How to manage successful data-driven operations in startups
On July 11th 2019, we held a seminar with Partech portfolio companies on what makes successful data-driven operations.
The event was intended for Partech portfolio companies but we thought that some keypoints of the seminar could be made public to help all startups benefit from their expertise.
Here are some of the topics tackled and the speakers:
- Introducing automation at each stage of growth - James Kramer, VP of Risk & Data @ Yoco
- How to leverage data to insure a marketplace growth and liquidity -Oussama Ghanmi, CDO @ Ornikar
- Scientific marketing: Why have we created our own web tracking solution? — Sendinblue Armand Thiberge, CEO
- How we used data to drive our scaling organisation from 20 to 70 - Kate Toumazi, COO @ Privitar
- Tools to generate the right data to drive your company as a CFO/COO - Nicolas Godin, CFO @ Cardiologs
- Switching to data-driven decisions: Learnings from the food delivery industry - Edouard Nattée, CEO & Founder @ Foxintelligence
Have a look at some keypoints of the presentations:
Introducing automation at each stage of growth — James Kramer, VP of Risk & Data @ Yoco
Different stages of scaling require different types of automation.
Waht is Yoco?
Yoco is a South African start-up that was started by four friends in 2013 (for more info, see: https://partechpartners.com/companies/yoco/)
They saw huge opportunity in South Africa market. The fact that SMEs not adequately served from a financial perspective.
- This is because of how distributed yet small business in South Africa are, it makes it hard to address from a branch infrastructure perspective
First key opportunity in South Africa was around payment:
- 80% have access to credit cards, but only 10% able to take payment from them
Yoco is simple. You sign up online, purchase a credit card reader with the point of sale platform, which makes it able to track everything and run your business on the go. Also offers capital to borrow.
Differences between Yoco and traditional players
Yoco activation milestones
New customer activation is critical
- If its painful / difficult to use, people will go back to previous methods
- Fast and accurate delivery of credit card readers is critical for optimising Yoco customer activation
Four key stages of activation
- Signup and pay online
- Receive card reader
- Complete first transaction
- Receive first payout
How do you ship out thousands of readers every month etc., how do you keep this with scale?
When you start to think about scaling up any process, you have to think of it from a framework perspective
- Align the development of operations with life cycle stages
- Over development / under-development etc. needs to be avoided
Importance of matching execution capabilities with business and product maturity is key
Pilot → nail → scale → milk → kill — The life of any product
Process execution life-cycle
When they initially started the business, shipping the devices wasn’t really a priority. They would just go and deliver it in person.
Correctly timing the automation and resource investment of Yoco’s hardware delivery process has been key, today we focus on the scaling stage.
Started with In person delivery but gradually moved to calling a partner who delivers for you. This resulted in:
- Gains in efficiency, slight decrease in customer experience
- Previously delivered in person by Yoco
- Unstructured messy data
Then it went from calling a partner to excel file swaps, daily batches
- Using data to increase efficiency
- Batch data, KPIs now present
How can you scale this and continue into the future — how?
Defining the problems
Understanding the internal team’s day to day and experiencing the process as a customer helps to define the problem at a granular level
- Look at internal problems and customer problems
- Sit with all the different teams and agent and break down workflow, managing data becomes a bit of a mess
- Up to 23 hour delay due to single run daily batch processing
- Dealing with 2 companies whose internal systems are not in sync
- Information overwhelm — emails from yoco / phone calls from courier
Defining the solution
An automated process should always be simpler than a manual one, validating the solution as early on as possible is key.
About moving away from batch, and using integrated APIs:
- Once you have analyzed, you need to align
- Validate proposed solution with 3rd party tech resources
- Plan and cobuild timelines with BAs
- Validate process with operational managers
- Buy in from key leadership team members
- Size and scope granular technical work with engineering pod
- Validate current and proposed data model with BI team
Implementing the solution
Clear structure and processes around build, change management and success measurement enables rapid deployment, iteration and learnings
- Build — agile sprints
- Simple scope per print
- Regular retros
- Strong co-ordinators close to day-to-day operations and also the development team
Measure success and realise value -
- Measure against KPIs specified
- Identify other problems
Using data in operations 1/2
Using data in operation 2/2
Proactive gathering of qualitative and quantitative data enables total transparency and quickly surfaces issues
- Using data both qualitatively and quantitatively
Prioritising data and operations, alongside product, as an opportunity to improve customer experience will reap significant benefits as you scale.
Digitising manual processes with real-world implications takes time, it’s different to optimising an already digitised process/system.
Proactively thinking about what data you require is critical when designing the system — this is different to a software engineering mindset.
Build a strong change management plan — when processes touch the core customer experience it’s not enough to simply flip a switch.
- All good building these automated processes but you have to think about how you will implement from an operational perspective.
How to leverage data to insure a marketplace growth and liquidity — Oussama Ghanmi, CDO @ Ornikar
What is Ornikar?
