The reverse ETL idea was one of the better ones to come out of the modern data stack era. Companies spent a decade consolidating customer data into a warehouse, at considerable expense, and then discovered that the data was trapped: the warehouse could produce a beautiful dashboard showing which customers were about to churn, and getting that list into the email tool required either a manual CSV export or a custom integration nobody wanted to maintain. Reverse ETL closed the loop by syncing warehouse data out to operational tools, and Hightouch built a good product on that observation. The documentation explains this well.
The warehouse-native argument is the strongest part and it is a genuine architectural advantage rather than positioning. Traditional customer data platforms ingest a copy of your customer data into their own store, which means you now have two sources of truth, two sets of access controls, two places where a deletion request has to be honoured, and a definition of customer that can drift between them. Hightouch queries your warehouse directly and syncs the results outward, leaving the data where your governance already applies. For anyone dealing with GDPR deletion workflows or trying to maintain a single definition of a customer segment, that difference is substantial and the docs make the case clearly.
The destination catalogue documentation is thorough, with real per-destination detail about field mapping, sync behaviour and the quirks of each API, which is the kind of material that only gets written after supporting a lot of real syncs. The separation of paths for data teams, marketers and administrators reflects how the tool is actually used in an organisation, where the person defining the audience and the person owning the warehouse are usually different people with different vocabularies. What does not exist is any training. No academy, no courses, no certification, no learning path.
This is a product that sits between two functions, data and marketing, that famously struggle to communicate, and a structured course teaching each side enough about the other's concerns would be genuinely valuable. Instead there is documentation organised by feature, which serves the data engineer reasonably and the marketer less well. The agentic repositioning is the newer development and deserves a measured read. Hightouch now describes itself as an agentic composable CDP, with agents built in to help marketers explore data, build audiences, generate creative and analyse campaign performance.
I have no objection to any of those capabilities existing. The material supporting them is thinner than the material supporting the core product, which is what you would expect for something recent, and the gap between the positioning and the documented practice is wide enough to notice. My general view on agent-generated audiences is the same as my view on agent-generated pipelines: generation is the easy half and verification is the hard half, and an audience definition that looks right and quietly excludes a segment is a failure mode that will not announce itself. Treat it as a direction rather than a settled practice.
The omission I would most want addressed is audience sprawl. Making audience creation easy has an entirely predictable consequence, which is that a year later there are three hundred audiences, forty of them syncing to destinations nobody checks, several defined in mutually contradictory ways, and no process for retiring any of them. This is the exact analogue of dashboard sprawl in BI and feature flag debt in release management, and it is the structural blind spot I keep finding: a vendor cannot teach restraint about its own product, because its revenue grows with usage. Nothing here teaches you to audit, consolidate or retire audiences, and somebody should.
Sync economics get less attention than they should too. Destination APIs have rate limits, syncs consume warehouse compute, and a high-frequency sync of a large audience to a destination that throttles you is a cost and reliability problem that is easy to create and annoying to diagnose. The documentation covers the mechanics. It does not really teach you to think about the trade.
Three point four. Competent documentation of a genuinely good architectural idea, with no teaching layer at all, a repositioning that has run ahead of its material, and silence on the sprawl problem that the product's ease of use reliably produces.