Should we roll out AI to the whole company at once, or start with one team?
Start with one team when you’re still working out which tools, uses and rules fit your firm, when budget is tight, or when people are skeptical and need proof. Roll out company-wide when leadership has already chosen the tools, the basic rules are written, the use cases are common to almost everyone, such as drafting, summarizing and research, and you can support every team with training at the same time. Many firms do both in sequence: a short, focused pilot first, then a wider rollout built on what it taught them.
The deciding question is how much you already know about what works in your firm.
Who this is for
Owners, managing partners, COOs and operations leads who have decided to adopt AI and now need to choose how broadly to start.
The short version
- A one-team start is lower risk and gives you real evidence, but it’s slower and can leave the rest of the firm waiting or improvising.
- A company-wide rollout moves faster and treats everyone the same, but it spreads support thin and amplifies any early mistakes.
- Licenses alone don’t make either approach work. Training, clear rules and a named owner matter more than the size of the first wave.
What each approach means
Starting with one team means picking a single group, such as finance, operations or a project team, giving them tools and training, and focusing on a few specific workflows for a set period. The goal is to learn what works and show results before expanding. Our guide to running an AI pilot that doesn’t stall covers how to set one up.
A company-wide rollout means giving everyone access at roughly the same time, usually with firm-wide training, a shared acceptable use policy and a common set of starting use cases.
How the two usually compare
These are typical patterns, not rules. Results depend on your firm’s size, culture and how well each approach is run.
| Factor | Start with one team | Company-wide rollout |
|---|---|---|
| Speed to firm-wide use | Slower | Faster |
| Upfront cost | Lower, fewer licenses and less training | Higher, everyone at once |
| Risk if something goes wrong | Contained to one group | Spread across the firm |
| Quality of support | Concentrated and hands-on | Spread thinner |
| Evidence for leadership | Clear before and after results | Harder to isolate what worked |
| Fairness and morale | Others may feel left out | Everyone starts together |
| Shadow AI risk | Others may use unapproved tools meanwhile | Lower, an approved option exists for all |
| Best when | Uses and rules are still unclear | Tools, rules and core uses are settled |
When starting with one team is the better choice
- You haven’t decided which tools to standardize on, or you want to test more than one.
- You don’t yet have clear rules about what data can go into AI tools.
- Leadership or partners need proof before committing a larger budget. See how to get partners to use AI.
- The most valuable uses are specific to one function, such as estimating, underwriting or proposal writing.
- You have a team with a willing leader and a real, measurable problem to solve.
In these cases a broad rollout tends to produce scattered, shallow use. People log in, try a few prompts and drift back to old habits, which is one of the main reasons AI adoption stalls.
When a company-wide rollout is the better choice
- People across the firm are already using personal AI accounts, and you need an approved, safer option quickly.
- The main uses are general, such as writing emails, summarizing documents, meeting notes and research, so almost everyone benefits.
- You already have a policy, a chosen tool and prompting standards in place.
- You can run training for every team, not just hand out logins.
- A slow, staged approach would create an awkward split between teams that have AI and teams that don’t.
The honest trade-offs
Pilots can become a holding pattern. A one-team start only works if it has an end date, clear success measures and a decision at the end. Without those, it can drag on while the rest of the firm waits.
Broad rollouts can waste money quietly. Paying for seats that sit unused is common when access arrives without training or a reason to use it. Our page on why licenses are not a strategy explains why, and controlling AI costs across a team covers how to keep spend in check.
The first team shapes the story. If the pilot team is skeptical or overloaded, a weak result can set the whole program back. Pick a team with a motivated lead and time to engage.
Firm-wide mistakes are firm-wide. A data handling error or a poorly chosen tool affects everyone at once in a broad rollout. Settle the basic rules before you open access.
A middle path that often works
Many firms combine the two. Everyone gets a short baseline session on safe, everyday use and the firm’s rules, so no one is left out or tempted to use unapproved tools. At the same time, one team goes deeper on a specific, measurable workflow with more hands-on support. When that team shows results, its playbook becomes the model for the next teams.
This gives the whole firm a safe starting point while still producing real evidence. It also gives you a clearer picture of how long training takes before you plan the wider rollout.
Questions to ask before choosing
- Have we chosen our tools and written our rules? If not, start smaller.
- Are the most valuable uses general or specific to one team? General favors a wider rollout. Specific favors a focused start.
- Are people already using AI on their own? If so, an approved option for everyone may be urgent.
- Can we support everyone with training at the same time?
- How will we measure results, and who decides what happens next?
If you’re still mapping out the steps, see our AI implementation roadmap.
How Ruby helps
Ruby Advisory helps firms decide how broadly to start and then makes that start work. Our recommended path for most firms is the middle path: give everyone a short baseline session on safe, everyday use and the firm’s rules, pick one team with a real problem for a focused pilot with a set end date and clear measures, then expand team by team using what the pilot proved. We help with:
- AI strategy: choosing the first team, tools and success measures, and writing the rules.
- AI training: baseline sessions for the whole firm and deeper, hands-on training for the pilot team.
- AI engineering: building the workflows or agents the pilot needs and keeping them running.
To talk through how to start at your firm, book a call.