How do I roll out AI across multiple companies or business units?
It depends on how much the units have in common. If they share systems, roles and workflows, such as several offices of one firm, set one policy, one approved tool set and one training program centrally, then let each unit pick its own first workflows. If they are truly separate businesses, such as a private equity portfolio, standardize only the guardrails and the playbook, start with two or three willing companies, and let results pull the rest in. Mandating the same tools everywhere at once is the most common way these rollouts stall.
Who this is for
Operating partners and portfolio operations teams at private equity firms, COOs of multi-office firms, and leaders of groups with several business units who are asked to “get AI going everywhere.”
The short version
- Decide centrally on guardrails, a playbook and a way to share results. Leave workflow choices local.
- Start with a few units that have a willing leader and a clear problem.
- Measure each unit on its own named workflows, then compare.
- Expand in waves, using the first units’ results and people to teach the next.
What to standardize and what to leave local
| Decide centrally | Leave to each unit |
|---|---|
| AI acceptable use policy and data rules | Which workflows to start with |
| A short list of approved tools | Who the local AI champion is |
| The rollout playbook and templates | Pace within the agreed timeline |
| How results are measured and reported | Unit-specific prompts and agents |
| Shared training materials | Extra training for specialist roles |
The central team’s job is to make it easy and safe to start, not to run every unit’s project. Separate businesses with their own leadership will resist anything that looks like head office choosing their software.
The playbook
1. Set the guardrails first (weeks 1 to 2). Publish one AI policy that covers confidential data, client information and which tools are approved. Units can add stricter rules but not looser ones. See how to write an AI acceptable use policy.
2. Take a quick inventory (weeks 1 to 3). Ask each unit three questions: which AI tools people already use, which recurring tasks eat the most time, and who is keen to lead. Short calls beat long surveys. You will usually find shadow tool use, which tells you where demand already is.
3. Pick the first wave (week 3). Choose two or three units with a leader who wants to do this and a clear, recurring workflow problem. Avoid starting with the unit in the most trouble. It needs focus elsewhere and a slow start there will color everyone’s view.
4. Run a pilot in each first-wave unit (weeks 4 to 10). Each unit picks one or two named workflows, records a baseline, gets hands-on training on that work and runs for about six weeks. Our six-week pilot playbook gives the detail, and how to find workflows to automate helps units choose.
5. Share results across units (week 10). Hold a short session where first-wave leaders show what changed, what did not and what they would do differently. Peers are more convincing than head office.
6. Expand in waves (from week 11). Bring in the next group of units, pairing each with a first-wave unit. Reuse the templates, prompts and training that worked. Stop doing what did not.
7. Decide what to build once and share. When several units need the same thing, such as a reporting agent or a document search tool, consider building it once centrally and configuring it per unit. See custom AI vs off-the-shelf tools.
How to measure progress across units
Measure each unit on its own workflows, using the same simple layers everywhere: whether people use AI on the target work, how often, what changed in time or review rounds, and whether quality held. Report these in one shared format so leadership can compare without forcing every unit onto the same use case. The method is in how to measure whether AI training worked.
Watch tool spend too. Licenses bought per unit add up fast and often go unused. See how to control AI costs across a team.
Roles that make it work
- A central owner with time set aside, not a side project for someone already full.
- A sponsor in each unit, usually the CEO or COO of that company or office.
- A local champion who learns fastest and helps colleagues day to day. See how to build AI capability inside your firm.
- Someone accountable for data and security who signs off on tools and integrations.
Common mistakes
- One mandated tool everywhere on day one. Units with different work need some room to choose within the approved list.
- Training without tool access. People forget what they cannot use the next day.
- Measuring logins instead of work. Activity is not value.
- Rolling out to every unit at once. The central team gets spread thin and early problems repeat everywhere.
- No policy until something goes wrong. Put the data rules in place before the first pilot.
When you do not need us
If you have a central operations team with time to run this, units that already use approved AI tools and a policy in place, you may only need the playbook above. Outside help is more useful when the central team is stretched, units are very different from each other or you need custom tools shared across several businesses.
How Ruby helps
- AI training: hands-on training for each unit, built around its own workflows, plus material you can reuse in later waves.
- AI strategy: running the inventory, choosing first-wave units and setting up shared measurement.
- AI engineering: building tools or agents once and configuring them for each unit, with controls built in.
If you need to get AI working across several companies or offices, book a call with Ruby.