Billing

A great introduction on all the various aspects of a billing system, from currency to invoices, including great illustrations on the logic of changing plans. All these topics are detailed in dedicated sections below.

Links to the Billing and Cost Management section which details the broad characteristics of billing for a cloud provider.

Get rid of integers and floats for monetary values. Use decimals. Or face the risks of anomalous x100 charges.

Huge and complete collection of pricing scheme, with their psychological effect and impact on revenue model.

A list of 15 different ways to sell products and services.

38 models for inspiration.

6 models of monetization relying on network effect.

A manifesto inspired by the DevOps movement in which pricing is no longer ossified, but is practiced as a responsive iteration process, and implemented as a flexible property of the system.

A collection of 1000+ pricing pages for inspiration.

“The most important reason I like this pricing model is that it heavily incentivizes both the customer and the service provider to act in everyone's best interest.” Also details issues with other pricing models.

The reason usage-based pricing makes even more sense for cloud services: to optimize ROI, keep all regular workload in traditional architecture and reserve cloud computing for elastic and experimental projects.

“The socially optimal operating point is unique, and can be sustained by a linear, usage-based tariff, which charges a fixed price per unit resource and unit time.” This paper justify usage-based pricing for cloud resources.

“The strategy of improving user service quality is discussed first, followed by the pricing strategy for cloud resources to maximize revenue.”

“Billing is not complex on purpose: it's the price to pay for elasticity.” (source) - Or why you're likely to get an endless stream of complaining users if choosing utility pricing scheme: while accurate to the (milli-)cent, this model is frustrating for customers not ready to invest time grasping the underlying concepts.

The starting point about quantization of usage.

Implementing usage-based pricing is tricky, and this algebra will help you organize temporal reasoning. Also see this Stack Overflow question with clean schema.

Behind the quest of this developer to track its expenses, you get a glimpse on cloud billing difficulties. While not explicitly pointed out, pricing cloud resources is hard and the result of quantization, granularity and rounding, between space, time and currencies.

An example of really tricky burstable instances which accrues and limits its own amount of CPU usage credits.

Personal anecdote on the design of the pricing plan for AWS Simple Email Service.

Usage-based pricing is more optimal and more fair: it “allows a customer to start at a low cost, minimizing friction to getting started while still preserving the ability to monetize a customer over time”.

Before the cloud, there was another metered resource priced by its usage: electricity. Here is a detailed (and multi-lingual) taxonomy of its vocabulary from the International Electrotechnical Commission.

Open-source metering & usage-based billing in Ruby.

An open-source, fast (50k+ CPS) and scalable (load-balancer + replication included) real-time billing for ISPs and Telecom Operators, written in Go.

“In low-touch SaaS, the most common way packages are presented are as different columns in a pricing grid, with each column corresponding to a plan, offered at a different price, with differential access to features or maximum allowable usage along some axis interesting to the business.”

Open-source project to manage pricing & packaging infrastructure.

Open-source license key generation and verification tool in Go.

Beyond linear pricing, there is a place for additional platform fees and free-tiers in the pricing structure.

"There are two ways to make money. You can bundle, or you can unbundle." - Jim Barksdale.

Which numbers should you use? How high should it be? Should it be rounded? This guide has 42 tricks to help you choose the best price.

Pricing is an offensive tool to reinforce your product value and underscore the company core marketing message.

Explain how to articulate the pricing of SaaS businesses around the product strategy.

Explains why it is a sloppy analogy.

A couple of ways to re-license your product.

“In insurance, the price is the product. (…) How do I change price to make shitload of money?” Or how to deliver results (optimal price for a customer) instead of delivering data (calculating a customer's risk, which had been the standard approach used by actuaries previously).

Used in surveys to determine the price for a new product or service. The results can be used to produce a demand chart and a revenue curve.

PSM is a market technique for determining consumer price preferences. Allows for drawing a revenue curve to estimate the price point delivering the maximum revenue.

“The most compelling argument against simply picking a price, though, is that it limits how much you can learn about your market.” Then the author setup vickrey auctions to discover price.

