Network Effects Examples: Lock-In or Plain Inertia

Switching costs are the full price of leaving: moving the data, retraining people, the quarter when something breaks. They become a moat only once a customer has added that price up and stayed anyway. Network effects are the other half of the same mistake: eBay has one, Netflix does not, and both get claimed on the same evidence, customers who have not left. But a locked-in customer and a bored one look identical until the first price increase, so the honest test is behaviour under pressure, not tenure.

What switching costs really cost#

It is everything the customer swallows to move elsewhere: migrating data, rewriting procedures, retraining a team, the risk of an outage at the worst moment. You see it most clearly where one system carries a company's daily work. An ERP system like SAP stops being software after a few years and becomes the nervous system of the operation: it tracks stock, issues invoices, tells people what to do next. This branch of the economic moat comes from what happens across the week that system stops working, not from the licence price.

Say the subscription costs $2 million a year and a rival offers the same scope for $1.4 million. Six hundred thousand looks like a decision that makes itself. Except that the migration is eighteen months of team effort, retraining four hundred people, and one quarter in which the warehouse might invoice late. Which means the IT director is weighing $600,000 of annual savings against the risk that invoices go out late for three months. On that arithmetic, a 5% increase goes through without a single phone call.

Consumer products rarely carry the same weight. Microsoft's office suite holds companies through integrations, file compatibility and departmental habit, so procurement does not even invite the competition to talk. Your music app holds you with playlists, which is one afternoon of work. A switching cost that fits into an afternoon is not a moat, it is a mild annoyance.

Hence the reading rule: switching costs are measured in months of the customer's work, not in the size of the invoice. A contract holds a customer like a mooring line, and the question is always the same one. Anchor chain, or a thread nobody has pulled yet?

Why high retention proves nothing#

The consensus here is reasonable. If customers have stayed for eight years, renewed their contracts and swallowed every annual increase, leaving must hurt. A slide with, say, 95% retention and net revenue retention above 100%, under a row of logos that have been with us forever. Most analysis stops there.

The cleanest proof came out of UK banking, where the regulator had portability built to order. The account switching service moves your account to another bank in seven working days and redirects every payment, free of charge and under a guarantee. Even so, the Competition and Markets Authority found that "at the moment only 3% of personal and 4% of business customers switch to a different bank in any year", and used that finding to open the market up to open banking. So 97 out of 100 customers stay put where leaving costs a week and no money at all.

The service works and gets used: in 2024 it handled 1.2 million switches, the second year running above a million. Changing banks is neither exotic nor expensive, and almost everybody stays anyway. Read UK bank retention alone and you would see a textbook moat. What you are looking at is inertia.

Inertia is not a flaw in the customer, it is the default state. Change needs a decision, a decision needs a reason, and usually nobody supplies one. The rival who does supply one walks off with a customer whose tenure looked like devotion.

There is one arrangement in which retention does settle the question. The company raised prices that year, customers stayed, and revenue from those same customers grew faster than the price list. That is no longer a story about nobody leaving, it is a decision taken with an alternative in plain sight. Most companies have two or three such years. The rest were quiet.

The three moments that tell the truth#

So how do you know a customer stayed because they could not leave, rather than could not be bothered? Tenure cannot answer that. Behaviour can. Switching costs are visible only in the moments when a customer had a reason to price the exit and stayed regardless. There are three of them, and each leaves a trace in the filings:

1. A price increase. Did the company raise prices and keep its customer count? The split of revenue growth into price and volume sits in management's commentary on the results, and it is the same pricing power test you would run on any company.
2. A budget cut. What did the customer do in a weak year: stay, drop to a cheaper tier, or renegotiate the rate? A recession is a free experiment nobody had to design.
3. A migration tool. Did anyone lower the cost of leaving, whether a rival with a data-transfer tool or a regulator with a portability mandate? If so, the clock started, and only the first year after that event tells you anything.

The answers rarely appear in the investor deck. They appear in the risk section, where the company writes in its own words that customers may leave and rivals may cut prices. That is how you read risk factors, and it is the most honest passage in the whole annual report. The second helping is in the Q&A part of the earnings call, when analysts push on rising discounts. A strong answer has a number in it. A weak one has a tone.

Work out, too, who actually pays the switching cost. Inside a company the person using the product, the manager approving the purchase and the department signing the transfer are three people with three different worries. For the user, switching means a new interface and three weeks of irritation. For the director, it means a failed migration carrying their signature, and that second worry holds far harder. Which is why tenders go less often to the best product than to the vendor nobody gets fired for choosing.

Network effects examples that actually hold#

Switching costs hold one customer at a time. A network effect holds all of them at once, because every new participant raises the value of the product for everyone else.

