How Spenmo Sucessfully Built A Self-service Culture With Holistics
By honing in on three crucial components - People, Process, and Tool - the 8-person data team at Spenmo was able to establish a thriving self-service culture that empowers 200+ business users to answer their own data questions without writing SQL, which allows the data team to reclaim valuable time to focus on high-impact projects.
How It Started
Based in Singapore, Spenmo is an all-in-one software consolidating corporate cards, multi-currency dashboards, automated bill pay, and employee expense claims.
Michael Han - Head of Data at Spenmo, oversees a team of eight members - including one data engineer, one analytics engineer, one data product manager, and four data analysts who successfully enabled self-service for around 200 business users.
But in the past, it was a different story.
When Michael joined Spenmo, the business was experiencing exponential growth, and the volume of internal data requests came with it. It was clear that the data infrastructure would have had to scale with the business, or the data team risked getting bogged down with supporting simple requests instead of pursuing more impactful data work.
Michael, who had previously experienced the burden of a bulky analytics setup with a lot of legacies in larger tech companies, knew that his organization needed a self-service analytics setup. As a long-time follower of the Holistics newsletter on Business Intelligence, Michael decided to adopt Holistics.
The rest is history - the partnership between Spenmo and Holistics is now two years strong and thriving.
In a recent talk at dbt Singapore Meetup, Michael and Jing Yu - Senior Analytics Engineer at Spenmo - shared their self-service strategy and lessons learned setting up self-service culture at Spenmo. All photos in this case study are credited to Michael and Jing Yu’s slide deck.
01. Low-Impact Work That Damages Analysts’ Productivity
Michael defined low-impact work as maintenance, repetitive work that is devoid of any meaningful progress toward achieving business objectives - and he described them as brutally painful.
Having been in analytics for 8,9, and 10 years. It’s always painful whenever you have to go back and rewrite your SQL pipeline or you have to deal with downtime on the databases. A lot of times when something breaks, your day is gone. Because whether that comes to debugging, finding the problem, getting to the person to fix the problem, rerunning the pipeline, or making sure the data landed in the dashboard, that whole thing is just brutal.
Because of this challenge, Michael was a meticulous architect when it came to building Spenmo’s data stack. He knew that the tools they chose would play a crucial role in the productivity of the data team.
02. Supporting 200+ Business Users With An 8-Person Data Team - Without Letting Ad-hoc Requests Clogging Up Analyst’s Capacity
As companies expand and their analytical needs belly up, data leaders often resort to hiring more data analysts to cope with the increasing demand. However, every data request has to be routed through these analysts, leading to frustrating request queues, and the bulk of the analyst’s job is turned into English-to-SQL translation.
Having previously worked in large tech companies with cumbersome legacy systems, Michael foresaw the damaging effects of this “data monkey” work - which often lead to demotivation and mounting frustration among data analysts.
As the business user base at Spenmo expanded to 200, and the number of available data analysts remained at 8, the need for a self-service solution became urgent.
It comes down to resourcing issues. On a very high level, we’re in an organization of 200-300 people and the Data Headcount is 5%, which means one data analyst supporting potentially 19, 20 stakeholders. So realistically, if you’re just dealing with ad-hoc requests - you’re dealing with like “hey I’m interested in XYZ, can you pull this data for me” - there’s no work being done.
The Solution: People, Process and Tools
Michael and his team took a multi-faceted approach to tackle the issues of low-impact data work and lack of self-service data exploration. The team crafted a three-pronged approach that included a powerful combination of people strategy, streamlined processes, and fitting tools to break through the data roadblocks that once stood in their way.
01. People: Empathy and Education
The data team at Spenmo recognized that various business units had different data requirements and needs, necessitating a tailored approach for each team.
- For example, they understood that the strategy team needed ad-hoc analysis while the operation team only required monitoring dashboards. To cater to the operational team’s needs, Spenmo data analysts developed evergreen dashboards that required no further changes.
- For the strategy team, instead of creating a plethora of ad-hoc reports, the analysts guided them on how to ask the right questions, what type of data request to make, and how to get the correct answers using Holistics.
So we tell them first like - what it is that you’re looking for? And then once they understand what they’re looking for, then they make a request, and we show them how to ask the right question, what is the context of this request? And then we teach them how do you use Holistics.
Spenmo’s data team puts their users first and constantly asks themselves: “how can I help them succeed?”.
The answer to this question is the variety of thoughtfully designed datasets that can cater to different analytical needs and maximize the number of use cases to be served by each data table.