“To find the optimal price that provides maximum revenue for your product you will need to field-test several price points; you then reconstruct the revenue curve and find the peak.”

Some cases of price personalisation are prohibited in EU due to consumer protection and data protection rules.

All GCP SKUs available as an API.

The open-source variant of the commercial Akeneo products.

An open-source UI and database to manage product metadata, written in PHP Symfony.

A streamlined open-source project, focusing on PIM-only features.

Comparative study of the 3 projects above.

A comparison matrix which use ECU (Amazon's vCPU) as a CPU power measure unit.

Easy Amazon EC2 Instance Comparison.

“Comprehensive introduction to forecasting methods and to present enough information about each method for readers to be able to use them sensibly.”

Discusses how Uber apply data science and machine learning in their financial planning platforms.

Great introduction on how time series can be used to evaluate business problems.

A more detailed tutorial on how to use past data to make temporal forward predictions. And be aware of the differences between sktime and the Prophet project mentioned in the article above.

Python library for user-friendly forecasting and anomaly detection on time series. It wraps a huge number of models, including Prophet. Great for experiments, but bear in mind that all the models in Darts expects that your data comes at a very regular interval, and make a lot of assumptions about their shape.

Open-source tool to stay under budget by uncovering hidden costs, monitoring increases in spend, and making impactful changes based on custom recommendations.

An example of consumption trend line for resource consumption.

A full-featured hosted product to perform forecasting on any time-series.

“The biggest cost savings there are: 1. Turning things off that you're not using; 2. Then spot instances; 3. Then reserved instances.”

This paper describes the pricing strategy model deployed at Airbnb.

A collection of papers on bidding and auctions.

Hinted by an HN comment, in which yes, “‘Asking people what they'd pay for and how much rarely works.’ (…) Using a Vickrey auction, similar to Google's ad auction mechanism, can elicit a person's maximum willingness-to-pay.”

“Which comes first, the supply or the demand? Chicken or egg?”

How to Kickstart and Scale a Marketplace Business: Constrain the marketplace; Decide which side of the marketplace to concentrate on; Drive initial supply; Drive initial demand. A 4-parts series with dozen of interviews of people with direct experience building and scaling marketplaces.

A bit of vocabulary: “In a casino, the term rake refers to the commission that the house earns for operating a poker game. (…) While casinos use the term rake, a plethora of interesting word choices exist which all describe the same thing – keeping a little bit of the revenue for the company that is running the service.”

Paper exploring the “mechanisms for the allocation of processor time and storage that are compatible both with programming practice and with market mechanisms.”

“How should the consumer modulate its demand to optimize its profits? (…) How should the provider and consumer negotiate the specific pricing structure they will employ?” Covers non-linear models, tiered pricing, elastic demand, consumer et provider strategies.

From AWS's original auction-based market (2009-2017) to provider-managed pricing across all major clouds today. Documents how transparent bidding was replaced by opaque algorithms.

“Amazon's spot price is unlikely to be set according to market supply and demand. Rather, price oscillates within a very narrow band most of the time, which is more likely to be an artifact of some pricing algorithm with predetermined reserve price.”

“We examine the economics of demand and supply in cloud computing. (…) These results explain why fixed prices currently prevail despite the seeming need for time-varying dynamics. Examining the actual CPU utilization provides a lens into the future. (…) demand fluctuations would be on par with the three classic industries where dynamic pricing is important (hotels, electricity, airlines) and dynamic prices would be essential for efficiency.”

“The key idea of our algorithms is to use pricing signal to regulate user demand. One may argue that such an algorithm gives poor service to the user as it pushes tasks out of the queue in order to maximize cloud broker's profit.”

“Pricing policy of spot resources should be carefully designed and the impact on both present and future should be considered.”

“Cloud providers possessing large quantities of spare capacity must either incentivize clients to purchase it or suffer losses. Amazon is the first cloud provider to address this challenge, by allowing clients to bid on spare capacity and by granting resources to bidders while their bids exceed a periodically changing spot price.”

