Artificial Intelligence for Daily Business Analyses

JON REILLY: Hi, I'm Jon from Akkio, and we make it really easy to adopt machine learning in your business. Our backgrounds are actually uniquely suited to this task. Half of the founding team comes from Sonos, where we cut our teeth on product management and learned how to make really easy to use consumer products. And the other half of the founding team came up in a technical 3D printing company where they learned to make advanced software components to continuous carbon fiber strands inside nylon matrices.

And so that combination is really critical to our ability to deliver for our users because we're making a software application that's as easy to use as a consumer application, but it's an incredibly powerful business tool that allows you to optimize your data value extraction. We were actually all working together at a company here in Boston called Markforged that does strong part 3D printing. And two of us had met before at Sonos, like I mentioned, but the larger team met entirely at Markforged.

And while I was working there, I was leading the product team Abe was leading some of the software innovation teams. He's our technical CEO. And Craig was leading product. And then we had Ekin, who's a great user experience designer. And collectively, we were building the business, and I ended up getting involved in marketing. s while I was working in marketing, I realized that a lot of the workflows that are happening in the marketing organization need optimized-- your lead flow.

Which leads does the sales teams contact first? Which ones are not worth talking to? How do you respond to a customer support inquiry about your product?

And all of these were high-volume workflows that our team was struggling to handle. And the answer whenever you have a high-volume workflow like that is really optimization of your focus, so focusing on the most likely to convert customers, helping out the people with the biggest issues who can't self-serve in the support side of the business. And it was really just asking for machine learning, as machine learning is a perfect pairing to solve these types of problems.

But when we surveyed the market, we couldn't find any tools that would enable us to do that with our own data. And that's a problem because your data is something that only really well. You're an expert in what it means and what the driving factors inside of it are, and even like what is called, what the titles are.

And so we tried to use a couple of external contractors to get this done. We worked with some application-specific solutions. But none of them really were able to understand the core of our business. And so there was this communication gap in terms of taking advantage of our data.

And that's where the core idea from Akkio, about Akkio, came from, was that if we had a platform that made it easy for a non data scientist, a non-technical person, to look at their data, understand what's going on, and then leverage it in real-time decision making, that would be a massive improvement in efficiency in how we operated. And so that was the seed of the idea. And we got together and started Akkio based on that.

Akkio is built for analysts, and for operators, and anyone in a business who's working with a data-driven workflow. If you're using tools like Excel to do exploratory data analysis, or Google Sheets, or Snowflake, or things like-- you might even be using Tableau or Power BI. Anybody working in a business who is doing data analytics, trying to extract value from the data, we're a tool for them. And we make it about 10 times faster than your normal workflows to manipulate that data using natural language.

So you can just, in a chat interaction method, ask your data to be changed or manipulated or transformed or joined. And that's a massive acceleration in everyone's workflow and time commitment to get the data ready for analysis. But then we actually automate the process of extracting the patterns inside that data, and make it really easy to understand what's going on, and allow you to leverage it to extract value in your business.

Data is your business's new gold, and every business is going to need to adopt ML over the next five years in every workflow that they run if they want to remain competitive. And Akkio makes that really easy to do. Akkio is designed to work across a large range of data. So we have customers extracting value from thousands of records, startups and small businesses that are building us into their workflows.

But we also have customers running us against their Snowflake instances and building models on 260 million rows of data. So it works at pretty much any scale. And really, the question we get a lot is, how much data do I need to be able to extract value with machine learning? And I like to say, it depends on the complexity of the pattern that's hidden in your data. We process tabular data, but I'll use images as an example.

If you're teaching an ML model to learn the difference between a dog and a house, that's a pretty easy task because there's not a lot of overlap in that Venn diagram. But if you're trying to teach it to recognize the pattern between a dog and a cat, you're going to need a lot more examples. And if you're trying to figure out the difference between a dog and a wolf, you're going to need millions of examples because they're very, very similar.

The same thing applies to any tabular business problem. If your pattern is simple and straightforward and you're just trying to automate a repetitive task, you can get with as little as 1,000 examples of that task to teach the ML engine in order to automate it. But if it's a complex pattern, like what your best fit customers are, extracting the cluster of your best fit users in terms of lifetime value, and you have lots of similar users in your stack, then you may need hundreds of thousands of examples, or millions of examples of users in order to find those patterns. So really, it depends on the complexity of your business problem. But Akkio is a tool that makes it easy to get quick wins regardless of the size of your data.