02. Process: Documentation and Request Flow
At Spenmo, the data team implemented a meticulous process to promote self-service among business users.
Data analysts work closely with the teams and set clear expectations, being transparent about what they can deliver. For instance, when approached with a data request, the data team informs the business user that the turnaround time could take up to 8 weeks. However, they offer to guide and support the user in answering their own data questions with Holistics - which can be accomplished with just a one-hour training session.
One of the ways we do it (encourage self-serve) is basically just saying “if you have things that you want us to help you with, here is a channel for you to raise that request. But know that the turnaround time when you make that request is 08 weeks.” If you didn’t tell us when you started the project, then you need to build it yourself. And we encourage that by giving a lot of help if they try to do it themselves, like” if you need help with it, you build it, then we’ll tweak it for you, or we’ll have a session with you to help you build it.
They went the extra mile to create an extensive documentation system for data tables they loaded into Holistics with details specific to every dimension and fact table - which helps to ensure that:
- Business users are aware of which data is available.
- Business users understand which datasets to use for exploration.
- Business users understand what data to expect in every column.
- Business users understand the source table which was used for populating the data (important for validation).
03. Tool: Empowered With Holistics
As Spenmo’s data team pursued the vision of setting up a maintainable, extensible, and reusable data stack, they sought a BI tool possessing these qualities. After a two-week trial, Holistics emerged as the clear choice. Holistics’s product philosophies are aligned with Spenmo’s vision, complemented nicely by the balance between value for money and rich functionalities.
Spenmo’s favorite features include:
Feature: Semantic Modeling Layer
With Holistics’ Modeling Layer, Spenmo analysts love how they can easily manage data logic centrally, define the data model once, and reuse any part of it across the system. This eliminated the need for analysts to repeatedly write SQL queries, saving time and reducing errors - which fits in nicely with Michael’s philosophy of designing a maintainable analytics stack and allow analysts at Spenmo to spend less time on maintenance ad-hoc work, and more time driving business forwards with data.
Spenmo’s analysts also rave about the ability to include metric calculation as part of datasets - which allows users to quickly drag and drop from these metric fields instead of creating the calculation themselves.
Feature: Analytics As-Code
As analysts at Spenmo are eager to adopt software engineering best practices in BI workflow, they found Holistics’s AMQL is a nice touch - as it’s tailored for better reusability, composability, development productivity than currently possible with existing tools:
AMQL (Analytics Modeling & Querying Language) is an integrated set of 2 analytics-as-code languages designed by Holistics to enable data analysts to define analytics logic in code:
- AML: a declarative language used to describe data semantic model and analytics objects such as database tables, their relationship, and visualization.
- AQL: a query language that leverages data semantic model defined in AML to query SQL databases in a higher abstraction manner, especially composing and reusing metric-based queries
Using Holistics’ AMQL, analysts can define analytics logic and govern it with Git version control, allowing for code review and promoting reusability. AMQL allows analysts to easily write code to define metrics, create reports, and build dashboards with a powerful, intuitive syntax.
Feature: Intuitive Drag-n-Drop Report Builder
Holistics empowers business users at Spenmo to efficiently explore data and build reports on their own, eliminating the need to wait for the data team. This is facilitated by a robust documentation system and the data analysts’ close guidance, allowing non-technical users to quickly grasp the know-how necessary to answer their own data queries with confidence.
Aggregated and snapshot datasets complement this functionality (drag-n-drop report builder) really well. Having these pre-aggregated datasets simplifies the drag-and-drop process for non-technical users - they can achieve complex aggregation results with less effort spent on figuring out the drag-and-drop steps.
The number speaks for itself.
With a well-designed self-service strategy, supported by a BI tool designed for data analysts’ productivity and self-service data exploration, Spenmo has seen significant improvements in the organization’s self-service culture.
- 25% Dashboard Platform Stickiness: % Daily Active User/Monthly Active User. This is equivalent to 7.5 days of usage per active user.
- 99 Daily Active Users: Spenmo’s 8-person data team effectively serves 40% of the company’s analytics demand, while still having time to pursue high-impact, high-leverage data initiatives.
There will always be some amount of ad hoc. But we are quite strict about it - we’re only doing a certain amount of ad-hoc per month. We can keep this low with Holistics. So our team overall is very productive to say, okay, we get to hire people on doing hard things. We get to build a very complicated dashboard. We build important data integrations. We do very important analysis.