Don't buy one AI tool for everyone. Tier it by audience, because your people don't all do the same work. We rolled out Amazon Quick to 200 AI Champions at a North American beverage distribution business alongside Microsoft 365. This is the tiered licensing model, the setup steps most people skip, and the prompt framework we taught.
Most organizations ask us the same question when they start thinking about AI: which tool should we buy?
It's the wrong question, and we can prove it.
Earlier this year we ran a three-month AI enablement program for a North American beverage distribution business with thousands of employees. They didn't pick one tool. They picked three, gave each one to a different audience, and gated all of it behind a policy. Amazon Quick went to 200 AI Champions. Claude went to a smaller group of specialists. Everybody else got the Copilot Chat capability already sitting inside the Microsoft 365 licenses they were paying for.
That structure is the single most useful thing we've learned this year, and almost nobody is doing it.
The short version
- Don't buy one AI tool for everyone. Tier it by audience, because your people don't all do the same work.
- Copilot Chat is already included in most Microsoft 365 licenses. Start there for the majority and spend your budget where it earns more.
- Amazon Quick is an excellent fit for a champion tier, because it connects to Teams, Outlook, Calendar and your local files, and it runs as a desktop app alongside Microsoft rather than fighting it.
- Make policy training a prerequisite for access. No completion, no license.
- Four setup steps decide whether Quick is useful or disappointing, and most users skip all four.
- Teach one prompting framework and repeat it everywhere: Goal, Context, Expectations, Source.
Most organizations ask the wrong question
"Which AI tool should we buy" assumes one answer fits a warehouse supervisor, a trade analyst and a finance director. It doesn't.
What actually works is asking three questions instead:
- Who needs AI to be available and safe? That's everyone, and you've probably already paid for it.
- Who needs AI to be connected to their daily work? That's a meaningful minority, and they're the ones worth a real license.
- Who needs AI to do advanced and agentic work? That's a small group, and they need something different again.
Answer those three and the licensing decision writes itself.
The three tiers
| Tier | Who | Tool | Why |
|---|---|---|---|
| All employees | Everyone, thousands of people | Copilot Chat, already included in Microsoft 365 | Baseline familiarity and safe habits, at no additional license cost |
| AI Champions | 200 selected users across every function | Amazon Quick | Deep daily use, connected to their real work, and they act as peer leaders for their teams |
| Specialists | A small group doing complex work | Claude | Advanced and agentic use cases that need a different kind of capability |
The part people miss is that the middle tier does double duty. The Champions aren't just license holders. They're the adoption mechanism. You're not buying 200 licenses, you're building 200 people who can answer a colleague's question without opening a ticket.
That's also what makes the economics work. You're paying for depth where depth pays off, and using included capability everywhere else.
Policy first, access second
Before anyone received a license, they had to complete a mandatory eLearning built from the organization's Acceptable Use of AI Systems policy. Completion was documented as a prerequisite for getting access to either Quick or Claude.
This is worth copying for three reasons:
- It gives the policy teeth. A policy nobody reads isn't governance, it's paperwork.
- It reduces shadow AI. People use unapproved tools when nobody has told them what's approved. Telling them, clearly, at the moment they get access, works.
- It creates a clean enrollment moment. "Finish the training, get the tool" is a far better message than "here's a tool, please be careful."
If you do nothing else from this article, do this one.
The four setup steps almost everyone skips
Here's where Amazon Quick surprises Microsoft organizations. It doesn't sit off to one side. Set up properly, it reaches into the tools your people already live in.
We built a getting started guide for the Champions because the difference between a configured Quick and an unconfigured one is the difference between a useful assistant and a disappointing chatbot. Four steps:
- 1
Connect your work tools.
Under Settings, Capabilities, Connectors, sign in to Microsoft Teams, Outlook and Outlook Calendar. That gives Quick the ability to read chats and channels, read and draft email with your approval, and check availability to prep for meetings. Most people never open this screen.
- 2
Grant access to your local folders.
In the desktop app, add the folders you actually work in and turn the indexing toggle on. Now Quick can search your own files. This is the step that converts it from a generic assistant into one that knows your work.
- 3
Configure your activity feed.
Choose which notifications surface, Teams, email, calendar. This is what makes it feel like a workspace rather than a chat window.
- 4
Write your response preferences.
Under Settings, Customization, write a short blurb about your role, your communication style, and how you like information delivered. It shapes every response you get from then on.
That fourth one is the sleeper. Two minutes of typing, and the quality of every answer afterwards changes. In our sessions it was consistently the thing people said they wished they'd known first.
The prompt framework we taught
We teach one framework, and we repeat it in every session, every guide and every follow-up:
Goal, plus Context, plus Expectations, plus Source.
- Goal: what you want done. Summarize, draft, analyze, compare, create.
- Context: background that helps. Audience, time period, market, brand.
- Expectations: how you want it delivered. Bullets, one page, a table, a tone.
- Source: the data it should use. An attached file, your email, a folder.
