"Sticky" is used for three different things. Two of them are symptoms and one is an asset. Confusing them is how teams congratulate themselves on engagement while their moat stays empty.
| What it is | What it means | What it is worth |
|---|---|---|
| Frequency | They open it often | A leading indicator. Nothing more. Frequency can vanish in a week. |
| Habit | They open it without deciding to | A mechanism. It produces the behaviour that builds the asset. |
| Switching cost | Leaving would cost them something real | The asset. This is the only one a competitor cannot simply out-market. |
The order matters: frequency is evidence that a habit may be forming, a habit is what causes the user to keep depositing value into the product, and the deposits are what make leaving expensive. A strategy that stops at "daily usage" has described the first link of a three-link chain and called it a moat.
Nir Eyal's Hook model has four stages: trigger, action, variable reward, investment.1 For a professional tool, two of these behave very differently than they do in the consumer products the model was built from.
Not all lock-in is equal, and it is worth naming which kind you are actually building.
| Kind | What holds them | How fragile |
|---|---|---|
| Data | Accumulated content they would lose or have to move | Depends entirely on whether it can be re-derived elsewhere. See below. |
| Workflow | Muscle memory, a routine built around the tool | Strong while the routine holds, gone after a two-week interruption. |
| Integration | Connected systems, configured plumbing | Real, and it is the reason connection requests are worth their friction. |
| Social | Colleagues and counterparties on the same tool | The strongest kind, and mostly unavailable to a solo-professional tool. |
| Contractual | An annual commitment | Not a moat. A delay. |
"The longer you use it, the smarter it gets" appears in almost every AI product pitch, and it is usually a claim about personal accumulation: your data makes your instance better for you. That is a switching cost, and a real one. It is not a network effect. A network effect means other users make the product better for you, which compounds across the market and gets harder to catch as it grows. Personal accumulation does not compound across the market at all - it starts from zero for every single new user, including every user your competitor signs up tomorrow.
NFX, who wrote the standard taxonomy of these things, make the same warning from the other direction: data network effects are commonly overstated and routinely confused with plain switching cost from accumulated data.2 Getting this right changes what you build. If your defence is personal accumulation, then your defence is time: the number of months of history a user has, multiplied by how hard that history is to reproduce elsewhere.
Ask of any accumulated data: could a competitor re-derive it in an afternoon? Split the asset in two.
| Re-derivable | Not re-derivable |
|---|---|
| Anything extracted from a source the user still controls and can reconnect elsewhere: inbox, calendar, contacts, Drive, the CRM | Anything captured live and stored nowhere else: recorded meetings, voice notes, the corrections the user made to your summaries |
| A rival who connects to the same sources rebuilds it on day one. Impressive at demo time, worth nothing as defence. | Gone forever if the user leaves. This is the only part that is actually a moat. |
This is not an argument against reading the inbox and calendar - that connection is what makes the product useful in week one, and time-to-value matters enormously. It is an argument about which half you are counting as the moat. A product whose memory is 90 percent re-derived enrichment and 10 percent captured conversation has a far shallower defence than its own slide deck believes.
Two clocks run from signup, and a product needs both to be short.
A product can have excellent frequency, a genuine habit, and a time-to-lock of zero. It will look healthy in every dashboard right up until a competitor with better marketing arrives.
Read NFX's The Network Effects Manual, in particular the data network effects entry and their caution about it. If you only read one section, read that caution - it is the antidote to the most common claim in AI pitch decks, including this one.
Want to work out the actual time-to-lock for a specific product design? Bring the design to your teaching agent and work it through together.