A B2B SaaS company serving the logistics sector
A Toronto-based B2B SaaS company in the logistics-tech space, around 25 people, post-Series A and approaching their Series B raise. The product is a configuration and compliance platform used by mid-market freight carriers, heavily document-driven, with each customer maintaining hundreds to thousands of regulatory, contractual, and operational documents in the platform's library. Customer feedback was consistent: finding the right document was the worst part of using the product.
Off-the-shelf full-text search was already in the product and had been incrementally improved for years. The team had reached the limits of what keyword matching could do for a corpus where the same regulation might be referenced as “HOS rules,” “hours-of-service,” or “Part 395” in three different documents. The CTO believed semantic search was the right answer technically, but the engineering team had no production experience with vector databases or embeddings pipelines, and there was no budget to hire a permanent AI specialist before the Series B closed.
We started with a six-week Proof of Concept against an anonymised slice of three representative customers' corpora, the high-volume customer, the most-complex customer, and the most-document-heavy customer. The PoC validated three things: that semantic search materially outperformed keyword on this corpus, that the indexing economics worked at the customer-count scale they expected to grow into, and that the rollout did not need to touch the existing search infrastructure on day one. We then took the engagement into a four-month AI Development & Implementation phase, building production-grade indexing, the customer-facing UI, monitoring, and the safety rails the engineering team needed to operate the feature without us after delivery.
Semantic search shipped to all customers fourteen weeks after the PoC began. Sixty percent of active customer accounts used it at least once in the first sixty days. The feature was specifically cited in the company's Series B materials. The engineering team has owned and extended the feature with no further engagement from us, including a customer-specific embedding-model tweak that we had explicitly designed the system to support. The original keyword search remains in the product as a fallback, used by roughly 8% of queries; we recommended against retiring it.
Published February 24, 2026. Sector and figures are generalised where necessary to keep the client comfortable being represented here. The structure of what we did and what changed is exactly as it ran.
Tell us the situation on a free first call. We will say which engagement family fits and what it would cost. If none of our offers fit, we will tell you that too.
Schedule a free consultation