When you get first into the Akkio platform, the first thing you do is you connect your data. And you can connect your data by uploading an Excel sheet or a CSV or a JSON file. Or you can connect a live data source, and that could be a Snowflake instance. It could be BigQuery, your CRMs, like HubSpot and Salesforce, Google Sheets, and a bunch of more integrations that are in the pipeline.

Once your data is connected, the next task you do is prepare that data for analysis. That means transform it, filter it, adjust it, maybe merge different data sets from different sources together. We make that incredibly easy to do. We shortcut the time it takes you to work with data by an order of magnitude.

And then, in the click of a button, you can train a machine learning model to predict your key outcome, whatever that is. And if that's your sales leads, then you look at your closed won and closed lost business, you pick that as your prediction target, and we'll learn from every other feature you have in your data about those leads to figure out what the patterns are that drive someone to purchase your product or not purchase your product. And then we show that to you in really easy to understand graphics so that you can look and say, OK, this feature is going to cause this user to be this much more likely to buy our product, whereas a user with this type of feature is way less likely to buy our product. Here's where we should focus our sales and marketing dollars in order to optimize our sales.

And then you can even take that model once you've built it and deploy it and take a new lead that comes in that you haven't ever seen before and run it against the model. Very easy in real time. And we'll tell you the probability that lead is going to turn into a sale, and the probability that lead is going to have a high-value sale or a low-value sale.

And you can do that-- so then, of course, you can take that information and you can act on it appropriately. So call the high-value leads first, call the next ones next, and maybe ignore the long-tailed leads that are never going to buy your product or churn very quickly. You can imagine very easily how that makes your sales process more efficient. But it's not just sales. You could apply that same process to any workflow in your business.

We have people predicting customer churn. So looking at your customers' history, the products that they use, how often they're engaged with the platform, and then identifying customers you've churned in the past or not churned in the past. And we can surface really low signals in your data that are hard to see through the noise if you're just looking at it in a manual, exploratory way that are the early indicators of probability at a churn so you can take action to intervene with those customers. You can flag the customers who are at risk early, and you can understand the risk factors associated with the churn so you can help reduce their probability of churn.

You can also do things like forecast your revenue or your inventory. If you're forecasting your revenue, we'll tell you where we think your revenue is going to go over the next period of time. We'll also tell you what the driving factors that are impacting your revenue are, how those are trending, and where they're going to impact your revenue over the next period of time. So it's really clear to you the three basic questions we're trying to answer for business-- any business-- are what's going to happen, why is it going to happen, and what can you do about it. And those answers are in your data. You just need machine learning to help extract them.

We have a company using us to optimize their marketing spend out of their Snowflake instance. So they build models directly against their 250-million-record snowflake database in real time. Then they use those models to look at new records as they come in, new interactions with their website, and they predict lifetime value. And they take that lifetime value prediction and they use it to optimize their spend with Google advertising. And they do that in real time.

We have companies who are using us to predict all sorts of different outcomes in their financial organization. So forecasting what's going to happen with your revenue, and your orders, and your payments from your customers, including looking at seasonality and ROI on your different expenditures. We have a large international shipping company that's using us to analyze the drivers of shipping cost and value delivery to their customers so their customers can pick the best shipping routes, minimize carbon footprint, and minimize cost when they're shipping things, and understand the probability that a package is going to arrive on time at a customer.

So just a broad range of potential applications in any business. Anywhere where you have data and you're optimizing a workflow, you're going to want to have the ability to customize a machine learning model for that specific workflow, and that's what we make really easy. We're a platform that makes it easy to take a specific problem in your business that is a data-driven problem and solve it.

The vast majority of solutions are probably a lot more application specific, meaning they're a only lead scoring ML tool. The problem with those tools is they don't really know your business very well, and they're a least common denominator average solution to the problem. Their right to exist as being experts on lead scoring.

Our right to exist as being experts on making it easy to customize an ML model for your unique business problem, and enabling the subject matter expert in your business to take advantage of machine learning because they're the one who actually understands what's going on. And we put a machine on that team so that they can leverage it to make better decisions.

[MUSIC PLAYING]