“Google's prices are fixed, whereas AWS uses a market model”.

When you're big enough that creating an internal secondary market for instances makes economic sense.

Tool that easily and reliably converts (all) your existing AutoScaling groups to cheaper spot instances.

A collection of budget pacing resources for targeted online advertisements.

How to scale to 5M bid request/s, 2ms Max Response Time.

Open Source Bidder and Demand Side Platform (DSP).

A general introduction to history and vocabulary of accounting.

Describes how to view accounting in the form of graphs in which money flows, then how these movements are materialized in the financial statements of a small company.

Same premise as above, but way more detailed and complete, as it adds reporting and implementation details.

An interactive form to play with accounting concepts.

Focus on the daily operations of recording and maintaining the transactions of a business.

“Reconciliation is a business process which arises almost entirely because of a lack of structured data in the pipelines that convey money between businesses”. There are some hack to streamline the process, like adding an arbitrary discount to produce unique trailing decimals, or setup a couple of virtual bank accounts as proxy.

Extensive list of open-source personal finance projects that might be great to look into for inspiration on double-entry accounting and bookkeeping.

And a list of graphical accounting tools: GNUCash (GTK+), Grisbi (C), Firefly III (PHP).

A complete tutorial on personal finance tracking with GnuCash.

Free Desktop book-keeping software for small-businesses and freelancers.

YAML accounting and JSON tax forms.

API that store, validate and reports asset movements. Also known as a Ledger. Sequence is immutable, scalable, and easy to use.

An open source accounting system that aims to make the recording of double entry bookkeeping transactions programmable.

A scriptable financial ledger, designed to make it easy to model complex financial transactions.

Open source software designed for small businesses and freelancers.

Design decisions behind a billing and accounting system. Key takeaway: some companies still pays by checks; do not use floats; complex customer workflows; date and time issues; mutable data.

Explain the basic data model of a double-entry accounting system relying on Google Spanner for internal use at Square.

A distributed financial accounting database to ensure that the money either moves, or doesn't move, that it doesn't get lost somewhere in between. It has been tested on Jepsen, which tested its strong serializability.

Core functionality of a double entry accounting system for Django.

A 'managed account' is an allocation of money that can be debited and credited.

“Triple‐entry accounting is a new and a more efficient way to address fundamental trust and transparency issues that plague current accounting systems. Triple‐entry accounting with Blockchain, when properly implemented, can fundamentally improve accounting.”

An excellent resource for implementing multi-currency accounting systems.

Helps you learn how to analyse and monitor your company's financial performance.

“Malone created a new accounting metric, something he called ‘earnings before interest, depreciation, and taxes’, or EBITDA.” This is how a cable company CEO made sense of cash flow, like real-estate business did. The article starts with this example and describes other cash flow games for SaaS models.

Allows you to fully understand how to use financial data to make better decisions for your business.

A framework for tech finance and business leadership teams to share the same language and processes of cloud operations and management.

Advanced abstract algebra applied to the analysis of accounting systems.

When contracts, invoicing and payments are not working together, you end up alienating your big enterprise customers. See the associated HN comments on bulk license purchase.

“The concept of Entitlements encapsulates the feature access settings under various product variants (aka Pricing Plans or Packages), bridging the gap between how a product is sold and how it behaves for a variant. In essence, entitlements are a set of permissions defining what a customer (paying or non-paying) can do with your software application.”

Explains the differences between various types of discounts and usage commitments in GCP.

“Implementing price reductions ranging from 9–70% for large purchases, we found remarkably little impact on revenue, either positively or negatively.” If discounts are a placebo and are still widely used in the industry, maybe they are a marketing hack to lure big customers?

Backblaze post-mortem on raising prices on their main offer. They wanted to create an extension program based on credit system. This became a full-time job for a handful of their most senior engineers, and resulted in a six month project.

The reason you have to add an time limit on credits: unused ones accounts as liabilities on our balance sheet.

A cautionary tale on how promo codes can be exploited to get unlimited free rides.

