New Hire Onboarding System

The Digital Sales Coordinator role I'm in has a steep learning curve. New hires need to know dozens of ad products, multiple internal systems (order entry, trafficking, pacing), policy nuances, and the unwritten workflows that move a campaign from sold to live. Most of that knowledge used to live in shadowing, Slack pings to senior coordinators, and tribal memory.

I built a two-layer onboarding system to fix that:

• Layer 1: Training presentations walk new hires through every system, workflow, and responsibility, with visuals and quick-reference checklists.
• Layer 2: A custom AI assistant sits on top of a consolidated knowledge base, so new hires can ask any question and get an accurate, role-specific answer in seconds.

Presentations are the curriculum. The AI assistant is the always-on reference desk.

I designed and built three training decks using Gamma. Each one covers a distinct stage of the role, with the same opinionated structure: overview → step-by-step → common mistakes → quick reference checklist.

1. Role overview

What the job actually is, day-to-day. Covers the five core responsibilities, the weekly cadence, daily routine, team structure, who you talk to and when, what systems you use, how campaign monitoring works, and a two-week onboarding checklist so new hires know what they should be able to do by when.

2. WideOrbit order entry training

Deep walkthrough of the order entry system: order setup, line item configuration, rate types, billing rules, the submission workflow, and a quick-reference checklist for the moves that new hires get wrong most often.

3. Campaign trafficking training

The handoff from order entry to live campaign: monitoring work orders, requesting creative, submitting data collection, post-submission monitoring responsibilities, and how to set up analytics correctly the first time.

On top of the presentations, I built an always-on AI assistant using Google Gemini's Gem feature. The setup for a new hire is one file upload, and from that moment they have an assistant that knows every product, rule, and procedure in the role.

The knowledge base behind it

The Gem itself is just configuration. The real product is the 437-line markdown reference document I authored that powers it. I consolidated 6+ training documents, PDFs, and my own operational experience into a single structured file with 11 sections:

1. Role overview
2. End-to-end campaign workflow
3. Systems and tools
4. Product database: rates, minimums, specifications
5. Order entry rules
6. Campaign handoff routing
7. Creative specifications
8. Pacing monitoring
9. 15 common mistakes (with how to avoid each)
10. Key contacts
11. Glossary

I structured every section specifically for AI retrieval: short chunks, clear headers, explicit rules, so the Gem returns accurate, role-specific answers instead of generic LLM hallucinations.

How a new hire actually uses it

1. Open Google Gemini
2. Upload the reference document as Gem knowledge (a one-time setup)
3. Ask any question: "what's the minimum CPM for this product," "how do I handle a cross-month campaign," "what does this acronym mean"
4. Get an accurate, role-specific answer instead of needing to find and interrupt a senior coordinator

I built both layers end-to-end. For each:

• I designed the training curriculum: what order to teach things in, where to put checklists, which common mistakes deserve their own slide
• I wrote every slide of every presentation. Gamma helped with visual layout and formatting, but the content (workflows, rules, checklists) is mine, drawn from months of doing the job
• I authored the 437-line reference document, consolidating six existing training docs and my own operational notes into one structured source of truth
• I configured the Gemini Gem with the instructions and uploaded the reference doc as knowledge
• I structured the knowledge base specifically for AI retrieval: explicit rules, short chunks, clear headers, so answers are accurate, not generic

Treat presentations and AI as complementary, not competing

Most teams pick one or the other: either a static slide deck or a chatbot. I built both because they do different jobs. Presentations give new hires structure. They don't know what they don't know yet, so they need a sequence to follow. The AI assistant handles the messy middle: the questions that come up while you're actually doing the work, the ones that don't fit cleanly into "Week 1 Lesson 3."

Document for retrieval, not for reading

The reference doc isn't written like a textbook. It's written like an API spec: short headers, explicit rules, structured tables, because the AI agent needs to pull narrow chunks fast. A doc optimized for human reading produces an AI that gives long, vague answers. A doc optimized for retrieval produces an AI that answers a specific question with the specific rule.

Consolidate first, then automate

The first version of the knowledge base was just me reading through six existing training documents and my own notes and merging them. The AI piece came last. Most failed onboarding systems skip this step. They layer a chatbot on top of scattered docs and wonder why answers are wrong.

Build it for the next person, not just the next AI

The reference doc and presentations are useful even if the Gemini Gem disappears tomorrow. A new hire can read the doc, a new manager can update the slides, the same knowledge base could power a different AI platform. The system isn't locked to Gemini.

What the system uses. The intelligence behind the AI assistant is Gemini's underlying model. What makes it accurate is the knowledge base I authored.

• One-file setup for new hires: upload the reference doc into Gemini and the assistant works instantly. No multi-step configuration, no platform learning curve.
• Three structured training decks covering the entire role from day one through independent operation.
• Reduced dependency on senior coordinators for basic onboarding questions. Questions that previously required a Slack ping now get answered in seconds.
• One reusable knowledge base that consolidated 6+ scattered docs and could power any AI platform (not just Gemini).

• Curriculum design and AI prompting are the same skill in different costumes. Both come down to: what order do you reveal information, what context does the learner (or model) need before each step, where do the common mistakes go.
• The trick to making AI useful in a specialized role is the knowledge base, not the prompt. No prompt-engineering trick fixes a chatbot that has the wrong information. Get the source of truth right first.
• Documenting a role is how you scale your own knowledge. The 437-line reference doc is also my own personal source of truth. When I'm unsure of a rule, I check the doc I wrote.
• Onboarding is a force multiplier. Every hour I spent on the system saves multiple hours every time someone new joins the team. Compound returns.