Product
Adoption Is Decided in the Crowded Day
A demo wins the decision, the contract is signed, and six months later the tool is no longer used. This sequence says where adoption is really decided: not in the demo, in daily friction.
In AI tool selection committees, there is almost always a moment when a demonstration wins the decision. The salesperson shows, the audience is impressed, the contract is signed in the following weeks. Six months later, the tool is used only residually. This sequence is so widespread that it has become a structural fact of the market, and it says something important about what really determines adoption.
The most striking thing, in this sequence, is not that it happens, but that it recurs identically, again and again, without organizations seeming to draw the lesson. The same scenario replays, because the moment of decision and the moment of use are separated by months and entrusted to different people. One chooses on a promise, one lives with a reality, and the gap between the two appears only once it is too late to correct it. To understand this gap is to understand almost everything that decides a deployment’s fate.
Adoption is not won in the demonstration. It is won in daily friction.
Why the demo predicts nothing
A demonstration shows the best cases, on the best data, with the best operator, in the best context. It is useful for understanding a tool’s maximal potential, and it is insufficient for predicting its real use. Because real use never unfolds in demonstration conditions. It unfolds in a crowded day, on a sensitive matter, under the pressure of a demanding client, with imperfect data, at a moment when the user has fifteen other things in mind. In these conditions, what matters is not the breadth of the tool’s capabilities, but the usage cost it imposes at each interaction.
This usage cost is made of a multitude of micro-frictions. The time to open the tool. The time to give it the context. The time to formulate the request. The time to evaluate the answer. The time to bring it back into the workflow. The time to validate it, archive it, trace it. Each friction taken alone is minimal; cumulated, they determine long-term adoption far more surely than the absolute quality of the output produced. The demonstration erases all these frictions, because it is built to erase them; the everyday makes them all reappear, because it is built for nothing.
There is therefore a fundamental asymmetry between what the demonstration measures and what use demands. The demonstration measures maximal capability in ideal conditions; use demands minimal cost in hostile conditions. These are two things not only different but largely independent: a tool can excel at the first and fail at the second, and that is even the most frequent case, because what makes a demonstration spectacular, the richness, the visibility, the breadth of capabilities, is precisely what weighs down daily use.
The autopsy of a deployment that fades out
One can describe the typical trajectory of a failed deployment, so much does it repeat. The first weeks, the enthusiasm of the choice carries usage: one saw the demonstration, one wants to find the magic again, one makes the effort to go to the tool. This initial period is misleading, for it is carried by a motivation that will not last, and the usage figures it produces reflect not cruising use but the impetus of the start.
Then the reality of the everyday reasserts itself. Each interaction demands a little more effort than the old reflex, and under pressure, the old reflex wins. The lawyer with a deadline within the hour does not go to the tool that requires a detour; they do as before, because as before is faster at that precise instant. Usage does not collapse at once, it crumbles, interaction after interaction, until the tool is mobilized only for cases where one has time, that is, almost never.
The decisive point of this autopsy is the following: no one decided to abandon the tool. There was no meeting, no verdict, no moment when someone concluded the tool was worthless. It simply lost, each time, against the path of least resistance, in a series of individual micro-arbitrations no dashboard captures. It is a death without a death, an abandonment without declared abandonment, and that is precisely what makes it so hard to anticipate: it happens nowhere in particular, it happens everywhere a little.
No deployment dies of a decision. It dies of a thousand times the old reflex won.
This trajectory also explains why the usual remedies fail. Faced with crumbling usage, one generally reacts with training, relaunches, internal communication, as if the problem were that people had not understood the tool’s value. But they understood it perfectly; what they cannot do is pay the usage cost every day, under pressure. No training corrects a structural friction, because the friction owes not to a defect of understanding, but to the design of the tool. People are trained to want the tool, when the problem is that it costs them too much to use it.
The demo tests maximal capability. The crowded day tests usage cost. It is the second that decides.
The buyer is not the user
At the heart of this gap lies a dissociation few organizations face: the one who chooses the tool is almost never the one who endures it daily. The decision is made in committee, by managers who judge on a demonstration and a promise; the use is lived in the offices, by practitioners who were not in the room on the day of the choice. Between the two, the information does not flow up, or flows up too late, in too diffuse a form to overturn an already signed contract.
This dissociation distorts the whole selection process. The committee is sensitive to what gets demonstrated, because that is what it sees; it is blind to usage cost, because it will never experience it itself. It therefore optimizes, without malice, exactly the wrong criterion: it chooses the tool that makes the best impression in the room, not the one that will hurt least in the day. And since the end user has no say at the moment that matters, nothing corrects this bias before deployment, where it is too late.
Designing a tool that lasts therefore supposes designing for the user, not the buyer, even though it is the buyer who decides. It is a hard choice, because it amounts to optimizing for someone not in the decision room, at the expense of what would win that room. But it is the only choice coherent with the reality of adoption: a tool is not adopted by those who sign it, it is adopted, or rejected, by those who use it. Serving the second, even when it is the first to convince, is the bet that distinguishes a product designed to be sold from a product designed to be used.
Designing against friction
This reading of the market is one of the reasons MAX’s product philosophy concentrates on the systematic reduction of frictions rather than on the amplification of demonstrative capabilities. This translates into choices barely visible at first sight but determining in use: acting in existing tools, following the context without asking the user to redefine it, intervening at the right moment rather than on demand, keeping the trace without having it produced, remembering without asking to remember. Each of these choices removes a friction from the everyday, and the cumulative effect is a layer that becomes usable without sustained effort.
One must see that reducing friction is not glamorous work, and that is why it is so often neglected. Saving three seconds on an operation repeated a thousand times a day does not make a fine slide; it does not get told in a meeting, it does not get applauded in a demonstration. But it is exactly that work, accumulated over dozens of micro-frictions, that decides whether an entire organization ends up adopting or abandoning. Glamour is on the side of capabilities; adoption is on the side of friction. And one must choose which to optimize, because one cannot optimize both at once.
This philosophy has a price, to be accepted knowingly. It makes demonstrations less spectacular and initial evaluation less legible, because a demonstration of MAX does not reveal all its interest in five minutes: it is at the hundredth use that it becomes obvious why it works where others ran out of breath. It is a bet on real adoption rather than immediate buy-in, and that bet is uncomfortable, because it consists in giving up winning the committee easily to win duration with difficulty.
Why time decides in favor of low friction
In the long run, this logic always ends up prevailing, because friction is a recurring cost and recurring costs always end up being felt. A spectacular but use-costly advantage erodes as one uses it; a discreet but frictionless advantage strengthens as one gets used to it. Time works against demonstrations and for integration, because it accumulates repetitions, and it is in repetition that friction is paid.
It is a matter of arithmetic as much as of philosophy. A usage cost, even minimal, paid at every interaction, ends up exceeding any one-off gain, provided you multiply by the number of interactions. Conversely, an avoided friction is a gain that repeats endlessly, without one having to think about it. Over a horizon of two or three years, it is therefore almost never the tools that best seduced the committee that remain, but those that made themselves forgotten in the flow, because the arithmetic of repetition ends up outweighing the first day’s impression.
The market will not keep the most impressive demos. It will keep the systems whose friction stays lowest over time.
It is the only bet that pays at scale, and it is in this direction that the legal AI that will remain is built. The others will seduce in committee and disappear within six months, like so many tools before them, not because they were bad, but because they had optimized the wrong moment: that of the decision, and not that of use.