It is the first driving school marketplace in France and Europe:
- +900k users since 2015
- +1k new paying customer/day
- +700 instructors
- +97% customer satisfaction
Mainly have two products:
- Learn the basic rules of theory:
- Highway code product
- Customers buy annual subscriptions to access material, book their exam through the platform
2. Then the marketplace product:
- Connects independent instructors with students
- Two sided marketplace — instructors open slots, students book
There is a customer journey…
The flexibility of our theory training platform makes it hard to standardize the customer journey and therefore more complex the prediction of future new driving customers
Theory training→ theory test → driving lessons (median: 3 months)
Describing like this makes it simple, but it’s hard to standardize the customer journey.
People take different lengths of time to pass their test.
Growth of the theory training customer base is mainly driven by marketing expenses:
- Fully online
- High success rate
- 95% of first purchases
Growth of the driving marketplace is mainly driven by conversion, getting customers to advance from their theory test
Have to make sure there are instructors in the area and that they will get enough business…
The growth of our driving marketplace is a more complex process that relies on your capacity to insure a local supply at the right moment for our customers
Driving customers want available instructors within 5km radius immediately after passing the theory test.
Recruiting driving instructors is a long and complex process: 3 months to launch an instructor.
- Need to ensure sufficient weekly hours for launched instructors
Challenge: How to ensure marketplace liquidity in an exponential growth situation
2019 marketing plan has its eyes set on a 3x increase in the number of highway code users per month. How?
Supply team plan:
- How many instructors to meet potential driving demand?
- Where to launch new instructors?
- When to launch each instructor?
Until 2017, focussed only in France so far, launching in big cities
- When going all over France, we had to be more accurate in terms of demand and supply
Solution: Predict demand to connect marketing and supply plans
How did they do it?
Defined a roadmap of 5 phases
- Collect data
- CJ (customer journey) modeling
- Supply needs prediction
- Predictions automation
- Tools to the supply team
Combined different sources of data in order to model the CJ and redefine predictions
First-party: in collaboration with the product team, we put data traps along the CJ to collect driving preferences of our highway code users
Open Data: demographic data, local price sensitivity, school holidays etc.
Then modeling the CJ of typical customer taking into account regional specifities.
Cj models combined to data collected from different sources allow us to predict supply needs at a local level.
Data input → modeling → raw output → final output
- 2 years of historical data from theory users
- Marketing predictions of new monthly theory users
- Instructors CRM data
- Analytics DB: instructors activity data
- CJ modeling
- Assumptions +rules
- Local demand prediction until M+12
- For each region and month (until M+12)
After testing and refining the predictions based on the supply team feedback we automated the predictions and developed a dataviz tool to make the data actionable
Supply needs prediction automation
Turning data into action
Scientific marketing: Why have we created our own web tracking solution? — Sendinblue Armand Thiberge, CEO
What is Sendinblue?
Sendinblue is your all-in-one sales and marketing toolbox:
- 3,040 new paying clients every month
- 250+ employees
- 30k signups per month
- 80k active users
- 70M emails/sms per day
- 160 countries served
Main acquisition channels
- Brand (>50%) (word of mouth)
- Usually companies allocate a budget for each channel
- Sendinblue strategy is different
- Acquisition at Sendinblue is driven by CAC. Currently, limit themselves to an overall CAC of 250/month
Of course, you can’t only look at CAC by itself
- Always analyze their CAC with respect to marginal CA, which measures the sustainability of the current CAC with respect to the marginal cost of scaling
- Typically an exponential curve, first clients not that much but it goes up pretty quickly
- Everything is based on CAC so you need to make sure you measure it accurately
How do you do that?
Why is it not enough?
On GA, the proportion of clients coming from adwords is shown to be 25%, whereas the reality is more than 35%
This means that GA has some issues with accurate tracking. Why?
- Cross-device reconciliation
- Cross-user reconciliation
It also fails to take into account key metrics like LTV, churn, and payback. There’s a lot of server data that needs to be incorporated into your data and be computed
Build something in-house for more visibility
- Done in the early stage of Sendinblue
- Track all sessions with a domain cookie
- Use a library to record everything (pageviews and events): started with Segment → switched to Keen.io + GTM
- Process the events using Dataiku
- Analyze all channels using Qlik
Problem is to know what was the first acquisition channel, you can do that with Dataiku
Okay, but which attribution model should you use?
Your choice should be based on your understanding of the buying process and what makes the most sense for your business
At Sendinblue, they focus on the first click. Why?
The most difficult part of the customer acquisition process is to get people onto the website. Once they arrive, we’ve already won because we know how to convince them.
Solving cross-device and cross-user reconciliation
Check all the sessions and all the IPs
- Can work out the first session and the first source
- Much more important than you can think
Thanks to this scientific approach to marketing, you can develop a better understanding of your customers and gain more control over your spend
How we used data to drive our scaling organisation from 20 to 70, with Kate Toumazi, COO @ Privitar
What is Privitar?
Privitar is a privacy engineering software company, that enables organisations to use, share and derive insight from data safely.