On eBay, buyers turn up for the selection and sellers list where the buyers are. At the end of 2024 the company reported 134 million active buyers and 2.3 billion live listings, which is roughly seventeen listings per buyer. Over that same year $75 billion of transactions went through the platform, which is about $560 of turnover per buyer. None of those numbers is a moat on its own. The moat is that a seller has nowhere else to go for that pool of buyers, and a buyer has nowhere else to go for that pool of listings. A network built over two decades cannot simply be switched on, however much money a rival brings.

Network effects come in two kinds, and telling them apart does most of the analytical work. A direct one runs inside one group: a messaging app is worth as much as the number of your friends already on it. An indirect one runs between two groups that need each other: you carry a payment card because shops accept it, and shops accept it because people carry it. The first kind is local and falls one friend group at a time, since friends can move together. The second has to be attacked from both sides at once, which is why it holds better.

One rule covers the rest. Every time you hear a network-effect claim, finish the sentence "a new X raises the value for Y". If you cannot finish it, you are looking at scale, not a network.

Why Netflix has no network effect#

Netflix looks like a network and is not one. A million new subscribers do nothing for your evening, because the catalogue is the same for everybody. A bigger base funds a bigger content budget, but that is scale: it helps the company, not the subscribers with each other.

The company's own reporting shows it. In the second quarter of 2022 Netflix lost 0.97 million paid memberships, ending the quarter at 220.67 million. Just under half a percent of the base walked out over three months and took nobody with them. In a real network, a departure lowers the value for those who stay. Here the ones who stayed had no way of noticing.

Scale is not a network.

I do not believe in a network effect that can only be described by a user count. The distinction takes five minutes and changes the whole thesis: switching costs defend price, network effects defend market share, and scale defends nothing but unit cost.

Three mechanisms that look identical in an investor deck:

Every moat has an upkeep bill#

Even a real moat does not maintain itself, and somebody pays for it every year. A brand needs marketing and product quality, a network needs trust and active participants, a technology edge needs uninterrupted research. Switching costs work exactly as long as the product is worth its own friction. A company that raises the price list and stops investing in the product has turned a moat into a countdown.

I do not know which year that bill stops adding up, and nobody outside the company does. I know where it leaves a mark: in selling costs growing faster than revenue, because a customer who used to stay out of habit eventually has to be bought back with a discount.

While you are there, check whether the advantage described in the report is current or historical. Some moats were built for a market that no longer exists: for a customer who could not move their own data, for technology replaced once a decade, for a regulation two amendments ago. The question is not whether the moat was built, but whether it still bites given today's habits, tools and rules.

How long would a funded rival need#

One question organises all of this: if a superbly funded competitor wanted this company's best customers, how long would it take? An answer of "a few months" means the moat is shallower than the story about it. An answer of "years, plus trust, integrations and a change in human habits" means a business worth studying.

That answer differs between the two mechanisms. Switching costs can be worked around one customer at a time, with the rival paying for each migration itself; with enough money and patience, three years buys half the market. A network effect cannot, because the first person you move arrives at an empty product, and an empty product is a worse one. So my bet is that network effects are the more durable of the two when they exist at all, while switching costs are far more common and far easier to mistake for nothing.

Each mechanism breaks in its own way, and it is better to name that up front. Switching costs break when somebody writes a migration tool or a regulator mandates portability, as happened to UK current accounts. Network effects break more rarely but more violently, because nobody leaves a network alone: either nobody goes, or the group moves at once.

A moat with no named failure mode has not been researched, only admired.

A real moat is easy to over-read in the euphoria of finding one. It tells you how hard the business is to attack, and nothing about valuation, execution or demand. So the result of this test belongs among the assumptions you try to break yourself, not on a list of reasons to stop checking.

Checking conditions like that after every set of results is what Monitoring is for, without your involvement and without Credits.

Where to write the answer down#

A conclusion kept in your head wears out within a year, because what survives is the verdict, not the condition. In Taufolio that place is the Investment thesis: it comes out of a Full report and splits your reason for holding a company into assumptions, each with a measurable condition. "Switching costs are holding" is not an assumption. A sentence like "prices went up 4% and no customers left" is one, because it can be settled at the next set of results.

Only a year later does the difference between those two sentences show up. The first confirms itself, since customers always exist somewhere and management always says something about them. A sentence with a number in it can fail to confirm, and that is the whole of its value. What that split looks like in practice is on the product page.

A moat nobody has tested under pressure is a mooring line about which you know one thing: the boat has not drifted off yet. The next price increase, or the first weak year your customers have, will show whether it is chain or thread. The date is already in the earnings calendar.