Scaling Our SaaS Gross sales Coaching Platform with Rockset

Trendy Snack-Sized Gross sales Coaching

At ConveYour, we offer automated gross sales coaching through the cloud. Our all-in-one SaaS platform brings a contemporary strategy to hiring and onboarding new gross sales recruits that maximizes coaching and retention.

Excessive gross sales employees churn is wasteful and dangerous for the underside line. Nonetheless, it may be minimized with personalised coaching that’s delivered constantly in bite-sized parts. By tailoring curricula for each gross sales recruit’s wants and a spotlight spans, we maximize engagement and cut back coaching time to allow them to hit the bottom working.

Such real-time personalization requires an information infrastructure that may immediately ingest and question huge quantities of person knowledge. And as our clients and knowledge volumes grew, our authentic knowledge infrastructure couldn’t sustain.

It wasn’t till we found a real-time analytics database known as Rockset that we may lastly mixture tens of millions of occasion data in beneath a second and our clients may work with precise time-stamped knowledge, not out-of-date data that was too stale to effectively support in gross sales coaching.

Our Enterprise Wants: Scalability, Concurrency and Low Ops

Constructed on the ideas of microlearning, ConveYour delivers brief, handy classes and quizzes to gross sales recruits through textual content messages, whereas permitting our clients to observe their progress at an in depth stage utilizing the above inside dashboard (above).

We all know how far they’re in that coaching video right down to the 15-second phase. And we all know which questions they received proper and fallacious on the newest quiz – and might routinely assign extra or fewer classes primarily based on that.

Greater than 100,000 gross sales reps have been educated through ConveYour. Our microlearning strategy reduces trainee boredom, boosts studying outcomes and slashes employees churn. These are wins for any firm, however are particularly vital for direct sales-driven corporations that consistently rent new reps, a lot of them contemporary graduates or new to gross sales.

Scale has at all times been our primary challenge. We ship out tens of millions of textual content messages to gross sales reps yearly. And we’re not simply monitoring the progress of gross sales recruits – we observe each single interplay they’ve with our platform.

For instance, one buyer hires almost 8,000 gross sales reps a yr. Not too long ago, half of them went by means of a compliance coaching program deployed and managed by means of ConveYour. Monitoring the progress of a person rep as they progress by means of all 55 classes creates 50,000 knowledge factors. Multiply that by 4,000 reps, and also you get round 2 million items of occasion knowledge. And that’s only one program for one buyer.

To make insights out there on demand to firm gross sales managers, we needed to run the analytics in a batch first after which cache the outcomes. Managing the varied caches was extraordinarily laborious. Inevitably, some caches would get stale, resulting in outdated outcomes. And that may result in calls from our consumer gross sales managers sad that the compliance standing of their reps was incorrect.

As our clients grew, so did our scalability wants. This was an incredible downside to have. However it was nonetheless an enormous downside.

Different instances, caching wouldn’t reduce it. We additionally wanted highly-concurrent, prompt queries. For example, we constructed a CRM dashboard (above) that offered real-time aggregated efficiency outcomes on 7,000 gross sales reps. This dashboard was utilized by a whole bunch of center managers who couldn’t afford to attend for that data to return in a weekly and even each day report. Sadly, as the quantity of knowledge and variety of supervisor customers grew, the dashboard’s responsiveness slowed.

Throwing extra knowledge servers may have helped. Nonetheless, our utilization can be very seasonal: busiest within the fall, when corporations deliver on-board crops of contemporary graduates, and ebbing at different instances of the yr. So deploying everlasting infrastructure to accommodate spiky demand would have been costly and wasteful. We wanted an information platform that would scale up and down as wanted.

Our last challenge is our dimension. ConveYour has a crew of simply 5 builders. That’s a deliberate alternative. We might a lot moderately maintain the crew small, agile and productive. However to unleash their interior 10x developer, we needed to maneuver to the most effective SaaS instruments – which we didn’t have.

Technical Challenges

Our authentic knowledge infrastructure was constructed round an on-premises MongoDB database that ingested and saved all person transaction knowledge. Related to it through an ETL pipeline was a MySQL database working in Google Cloud that serves up each our massive ongoing workhorse queries and likewise the super-fast advert hoc queries of smaller datasets.

Neither database was reducing the mustard. Our “stay” CRM dashboard was more and more taking as much as six seconds to return outcomes, or it will simply merely trip. This had a number of causes. There was the massive however rising quantity of knowledge we have been amassing and having to investigate, in addition to the spikes in concurrent customers reminiscent of when managers checked their dashboards within the mornings or at lunch.

