Job at a glance
The role A working product and a working deployment are two different things. You are the person who closes that gap. You sit inside the client's head office. You get the deployment live, you get their teams using the dashboards, and you own whether the AI is returning something worth acting on. Every client is different: different languages on the floor, different store noise, different vocabulary for the same product, different CRM, different idea of what a good conversation looks like.
The core platform does not change for each of them. You are the layer that makes it fit, and you are the one the client meets. You are encouraged to spend time in stores. The engineers who do the best work here are the ones who have stood on a shop floor and watched where the pitch and the pipeline actually break. Nobody will make you go. You will also spend real time in the codebase, because you fix what you find rather than filing it.
What you build has a commercial edge to it. A pilot converts when the client sees the result they were promised, and an account grows when a second team inside it sees what is already sitting in their data. Both of those outcomes are yours to deliver, not somebody else's to chase. How you will work You design the deployment, you build it, and you own whether it holds up on a Saturday evening in a crowded store.
Nobody hands you the plan, and nobody hands you the spec. You write both. The decisions are yours. Which integration is worth the week, which vertical taxonomy needs building, what ships in the pilot and what waits, and when to tell a client that the thing they are asking for is the wrong thing to build. You go and find out what a client needs before anyone writes a line of code. You will not be doing it alone.
There are founders, AI engineers and product people around you, and they will build alongside you. What nobody will do is tell you what the client needs. That call is yours. The work compounds if you do it well. What you learn on one deployment becomes a specification, then code, then a pattern the next one starts from. A year in, the deployments you designed should be running without you, and a new client should take a fraction of the time the first one did.
What you will do Own the deployment end to end. Device provisioning, store connectivity, data flowing, first insight in front of the client. Get from kickoff to something real inside the pilot window, and know by the halfway mark whether it is in trouble. Get their HQ using it. A dashboard nobody opens is a failed deployment. Sit with the sales, marketing and L&D teams, show them what is in their own data, and make sure the people who asked for this are actually looking at it every week.
Make the AI work on their floor. Their languages, their store noise, their product vocabulary. Benchmark transcription and speaker separation on their actual audio, and fix what fails instead of explaining it away. Build the vertical. Intent taxonomies, objection maps and prompt libraries for the category you are deployed into. A jewellery floor and an electronics floor do not share a conversation model.
Wire it into their systems. CRM and POS integrations, so conversation data connects to what actually got sold and the insight can be checked against reality. Build what the client asks for. Custom reports, dashboards and agents. Ground everything in source conversations and verify it before it ships, because a confident wrong number costs an account. Close the pilot. A pilot converts on results, not on effort.
Know what the client agreed to judge this on, work backwards from it, and make sure the output in front of their leadership at the end is the thing they asked for. Grow the account. The same intelligence is worth something to marketing, L&D and category teams inside the same client. Spot which of them would benefit, show them what is already in their data, and hand a real opening to the account team.
Push it back into the product. Turn one-off client