The next level of fraud: “an online collective exploited a loophole on its platform to ‘steal’ tens of millions of yuan worth of discount vouchers”.

European directive on the application of VAT when vouchers are involved.

Point the negative consequences of users encountering a blank coupon field without having a coupon. See the update at the end of the article with research backing up this anecdote.

A list of test cases for developers to test their tax code against.

Centralize applicable VAT rates on foreign online services, for each country of residence (including territorial exceptions).

List of countries requiring application of taxes on foreign provided online services.

A multi-states US initiative to automate and standadize sales tax accounting and collection.

“If you think you can just setup a simple Stripe integration and move on, like us, you'd be sadly mistaken.”

European Union reference on the common system of VAT.

Answer: a provision in which a business transfer the responsibility of VAT handling to the customer.

Invoiced billing is a B2B-friendly payment happening after the service is consumed and an invoice issued. It seems to be a pain to setup on GCP but I suspect is a consequence of trying to reduce (costly) fraud.

The information required on a EU invoice is detailed by Article 226, Section 4 (Content of invoices) of Council Directive 2006/112/EC on the common system of VAT.

Here is a great way to rely on GCP's HSM to digitally sign documents and provide an immutable audit-trail. Might be applied to invoices as well as contract agreements.

Go beyond the solution above, and timestamp immutable document directly on Bitcoin's blockchain.

Because an invoice is immutable, the only way to fully or partially cancel it is to produce a credit note.

Flexible invoicing desktop app with beautiful & customizable templates.

A self-hosted open-source application for managing your invoices, clients and payments.

Library to generate a simple invoices.

For generating and displaying invoices (ideal for commercial Rails apps). It allows for flexible business logic; provides tools for tax handling, commission calculation etc.

Zero-maintenance and smart FOSS CLI tool for generating beautiful invoices for services and expenses.

Deep neural network to extract intelligent information from invoice documents.

European Union introduced a "standard" in XML format, which comes with a set of security vulnerabilities.

European standard on electronic invoice.

Python library to support the e-invoicing standard for France and Germany.

Most invoicing software can read and write UBL documents (XML) for data transfer.

JSON Schema, Open Source Go library, global tax database, with conversion tools, all in one.

A benchmark of top payment providers, their pricing and models.

RDBMS were build for banks around transactions to solve that specific issue. Then NoSQL came forcing us to carefully implement systems to avoid double spending.

Or why you should be prepared for and work around gateway provider's outages.

The importance of payment ops architecture.

European rules on applicable fees for late payments.

The root cause? People wearing a lot of wool in low humidity air, generating a lot of static.

How ACH works: A developer perspective, part 1, part 2, part 3, part 4, [part 5](https://engineering.gusto.com/how-ach-works-a-developer…

Dropbox's experience trying to account for an unreliable payment provider.

Under a constant stream of user complaints about payments, WordPress created a dedicated subdomains to help customers understand unexpected transactions. The trick was to add this URL directly in the bank statement.

Open-source backend for payment processing.

Open-source infrastructure for financial technology.

Another set of open-source projects for financial projects.

When a Netflix data engineer revisit receipts.

“CVS is a drugstore much like other drugstores, with one important difference: The receipts are very long.”

Some credit and debit card companies have "updating service" that allows new account numbers and expiry dates to be shared with merchants. Visa's implementation is called VAU and Mastercard's is ABU.

Payment Services Directive 2, explained.

A system checking the matching of customer's billing addresses with the one associated with the credit card.

PCI-DSS compliant card system built on PostgreSQL and PL/pgSQL.

Collection of miscellaneous notes primarily focused on the common cases of routable accounts, i.e. checking and savings.

Free and public IBAN validation and calculation webservice.

Swift / BIC codes for personal use only.

A repository that seems to scrape the website above.

EU standard for QR codes used to transfer money between bank accounts via SEPA.

“In this article, we will learn about India's Unified Payments Interface. A four-year-old payment scheme that has been accounting for 40-45% of digital payments across India.”

