Since the launch of Apple Silicon in 2020, multiple products and processors had been added to the Mac lineup. With many options stacking together, eventually a significant portion of customers was getting confused about what was the best Mac for their needs and the number of wrong-fit purchases increased, spiking returns and model exchanges.
This pattern, observed by the Apple Retail team, was occurring more specifically with online or unassisted in-store purchases. Customers that shopped with Apple Specialists were not affected, as the conversation did the translation, and customers would be satisfied with their configuration.
While Apple.com already had tools to help customers compare the existing models and their performance, that feature only compared models which a customer was already interested and often used less accessible language.
A different problem was happening with the Shop With a Specialist option, as it would often be discreetly shown on the website and was much harder to scale globally.
Bring the Specialist conversation approach to scale, both online and at stores, to help customers find the best Mac model for their needs and reduce the number of returns and model exchanges.
How can we get a customer buying the right Mac configuration for their needs on the first try, without a Specialist to advise them, and without steering average purchase values down?
The work ran in seven steps, though in practice they overlapped and doubled back on each other. We started with the Mac lineup itself and with the Specialists who sell it, aiming at rebuilding a conversation that already worked in a store. Then we picked a format, agreed how we would know if it worked, and only after that designed the flow and wrote the questions. Nothing shipped until two rounds of testing had answered both of the questions we had set ourselves.
We started by understanding what was actually on sale. How many processors, how many tiers, how many generations, and what the real performance differences between them were. Then every Mac option, and the kind of work each one fits best. There was already plenty documentation on all of this.
Then we talked to people. We interviewed Specialists and a few other stakeholders to understand how they turn what a customer tells them into a specific configuration, and how much of that we could map to a digital experience.
We looked at a few different things we could build, and decided a questionnaire was the right move for what we were trying to accomplish, while creating backlog tasks for future enhancements on other places.
Before designing anything, we agreed on what we would measure. The tool had to be better at figuring out the best spec for a customer, and it also had to be better at the value of what they bought. Both, or it was not worth building it.
Then the flow itself. How it should go, what order the questions worked best in. What to filter and what to weight, and how to make it feel personal the way the Specialists did it rather than like filling in a form.
Finalized the copy for the question set with a UX writer. Specifically steering users in a broad but familiar way, making results feel a safe bet while nudging towards better purchases.
We presented the solution and it was approved internally, but it still had to be proved. The first round was a batch of user tests with real customers. A Specialist would interview a customer and answer the questions on their own iPhone as they talked, and at the end we compared what the tool returned against what the Specialist thought was best for that person.
That told us it picked well. The second round had to prove it would not lower what people spent, which was the harder one. We ran a series of internal tests with Apple employees, comparing a control group against a group using the tool, so we could see what it did to sales before any customer ever saw it.
When a customer had to decide between the models it was faced with too many options.
The goal was to avoid returns rather than absorb them, which ruled out everything after the purchase: exchange programmes, order-modification windows, policy changes. Whatever we built had to act before or at the decision, and that is a narrow field.
One gate question opens the flow and decides how long it is. Pick Essentials, Work, Education or Creative hobbies, in any combination, and each one you pick adds its own follow-up. Three questions close it for everybody: where the Mac will live, what will be plugged into it, and budget, which is asked last so it cannot colour anything above it. At the end, a shortlist of specific configurations.
Four decisions carry most of the work. None of them are visible as features, and all of them are why the shortlist at the end is worth anything.
Ask About Life, Answer in Specifications No question mentions memory, storage or a chip tier. Every one asks about work, place or what gets plugged in, and each answer feeds several of nineteen hidden attributes. People are bad at judging what machine they need and good at describing what they do, so we only ever asked the second thing.
Constraints Filter, Preferences Weight Where the Mac will live is the only hard filter in the model. Answer anything other than a fixed place and every desktop leaves the pool before scoring starts. Everything else is soft, so a Pro machine the answers do not call for is pushed down rather than hidden, and can still surface when the rest of the answers earn it.
Ceilings for Capability, Averages for Taste Performance, memory, storage and graphics resolve to the most demanding answer somebody gives. Edit video once and you need a machine that edits video. Portability, screen size, battery and sound resolve to a blend instead. Capability is a floor you have to clear, comfort is something you can compromise on.
A Range, Not a Verdict Results come back in four slots: a top choice, an alternative, a laptop companion when both leaders are desktops, and one machine deliberately above the budget given. If nothing in the set is over budget, the model goes and finds one. That slot is what answers the objection that nearly stopped the project.
What the store was missing was never a page. It was the conversation a Specialist has, and the part of that conversation that turns a description of somebody's work into a configuration. Putting that step in every channel did not cost anything on the way in, which was the condition the whole project was set: customers who used it bought better matched Macs, and spent more doing it.
In a store, a bad decision does not stay a bad experience. A customer who buys the wrong Mac returns it or exchanges it, and that costs the restock, a Specialist's time and a machine that cannot be sold as new again. The project was argued for in those terms rather than in usability ones, and it was measured the same way at the end.
A customer answers eight questions at most, and none of them mention a chip or an amount of memory. Getting there took nineteen attributes, a rule about which answers set a floor and which get averaged, and a long argument about which single question was allowed to rule anything out. Every question taken out of the set made the model behind it harder, because what it used to ask still had to be worked out from the answers left.
Too much of this project went on looking for something nobody had built before, when the answer that worked turned out to be the one already running in every store: ask about the work, rule out what the place makes impossible, come back with a shortlist. The parts that survived contact with customers were the ordinary ones. Less appetite for the novel option and more time spent proving the safe one would have reached the same flow sooner, and with fewer of the objections that nearly stopped it.