The budget, and why it matters
Two hundred milliseconds at the server is the budget that keeps the total search interaction under a second on a normal connection. Below that, the buyer experiences the storefront as responsive. Above it, the experience degrades in ways that show up as bounce rate on a category, not as a complaint about search.
The budget is not aspirational. It is achievable on Adobe Commerce with the catalogue sizes most B2B businesses operate. The reason most builds miss it is not the platform. It is the layered defaults that compound until the storefront is paying for decisions nobody made on purpose.
The choices that matter, in order
Indexing strategy first. The default index footprint includes attributes that the storefront does not use and excludes attributes the buyer searches by. A lean index, scoped to the attributes the storefront actually queries, is the largest single win and it costs nothing to run.
Search engine choice second. Live Search is the path of least resistance and is adequate for most catalogues. For complex B2B with heavy attribute faceting, a dedicated engine, OpenSearch or Algolia, returns the budget at the cost of an integration to maintain. Pick on the basis of catalogue complexity, not on the basis of brand familiarity.
Front-end query shape third. The storefront should ask for the minimum set of fields it needs to render the results page. Most implementations ask for the full product record and discard ninety per cent of it. Trimming the query is dull work and it moves the number.
Caching last. Cache the things that do not change per buyer, do not cache the things that do. Most teams reverse this order under deadline pressure and ship a storefront that is fast for guests and slow for logged-in trade buyers, which is precisely the wrong way round for a B2B site.
"Search is fast enough when the engineer with a part number and the buyer with a category get the same sub-second answer across the full catalogue."
The trade-offs nobody volunteers
A lean index is harder to change. Adding a new search attribute is a reindex and a deployment, not a configuration change. The team has to accept the discipline. Most do, once the alternative is the storefront they have.
A dedicated search engine is a second system to monitor. The operational cost is real and it falls on the team that did not pick it. Decide who owns it before the contract is signed, not after the first incident.
A per-buyer cache strategy needs a cache key that includes the contract pricing context. Get the key wrong and you serve the wrong price to the wrong customer. That is a finance incident, not a performance one. Test it before launch and again after every pricing change.
What good looks like, six months in
The number to report is the ninety-fifth percentile server response time for search queries from authenticated trade buyers, measured at the application boundary. Under two hundred milliseconds, the site is doing its job. The number is easy to monitor and hard to argue with, which is why it is the right number to put on the dashboard the steering committee sees.
Hitting it is not a hero project. It is the result of four ordinary decisions, made in the right order, by a team that was given the time to make them properly.
