few people, or even one, and they are busy following their passion or maintaining their focus on a market niche. They often lack the time and the expertise to do their own analysis in order to identify best customers, estimate customer lifetime value, or even how and when to time e-mail reminders to prompt more sales.

It’s on Mailchimp’s AI to do the analysis, craft the program, and present it to the small business client, Dewey said. “They are acting as the executive,” he said, clicking to approve or disapprove the work.

Proof Analytics prefers to provide guidance that aids decision-making. With enough automation, one does not need a data scientist, just an analyst, Stouse said. Even then, a part-time analyst will be sufficient. Mitigating the risk in the decision is the goal, since “a bad decision [can be] costly,” Stouse noted.

Either way, the client can focus on running the business.


Opinions expressed in this article are those of the guest author and not necessarily MarTech. Staff authors are listed here.


About the author

William Terdoslavich
Contributor
William Terdoslavich is a freelance writer with a long background covering information technology. Prior to writing for MarTech, he also covered digital marketing for DMN. A seasoned generalist, William covered employment in the IT industry for Insights.Dice.com, big data for Information Week, and software-as-a-service for SaaSintheEnterprise.com. He also worked as a features editor for Mobile Computing and Communication, as well as feature section editor for CRN, where he had to deal with 20 to 30 different tech topics over the course of an editorial year. Ironically, it is the human factor that draws William into writing about technology. No matter how much people try to organize and control information, it never quite works out the way they want to.

Get the must-read newsletter for marketers.