You don't need all four every time. But watch what happens as you add them.
| Prompt quality | Example | What you get |
|---|---|---|
| Goal only | “Summarize this sales report.” | It guesses what matters and how to format it |
| Goal and Source | “Summarize the attached Q2 report, focusing on our top 10 brands by volume.” | A clear target, but no sense of who it's for |
| All four | “Summarize the attached Q2 report for our top 10 brands. I'm presenting to our division VP tomorrow, so focus on year over year trends and any brands significantly up or down. Format as a one page executive summary with bullets.” | Something you can actually use |
The third one isn't a harder prompt to write. It's the same request with the context a colleague would have given automatically. That's the whole lesson, and it lands in about ninety seconds when you show the three side by side.
We paired the framework with ready-to-use prompts for each function, written against the work those teams genuinely do. Account call prep for field sales. Variance analysis for finance. Promotion ROI for trade analytics. Delivery exception root causes for operations. Contract comparison for legal. Nobody got a generic prompt library, because nobody uses a generic prompt library.
Why 200 champions and not 3,000 users
The instinct is to give the good tool to everybody. Resist it.
- 200 is enough to reach everyone. Champions sit inside teams. Their colleagues ask them, not IT.
- 200 is small enough to support properly. You can run real onboarding, a peer kickoff, monthly meetings and office hours for 200 people. You cannot do that for 3,000.
- 200 gives you a clean measurement group. You know exactly who has the tool, what they were told, and what they did with it.
- 200 creates demand rather than satisfying it. When colleagues see a champion do something useful, they ask how. That pull is worth more than any launch email.
A broad rollout with no support produces a lot of licenses and very little behavior change. We've watched it happen often enough to stop recommending it.
What we measure
Adoption programs that can't produce a number get cancelled. These are the measures we collect and report:
- Active usage. Monthly and daily active users, against licenses assigned.
- Depth of use. Prompt volume, average prompts per user, and how many people have built their own agents.
- Business impact indicators. Time saved, estimated financial impact, license ROI.
- Sentiment and confidence. Survey-based, because how capable people feel predicts whether they'll keep going.
- Risk and cost signals. Reduction in redundant or shadow AI tools.
- Training reach. Coverage, completion rates, and whether people found it useful.
We run a baseline survey before anything starts. Without it you have no way to prove a change happened, and you'll be having a difficult conversation at renewal.
A simple way to explain the tiers to your leadership
You don't give every employee the same vehicle. Most people need something reliable to get to work, and you already own a fleet of those. A smaller group covers territory all day and needs something built for the mileage. A handful do specialized work that needs specialized equipment.
Buying one vehicle for everyone means overpaying for most of your people and underserving the ones who'd actually use it.
Licenses are the fleet. Tiering is knowing who drives what.
What IT leaders should do next
- 1
Check what you already own.
Most Microsoft 365 licenses include Copilot Chat. Confirm it's enabled before you spend a dollar on anything else.
- 2
Define your three tiers on one page.
All employees, champions, specialists. Name who's in each.
- 3
Get the policy done and turn it into training.
Then make completion the gate for access. This takes longer than you think, so start it first.
- 4
Pick your champions deliberately.
Across every function, chosen for influence and willingness, not seniority.
- 5
Send a setup guide before the first session.
Connectors, local folder indexing, activity feed, response preferences. People who arrive configured learn twice as much.
- 6
Teach one prompting framework and repeat it.
Goal, Context, Expectations, Source. Consistency beats cleverness.
- 7
Baseline before you start.
Usage, sentiment and confidence. You cannot show a change you never measured.
- 8
Publish which tool for what.
One page, plain language. Ambiguity between tools is worse than any single tool's limitations.
How Change Champions helps
We're a PROSCI-certified change management firm and a Microsoft Solutions Partner, and we've supported adoption for more than 60,000 users. We're platform neutral by design, because our clients rarely run just one thing, and increasingly they shouldn't.
For organizations rolling out Amazon Quick alongside Microsoft 365:
- Tiering and licensing guidance. Which audience gets what, and the user-facing “which tool for what” guidance that makes it stick.
- Policy enablement. Turning your Acceptable Use policy into mandatory eLearning, published in your LMS, tracked to completion, gating access.
- Champions network design. Selection criteria, role definition, onboarding, peer kickoff, and the monthly cadence that keeps it alive.
- Role-based workflow enablement. Guided demos, hands-on labs and office hours built around what each function actually does, using their own files.
- A resource hub. FAQs, prompt libraries, recordings and onboarding paths, so the answers outlive the sessions.
- Measurement and executive reporting. Baselines, usage, depth, sentiment, and a readout your sponsor can defend.
See our AI adoption consulting services for CIOs and IT leaders, or read our seven-step approach to Amazon Quick adoption and our Amazon Quick vs Microsoft Copilot comparison.
Ready to turn rollouts into real adoption?
Whether you're driving Microsoft 365 and AI adoption, reducing human risk, or leading a change program, a 30 minute discovery call is the fastest way to see if we're a fit.