We automated client deck creation for a 200+ person sales team - here's the exact stack we built

Spent the last 2 months helping a B2B enterprise automate their client deck workflow. Reps were spending 3-4 hours per deck pulling info from CRM + Notion + call recordings, then formatting in Powerpoint. With 200+ reps making 5-8 decks a week, the math was insane. Most AI for sales decks posts stop at "use ChatGPT or Gamma" which is nowhere close to what enterprise teams actually need. The goal was never "make AI build decks." It was make AI build the RIGHT deck for THIS client without the rep doing manual work. The stack: Data source - CRM (They currently use Salesforce, which was their existing stack - no big changes there) * Account data, deal stage, industry, stakeholders, pain points from discovery * Reps already maintain this, no extra work * Added a "deck trigger" field - rep marks it when a deck is needed Claude * Pulls account data from CRM via API * Maps it to a fixed content structure we built (problem framing, solution fit, ROI math, case study selection, pricing framing) * This is the part most people skip - without a fixed structure, Claude outputs are inconsistent across reps * Also handles tone-matching by industry (different profiles for financial services vs SaaS vs healthcare) Alai * Connected via API * Has our full design system pre-loaded (brand colours, fonts, layouts, approved iconography, tone of voice and even specific brand -approved templates it needs to pull from) * Uses memory to pull from approved decks - "about us", "leadership", "customer logos", "case studies" come from a vetted pool instead of getting regenerated badly every time What the rep actually does now: marks the deck trigger in CRM, gets a fully branded deck in \~8-10 mins, tweaks 1-2 slides if needed, sends. We went from 3-4 hours → \~15 mins of human time. The honest stuff: * CRM hygiene needs to be perfect here, notes need to be filled, data points like industry etc need to be updated precisely for content accuracy - we spent a week getting AEs to fully understand the importance of this * Tried Gamma & Beautiful AI initially for the design layer. Brand consistency was very basic - the output was not approved by the brand team, plus no memory feature meant repetitive slides kept being regenerated differently. (We are planning on implementing Gamma for their CX team's onboarding docs though.) * Setting the content structure in Claude is non-negotiable imo. Without it no two reps get similar quality. We are now working on pre-enriching crm fields as much as possible + automating meeting notes to CRM notes so that AEs can just review the update and don't need to spend too much time just maintaining CRM hygiene. Would love any suggestions on how to optimise further or happy to ans any questions around the stack choice, what we tested, etc