Machine Learning has been around since the 1950's. Yes, it is really that
old and has fallen in and out of favor over the last 7 decades. We don't
focus on the hype but rather on using Machine Learning as a tool in our arsenal.
We've had great success integrating both old and new Machine Learning techniques
and stay on the cutting edge of available algorithms and learning models.
We built some of the first sentiment analysis algorithms, support vector machines
and are currently pilot testing SLIDE algorithms. We won't build complicated
and cool sounding algorithms just for the sake of it, though. We only recommend
and implement Machine Learning algorithms when it is suited to the job at hand
and provides definitive results. Below are some examples of Machine learning
implementations we've completed.
Analytics and KPI Measurements
Numbers and data are how we make clients successful. We analyze clients
data to impower them, and lead them to make more informed decisions. In one
extraordinary case, we had a client that had built an incredible data warehouse.
This data solution was created simply because their competitors in the industry
were doing the same. They had little interest in actually using the system. We did and
leveraged their system to provide fantastic insights, but still couldn't get them to acknowledge
the results or the importance of their data. As an add on to their system, we developed a
simple analytics dashboard of some of their key metrics combined with the Machine Learning results.
They could not believe the reports and started using the dashboard on a daily basis. It has been
used to drive several development efforts throughout the company. Some
times visualization is as important as the algorithms themselves.
One of our clients exists in an industry without much direct competition.
They never had an eCommerce footprint, because they didn't see the need.
However, once one of their competitors opened an eCommerce platform, they
tried everything they could to bring their own to market. 8 months and
$5 million later, they were sitting on a site that generated less than 3%
of the company's overall sales. After all of these sunk costs showed so
little of a return, they called us in. With a complete overhaul of the site,
at a fraction of the time and cost, we increased the platform's
sales 500% week over week, making it the largest sales channel in the
company. One notable improvement was a Machine Learning address matching tool.
It single handedly increased customer matches in their billing system by 125%,
allowing for improved in footprint identification, accurate offer pricing and
Another client of ours had a much different problem. After meeting with them
and quantifying their data, we found a lot of their support calls dropping
into their sales queues. They had all of the old standard phone solutions,
but couldn't manage to keep support calls out of sales. We set up a number of
Machine Learning attribution models to find the underlying issues.
The problem, it turns out, wasn't their phone support, but their self help solutions.
We implemented a series
of sprints for their self help platform, each dropping inbound support
calls by 50-100 a day. After 6 sprints, they were seeing a consistent
300-600 less support calls a day company wide. This drastically reduced
the load on the support queue and thusly alleviated a much larger percentage
of support calls from the sales queue.
One particularly successful client already had an extensive eCommerce
footprint. It generated about as many sales as their inbound sales calls.
They had even automated some of the simpler orders, so no person was
involved until fulfillment. We noticed, however, that their platform was
far more sophisticated then what they made use of. We linked the platform's
data to a Machine Learning algorithms, focusing on automation, and worked with them on
finding existing pain points. After four months of development work,
we had increased their automated orders by 700% and pushed them to the
top sales channel for the company.
Any IT department worth their salt is always monitoring uptime. The last
thing anyone wants is for a site, service or feature to go down and to not
know about it. We routinely build out automated service monitoring at a functional level
(is the service still up), but some of our Machine Learning monitoring is where we
see far more granular and interesting results. Using historical data and learning algorithms, we create alerts
that trigger when a service goes outside normal operating parameters. These alerts
don't indicate a service is down, but that it is not operating normally, which can
lead to an outage. One client uses this to huge effect. We have reduced
downtime and outages by 400% by triggering these early warnings.