- Founded in 2014
- Enterprise B2B Software
- Average deal size 2 years ago £80k pa
- 20 People Org structure was 3 direct sales reps, 3 marketing, 2 services 4 G&A + 8 Engineering
What was in place with 20 px
- Marketing tool
- Sales tracking
- Company handbook/documentation
- JIRA for Engineering
What was preventing scaling faster?
- Unpredictable sales
- Lack of confidence to invest based on pipeline
- Slower overall scaling
Approaching the problem
Understand baseline situation
- Collect data
- Break up the problem
Identify key areas for improvement
- Outliers / benchmarks
Drive improvements and review for new issues
Continually reevaluate based on current scale
Challenges facing the business
Lead → Opportunity → Customer → Implementation → Customer Success
Lead — Lack of inbound leads, not enough pipeline generation
Opportunity — Long sales cycle, pipeline unpredictable and kept slipping, deal forecasting inaccurate
Implementation — Services not engage early, services hiring lagging
Customer success — renewal focus lacking
What did we change?
Hubspot → Salesforce, Yesware, Hubspot, Socialbble, Hotjar
- Storing History
- Driving consistency in approach across teams
- Following activities from start to finish
- Data cleansing
Prospecting opportunity / define need → Define use case → technical validation → proposal / negotiate → Closed / Won
Data provides transparency
These three aspects are key in providing insights:
- Performance by campaign
- Engagement by persona
- Performance by channel
- Performance by rep
- Avg time to qualif
- Performance of outbound vs inbound leads
- Quality of Marketing qualified leads
- Performance by rep
- Win/loss analysis
- Time at sales stage
- Sales cycle by ACV
Benchmarking — how do we know what good looks like?
- Formal benchmarking
- Peer discussions
- Over reliance on the data
- Over burdensome process
- Too small a data set
- Forgetting to listen to the field
- Doesn’t reflect the future
- Be bold!
How do things change over time?
Overall business changes that require a different approach
- No longer possible to share insights based on ad hoc conversations.
- Historical data now becomes meaningful/ trends
- New staff on-boarding.
- Inside sales activities
- Services process
Going from 70 to 150
- More real-time data for decision making
- Greater visibility of key data across the company
- Global expansion
- Upgrading systems to support greater analysis
- Product roadmap and market data
Tools to generate the right data to drive your company as a CFO/COO, with Nicolas Godin, CFO @ Cardiologs
What is Cardiologs?
Cardiologs is a medical technology company committed to transforming cardiac diagnostics by utilising medical-grade artificial intelligence and cloud technology.
- 200,000,000 people will die from heart disease in the next 10 years, in the world.
- In the US only, annual costs of heart diseases are projected to reach $1.1 trillion in 2035.
- Cardiologs aims to democratize expert, personalized, cardiac care, globally, through medical-grade AI & software.
The CFO job just got harder
People usually view the CFO as the “numbers person”. Now the modern CFO’s office has all the information and data to be proactively forward-looking and steer the business.
But there is a big problem
Resources are scarce and expectations are high!
Back and middle office should scale
The CFO office is not only about invoicing or budgeting, but about all the processes in your company because you want metrics!
There are tons of helpful products out there, don’t build things by yourself.
Think in terms of opportunity costs and start working on it as soon as possible.
What are the benefits?
- Decentralized processes: no written procedures, everything is set up in the tools (workflow, limitations) = accountability, trust, empowerment, focus on value-added tasks
- All the data you need are connected. You don’t need a CFO anymore!
Switching to data-driven decisions: Learnings from the food delivery industry, Edouard Nattée, CEO & Founder @ Foxintelligence
What is Foxintelligence?
Foxintelligence is the ultimate data source on your market and your competitors. At Foxintelligence they put technology at the heart of market research to bring unique information.
- Technology for scale
- Next gen consumer insights
- Privacy first
How do they do it?
Cleanfox — Data acquisition and collection
- The most advanced inbox cleaner
- Millions of real online transaction emails anonymized and processed in real time from a panel of 2.5M consumers
Foxbrain — Data processing and enrichment
- Collect data from order confirmation e-mails and enrich it with assets and AI
- Learn everything about the consumer, transaction product
Foxwatch — Data analysis and visualization
- Proprietary software to access the Foxintelligence dataset
- All the KPIs at granular level for your market
Covering all major merchants and brands across 5 verticals
What it means to be data-driven — Learnings from the food delivery industry
Defining a zone where data is king
- Food delivery can be translated in different equations.
- Replace unknowns by undisputable datapoints.
- Overall performance, sourcing, pricing, promotions, category management, CRM
Having the right person and the right setup
- Dedicated person/team for internal and external data sourcing and processing
- Someone/team with real data skills and direct data access
- Someone with a data roadmap
- Someone eligible to Exec committee and someone who can summon the CEO
- A real budget
What does not seem to work
- Delegation of the data responsibility
- Smart person with no technical background
- Anyone overwhelmed by operations
- Working in a cave
Making sure everyone can access the data
You don’t want to miss our next wrap-ups and blogposts? Subscribe to our newsletter and each month, you’ll receive our recommended reads!