A huge collection of everything that went wrong, and still is, with payment APIs from the past 2 decades. Any issue exposed in this article that gets unaddressed will end up as stolen money.

In which we learn that “once turnover hit a certain level, Paypal automatically put the business on a 21 to 60 day rolling reserve, meaning that up to 30 per cent of a company's revenue could be locked up for up to two months.”

“Idempotency prevents race conditions. Idempotency dictates that multiple identical requests from the same client do not result in a different final state.”

Describes a classic payment workflow, then how machine learning can replace hard-coded business rules and fine-tune the payment failures/retries loop to increase the success rates of charges.

Discusses a method to detect fraudulent activity (launching DDoS attacks, Bitcoin mining, …) by using non-intrusive, privacy-friendly data: billing data.

Section dedicated to fraud management related to user accounts, from our sister repository.

In this talk by Uber Eats, we learn from the biggest payment processor that “growing businesses with declining loss rates is extremely rare in the industry”. Also, fraud can takes several forms: chargebacks on non-perishable goods, promo abuse, refunds…

KYC is nuanced and fuzzy, because it is a stochastic process to reduce risk.

Papers sourced from several conferences on all sort of fraud: credit card, payment transactions, loans, customs inspections, money laundering networks, and more.

Open-source real-time transaction monitoring software for fraud and money laundering detection. This is just an engine to define rules, wight them, and apply them to transactions. There is nothing specific to payments or financial transactions.

Some examples of AML rules.

A practical handbook on how to identify patterns in transactions.

Describe a fraud technique known as "card testing", in which massive batch of stolen cards are checked for their validity against your API.

Suggestions involved either warning signals for trying to guess which orders might be fraud, or countermeasures to try to make things harder for the fraudsters.

Another tiny collection of features we can feed to a ML system to detect fraud. Comments on HN also unearthed more qualifying signals and derived geodata on the transaction.

Tutorial on how to rely on anomaly detection to implement suspicious card transactions.

An example of a platform's network effect, which allows to predict the likelihood of winning a dispute.

To prevent abuses, service provider must strengthen both credit card verification and the account creation process.

Or why relying too much on fraud automation might lead to disgruntled users.

Aggressively shutting down user servers is good from a business point of view to prevent fraudster abusing free resources, until it's not.

How to use statistical methods to classify applicants into categories to reduce risks. Lots of inspiration and research papers there to improve general scoring.

Digit distribution can be a signal of accounting fraud.

In the same spirit as above, this paper show that frequency of reported round numbers in election is a signal of man-made anomalies. Might be applicable in some fraud-fighting area.

“A loss function used in robust regression, that is less sensitive to outliers in data than the squared error loss.”

Simple way to detect outliers.

Standard manual accounting trick employed when isolating an error in a double ledger.

A new kind of fraud, in which apps silently keep charging users after the end of the trial period.

A billing team responsible for collecting payment is prey to this kind of fraud, in which the CEO is impersonated to commend exceptional payment transactions.

Skip the first 90% of this presentation by an AWS VP (which is about general system reliability). The last four slides are a good summary of what billing cloud services entails, especially the use of soft quotas to limit fraud.

Most of telephony frauds gravitates around billing and the metering of micro-transactions. This paper feature a taxonomy (on page 6) differentiating the root causes, vulnerabilities, exploitation techniques, and the way fraudsters can benefits from them.

Guide and recommendation on how to make auto-renewable subscriptions user-friendly.

Focus first on other things in the cart, like easing anxiety and second-guessing by adding reassurance (trust marks, testimonials) near the credit card and completion steps, as well as some guarantee language that you may have used earlier in the product exploration process.

Failed attempt to move the billing experience by from backend to frontend because of security.

A collection of pricing pages from various SaaS companies, to get inspiration on how to present your offers.

A great source of inspiration to get better visibility in your operations.

The 2 critical metrics being Customer Acquisition Cost (CAC) and Customer Lifetime Value (CLV).

Discuss the relationship between investments, profit and growth, and how it influence decisions on pricing, free trials and plan structure.

