Case studies / Applied AI
Applied AI · Industrial distribution
The AI recommendation engine behind $6.5M in incremental margin for a global parts distributor
A leading, PE-owned distributor of factory-automation parts was running procurement off “shopping lists” built almost entirely from historic sales volume — missing a huge range of live market opportunity, and unable to systematically use supply channels it didn’t fully trust. We designed, built, and operate the full data-and-AI platform behind their procurement recommendations engine — integrating external demand signals, live supply from multiple buying channels, and three purpose-built analytics algorithms into one daily-ranked deal list. In its first six months, the program generated $6.5M in incremental gross margin and a 16X return on program fees.
Outcomes
Generated in the first six months of the program, from roughly 15,000 high-ROI deals sourced through the recommendation engine and resold.
Measured by the client against its ongoing monthly investment in the platform.
Live across the US and Europe, re-ranking every available deal, every day, for buyers in every location.
Figures are the client’s own reported results, as published in Insight Factory’s client-facing case study for the program.
The challenge
A shopping list that only looked backward
The client’s procurement process was built almost entirely on one signal: what had sold before. That meant its buyers were blind to attractive, in-demand parts with no sales history to point to yet — exactly the kind of opportunity a fast-moving, eCommerce-heavy parts business can’t afford to miss. At the same time, the client had real supply sitting on the table it wasn’t using well: preferred-vendor catalogs, and a large daily stream of Excel “bid sheets” from businesses liquidating inventory — channels the client had previously underused because it had no systematic way to judge which sellers and listings were actually trustworthy.
What made this hard was the sheer range of signal that had to come together into one number a buyer could act on: external demand signals that had nothing to do with the client’s own sales history, live supply from several structurally different channels (an eCommerce marketplace, vendor catalogs, ad hoc bid sheets), and a trust layer that had to separate legitimate deals from listings that would waste a buyer’s time or money — all of it re-ranked daily, across two continents, in a way the client’s own team could act on immediately.
Before
- Shopping lists built almost entirely from historic sales volume and trends
- High-potential parts with no sales history invisible to procurement
- Valuable supply channels (vendor catalogs, bid sheets) underused over trust concerns
- Sourcing-channel performance and profitability compiled ad hoc, by hand
After
- Demand signals beyond historic sales feeding directly into what buyers see
- Three purpose-built analytics algorithms surfacing and ranking deals no historic-sales list would have found
- Every deal, across every channel, re-ranked by economic attractiveness — daily
- Suspicious sellers and likely-defective listings screened out automatically before a buyer ever sees them
- Every bid and purchase tagged for closed-loop reporting on results and ROI, by channel
How we approached it
One daily-ranked list built from many signals
One data repository, many demand signals
We integrated an extensive range of external demand signals — well beyond the client’s own historic sales — into a new cloud data repository, alongside a live stream of deal signals from every buying channel: the eCommerce marketplace, preferred-vendor catalogs, and daily bid sheets, each brought in via newly built APIs.
Three algorithms, one ranked list
One algorithm identifies attractive, in-demand products the client’s historic-sales-based lists were missing entirely. A second identifies high-ROI deals across every sourcing channel. A third ranks the economic attractiveness of every available deal, every day — so buyers across the US and Europe can prioritise their purchasing activity against one consistent, current signal instead of channel-by-channel judgment calls.
Trust and condition, screened by AI before a buyer ever looks
A generative-AI layer screens listings for signs of an unreliable seller and for likely product defects from listing images — systematically excluding them from the recommendations so buyer time and capital go only toward legitimate deals, not toward the wasted cost of bad ones.
Built as infrastructure, not a one-off list
This runs as a live, weekly-cadence managed service, not a static tool: every recommended deal is tagged with a bid ID at the point of purchase, closing the loop so results and ROI can be reported accurately by channel, and every rejected recommendation feeds back in so the engine keeps improving. It scaled from an initial pilot into an always-on program running across both the client’s US and European buying operations.
Under the hood
Next step
Sitting on multiple supply or demand signals that never get combined into one decision?
We build production AI platforms that turn scattered signal — market data, internal history, live external supply — into one ranked, trustworthy recommendation your team can act on daily, at scale.