How we cut deal presentation time from 30 minutes to under 60 seconds

Building a deal presentation by hand used to take 15 to 30 minutes per deal, and that time came out of someone's day every single time an opportunity moved forward. We built a pipeline that generates the same deck in under 60 seconds, for roughly $0.02 in model cost, triggered automatically the moment a deal's stage changes in the client's CRM. Nobody has to remember to build it, and nobody has to build it by hand.

What was actually broken before we automated it

The problem was never the presentation itself, it was who had to build it and how often that went wrong. One analyst carried the manual work of formatting slides for every deal that advanced, which made deck quality entirely dependent on whether that person was available, careful, and caught up on the latest numbers. Decks drifted out of the current template between updates, figures went stale between the moment someone pulled them and the moment the deck went out, and the whole process sat as a bottleneck between a deal moving forward and a deck actually being ready to send.

How does the pipeline turn a deal stage change into a finished deck?

The pipeline starts the instant a deal's stage changes inside the client's HubSpot, with no manual trigger required. An n8n workflow, built and hosted for the client and wired directly into the CRM they already ran, catches that stage change, pulls the relevant deal fields, and hands them to GPT-4o to draft the copy for each slide. That copy gets mapped into a version-tracked Google Slides master template through the Slides API, the finished deck is written to the client's Drive, and a link posts straight to Slack. A person never opens a blank slide.

• >Deal stage change fires the workflow instantly, no manual trigger
• >Deal data pulled directly from CRM fields, never retyped
• >GPT-4o drafts the slide copy from that structured data
• >Google Slides API populates a version-tracked master template
• >Finished deck lands in Drive and posts to Slack automatically

Why GPT-4o handles the copy, not the layout

GPT-4o's job in this pipeline is narrow on purpose: turn structured deal data into clean, consistent slide copy, not design a deck from a blank page. Keeping the model's job narrow is exactly what makes the output reliable. A model asked to freehand an entire presentation's structure will drift in tone and formatting from deck to deck. A model asked to fill fixed fields in a locked template, guided by a tightly structured prompt built around the deal's real data, produces the same quality and format whether it's generating the first deck of the day or the fiftieth.

The master file architecture that keeps every deck on-brand

Every generated deck starts from a single version-tracked master file, not a copy someone edited last quarter. That single source of truth is what stops decks from drifting out of brand over time. Layout, fonts, and slide order live in the master, and the workflow only ever swaps in the deal-specific copy and figures. When the template itself needs an update (new branding, a new slide, a revised disclosure), it changes once in the master, and every deck generated after that point picks it up automatically, with nobody left to update a dozen saved copies by hand.

What it costs, and how fast it runs

Is this worth building for your team?

This kind of pipeline pays for itself fast if your team builds the same style of document for every deal, client, or prospect and only the underlying data changes. If your presentations are genuinely bespoke every time, built around a one-off strategy for a specific opportunity, automation is the wrong fit for that document. But most recurring deal presentations, proposal decks, and status updates are more templated than teams admit, once you separate the data from the format.

mkdir builds this kind of pipeline inside the CRM and slide tools you already run, not as a separate system you have to adopt on top of what you have.