Metrics produced above have a greater reach, as they're used as important signals to validate product-market fit.

Simple yet effective interactive profitability calculator for startups.

Great breakdown of Visa business models and metrics.

A complete tour of the financials of a startup, which gives you extra-context on how the metrics you produce fit into the larger picture.

“On the surface, churn rate may seem like a natural proxy for changes in customer lifetimes. Let's dig into why that is not true.” Churn rate is not a meaningful metric to compute CLV: during the customer lifetime, the churn probability is not constant. Most of the time because of your free trial and vouchers. This article illustrate the influence of the distribution used to model the probability of a customer quitting.

A seminal paper, with a stronger approach compared to the one above: the exponential distributions is replaced by a geometric model, which is better suited to discrete time intervals like monthly contracts, the former being more appropriate to continuous time process.

Illustrates how retention can be better modelled using a survival function like Kaplan–Meier survival curves.

A refined model of CLV, used to segment users based on Recency, Frequency and Monetary value.

“How to use Python in a simplistic way to fuel your company's growth by applying the predictive approach to all your actions.” Relies on XGBoost binary classification.

A full-featured Python package to analyze your users based on their "alive" and "dead" states.

Python package to show the projected retention rates for each cohort and calculate the LTV of a given customer in that cohort.

“Data Engineers build data pipelines and infrastructure to ensure a constant availability of transformed data. Data Scientists analyze and build models from these data to develop new product features or drive the bottom line of the business.” As for AI professionals, their focus is on cognitive automation.

You don't need data scientists. “For ETL, hire data engineers. For reporting, hire BI analysts. The end.”

“Data science is best viewed as a form of company culture, rather than a set of technologies. However, many firms will try to create that company culture by acquiring data-science technology, rather than working on their culture.”

Because of precision: floats and doubles cannot accurately represent the base 10 multiples that we use for money.

“This is precisely the problem we have when trying to represent 10^-1, or 0.1 in binary. There is not an exact binary representation of 0.1 or 0.01.”

Floats were suspicious even before the first general-purpose computer (EDVAC) was built: “Von Neumann wasn't entirely sold on why we might need floating-point numbers. He was quite vocal in his critique of floating-point numbers.” The author of this article complement this critique by illustrating the issue of precision and rounding.

A collection of cases in which uncontrolled and untested spreadsheet models led to lost revenue, mispricing, poor decision making, fraud and systemic financial failure.

A blog collecting and spreading ideas on how to use Python more effectively in the business setting.

Connect and query your data sources, build dashboards to visualize data and share them with your company.

An open source solution to explore and visualize data and supports lots of databases.

Enterprise-ready business intelligence web application.

Open source convention-over-configuration product for the whole data lifecycle, all the way from loading data to analyzing it.

All of them have been abandoned and constitute prior art. This means there is nothing to prevent anybody to implement or commercialize these concepts.

“You need a slightly more sophisticated developer team to cobble together a billing platform” (source) - Google's Product Director take on building billing systems: you need a certain kind of engineers to tackle that domain. It is not for everyone.

The source of all new features added to the billing perimeter.

A feature matrix of the different scheme and their average dicounts.

Latest changes of GCP billing features.

From the unofficial Google Cloud Platform newsletter.

A wrap-up of recent GCP pricing features.

GCP's billing subsidiary.

All the latest billing updates on DO.

AWS sacrificed the simplicity of subscription pricing for S3 and went instead with a safe cost-following strategy because they had no clue how the users will use their service.

Invoicing of random fees with coded amounts were used as a hidden signal to bypass legal obligations. Maybe that's why you need a flexible billing system after all.

The IBM 1401 mainframe (1959) optionally supported pounds/shillings/pence (£sd) currency arithmetics in hardware.

Starting at page #445, the "10.5 Billing Equipment and Systems" section describe the history and technical evolutions of Bell's phone calls metering and pricing.

Accounting paintings were a significant genre in Dutch art.

A way to optimize pricing using a physical model made in plaster of Paris, from 1914.

A site mockup of an tax avoidance product for small businesses.