Nonetheless, the largest cause was merely that MySQL just isn’t designed for high-speed analytics. If we didn’t have the correct indexes already constructed, or the SQL question wasn’t optimized, the MySQL question would inevitably drag or trip. Worse, it will bleed over and harm the question efficiency of different clients and customers.

My crew was spending a mean of ten hours per week monitoring, managing and fixing SQL queries and indexes, simply to keep away from having the database crash.

It received so dangerous that any time I noticed a brand new question hit MySQL, my blood stress would shoot up.

Drawbacks of Various Options

We checked out many potential options. To scale, we thought of creating extra MongoDB slaves, however determined it will be throwing cash at an issue with out fixing it.

We additionally tried out Snowflake and appreciated some features of their resolution. Nonetheless, the one large gap I couldn’t fill was the dearth of real-time knowledge ingestion. We merely couldn’t afford to attend an hour for knowledge to go from S3 into Snowflake.

We additionally checked out ClickHouse, however discovered too many tradeoffs, particularly on the storage facet. As an append-only knowledge retailer, ClickHouse writes knowledge immutably. Deleting or updating previously-written knowledge turns into a prolonged batch course of. And from expertise, we all know we have to backfill occasions and take away contacts on a regular basis. Once we do, we don’t wish to run any experiences and have these contacts nonetheless displaying up. Once more, it’s not real-time analytics in case you can’t ingest, delete and replace knowledge in actual time.

We additionally tried however rejected Amazon Redshift for being ineffective with smaller datasets, and too labor-intensive usually.

Scaling with Rockset

By YouTube, I realized about Rockset. Rockset has the most effective of each worlds. It may well write knowledge rapidly like a MongoDB or different transactional database, however can be actually actually quick at advanced queries.

We deployed Rockset in December 2021. It took only one week. Whereas MongoDB remained our database of file, we started streaming knowledge to each Rockset and MySQL and utilizing each to serve up queries.

Our expertise with Rockset has been unimaginable. First is its pace at knowledge ingestion. As a result of Rockset is a mutable database, updating and backfilling knowledge is tremendous quick. With the ability to delete and rewrite knowledge in real-time issues lots for me. If a contact will get eliminated and I do a JOIN instantly afterward, I don’t need that contact to indicate up in any experiences.

Rockset’s serverless mannequin can be an enormous boon. The way in which Rockset’s compute and storage independently and routinely grows or shrinks reduces the IT burden for my small crew. There’s simply zero database upkeep and nil worries.

Rockset additionally makes my builders tremendous productive, with the easy-to-use UI and Write API and SQL assist. And options like Converged Index and computerized question optimization eradicate the necessity to spend beneficial engineering time on question efficiency. Each question runs quick out of the field. Our common question latency has shrunk from six seconds to 300 milliseconds. And that’s true for small datasets and enormous ones, as much as 15 million occasions in one in all our collections. We’ve reduce the variety of question errors and timed-out queries to zero.

I not fear that giving entry to a brand new developer will crash the database for all customers. Worst case situation, a foul question will merely devour extra RAM. However it should. Nonetheless. Simply. Work. That’s an enormous weight off my shoulders. And I don’t should play database gatekeeper anymore.

Additionally, Rockset’s real-time efficiency means we not should cope with batch analytics and rancid caches. Now, we will mixture 2 million occasion data in lower than a second. Our clients can have a look at the precise time-stamped knowledge, not some out-of-date by-product.

We additionally use Rockset for our inside reporting, ingesting and analyzing our digital server utilization with our internet hosting supplier, Digital Ocean (watch this brief video). Utilizing a Cloudflare Employee, we often sync our Digital Ocean Droplets right into a Rockset assortment for simple reporting round price and community topology. It is a a lot simpler approach to perceive our utilization and efficiency than utilizing Digital Ocean’s native console.

Our expertise with Rockset has been so good that we at the moment are within the midst of a full migration from MySQL to Rockset. Older knowledge is being backfilled from MySQL into Rockset, whereas all endpoints and queries in MySQL are slowly-but-surely being shifted over to Rockset.

When you’ve got a rising technology-based enterprise like ours and wish easy-to-manage real-time analytics with prompt scalability that makes your builders super-productive, then I like to recommend you take a look at Rockset.