Likvido is a fast-growing and ambitious tech start-up based in Copenhagen with 30 employees working from all over the world. We were founded in 2018 with the vision of closing the €400 billion financing gap for SMEs, and free SMEs from manual and time-consuming bookkeeping tasks.
We approach this vision in three steps:
Step 1: Automating their order-to-cash process and moving SMEs’ operating system online
Step 2: Rethinking credit scoring based on alternative data
Step 3: Providing SMEs with financial products “on the go.”
We have built a SaaS suite that automates the entire invoice-to-cash flow for SMEs and puts us in the centre of their cash flow. We call it an Accounts Receivable Automation Software. When the entire invoice-to-cash flow is digital and automated in our SaaS solution, we have a massive amount of structured data we can use to make real-time transaction-level credit scoring which we can use to provide SMEs with a range of financial products.
At the same time, our customers save massive amounts of time, get paid faster and get actionable insights.
1. The credit rating of new customers: Before selling on credit our customers can get a simple credit rating on new customers, warning them if there is a high risk of that specific customer paying late or defaulting on the payment.
2. Invoicing: Our customers can create a new invoice directly from Likvido, or use our two-way synchronisation to their accounting software to import invoices.
3. Payment, reconciliation and bookkeeping: We handle the payment from our customers' debtors, and have automated the reconciliation and bookkeeping process.
4. Paymentreminders:Ifourcustomerdoesnotget paid on time, a customised “marketing automation” campaign makes sure that the debtor gets reminders and late payment fees.
5. Third-party debt collection: If debtors do not pay after reminders, we offer third party debt collection in 180 countries recovering most of our customers' bad debt.
6. Task management and workflows: Our customers can automate the entire AR-flow, or combine it with internal tasks (such as calling debtors). To manage all manual tasks, we have built a rule-based task- management system.
The direct benefits for our customers include:
Real-time transaction-level credit scoring
A credit scoring model used for a traditional bank loan looks at historical data from financial statements to predict if the SME can repay a loan over the next 2-5 years. This process can be very difficult because the data is often outdated, and it is always hard to predict what will happen in a couple of years.
What we do is very different.
We use real-time transaction data to predict if a single invoice is going to be paid within the next couple of weeks. Naturally, this is not simple, but with real-time data and a shorter period to cover, chances of accurate predictions are much higher.
We use both traditional credit data and alternative data to power our machine learning (ML) algorithms.
Traditional credit data (business information)
Transaction data (from ERP / accounting software)
When signing up for Likvido, our customers synchronise Likvido with their accounting / ERP solution. By doing so, we get access to all their accounting data. Today, we have access to data from 520 SMEs, which lets us analyse approximately 2 million transactions.
The data we get includes:
Bank account information (via AISP)
Since PSD2 has been implemented, we can get access to all European bank accounts with few clicks. We use these data to look into our customers:
ML models and output
We use machine learning algorithms and statistical models to predict:
• The dilution of incoming and open transactions: What is the risk that the invoice is not going to be paid at all? (payment default)
• The lateness of payments: When will the invoice be paid
These predictions are calculated into a single numeric score for each of our customers, to make the prediction actionable. As our machine learning models are developing patterns from its history, the self-improving algorithms uncover intricacies of SME behaviour, allowing for a far more accurate and finely-tuned means of assessing the true creditworthiness of SMEs.
There are several algorithm choices including Penalized Regression, Random Forest and Boosted Trees. Random Forest (R.F.) for example, has certain key benefits including ease of use; lack of significant hyperparameters; lack of fitting bias; ability to allow large amounts of features as inputs; and resilience to outliers and feature collinearity. The following table outline the pros and cons of three common ML algorithms:
A major challenge in an ML model is controlling the trade-off between bias and variance (we want low bias low variance). For example, having distinct individual models for each client could result in high bias and low variance. From a machine-learning perspective, it is beneficial to leverage data across all suppliers to build one model. The algorithm is unbiased and develops unseen connections across different types of clients to produce highly predictive performances. A single model is much more robust and generalisable if we can include more buyers and sellers. The generated features become more important, and the model will depend less on I.D. fields, particular to a single dataset. ML models benefit from diversity in the inputs, and pooling across many sellers and buyers will allow the algorithms to lower variance, rather than concentrating on the precise details of a single client that may not apply to others, which leads to higher bias.
By automating the invoice-to-cash process, we get data to make real-time transaction-level credit scoring. We can now use these data to provide our customers with tailored financial products on the go, based on their creditworthiness and needs.
The two most obvious products are:
• Invoice based credit insurance: We let our customers insure their total revenue or a single invoice
• Invoice based lending: We provide working capital based on the creditworthiness of our customers' invoices
We will partner with institutional investors who can provide capital and credit insurance companies who can underwrite credit insurance.
59 sage.com
60 sage.com
61 Nuance: The Remind Me Generation