Personalized recommendations in B2B email don’t work the same way as B2C. You can’t run a Netflix-style collaborative filter on a list of 4,000 industrial engineers. The data is too sparse, the buying cycle is too long, and a wrong recommendation damages credibility instead of just lowering CTR.
The approach that worked for me with an IIoT startup selling predictive maintenance to manufacturers: anchor every recommendation in two things only. The recipient’s industry vertical (food and beverage, automotive, pulp and paper, pharma) and the type of asset they had shown interest in (rotating equipment, conveyors, HVAC, electrical). Nothing else. No engagement scores, no AI scoring, no lookalike modelling.
We built a small content library mapped to that 4-by-4 grid. Sixteen combinations, each with a tailored case study, a benchmark figure, and a relevant pilot result. A maintenance manager at a pasta factory who had downloaded a paper on motor vibration got the food and beverage / rotating equipment cell. A reliability engineer at an auto plant who had attended a webinar on conveyor failures got the automotive / conveyor cell.
The data came from 3 places: form fills (industry plus role), tracked content downloads, and webinar attendance tagged by topic. No CRM enrichment vendor, no machine learning, no behavioural model. Just two questions answered well, mapped to a content matrix.
The result was a click-through rate of 6.8% on segmented sends versus 2.1% on the generic newsletter. More importantly, reply rate tripled.
The lesson: in B2B, personalisation isn’t an algorithm problem. It’s a content problem. Two pieces of context, used honestly, beats a sophisticated model running on the wrong inputs.
