Your firm should start with AI by listing the recurring work your people already do, scoring it for frequency, effort and risk, and picking two or three workflows where AI can help soon and where someone is willing to own the change.
Starting with a tool, a vendor demo or a firm-wide program tends to produce activity without much change. Starting with specific work tends to produce results you can see and build on.
Who this is for
Managing partners, CEOs and COOs who know AI matters, have read plenty about it, and want a sensible, low-risk first move for their own firm.
The short version
- Inventory the work first. You can’t prioritize what you haven’t written down.
- Score each workflow on a few simple criteria, then pick a small number to start with.
- For each one, decide whether it needs training, a policy decision, or something built.
Step 1: Build a workflow inventory
Ask each team lead to list the recurring tasks their team does in a typical week or month. Keep each entry short: what the task is, who does it, roughly how long it takes and what tools or documents it touches.
Good prompts for the conversation:
- What do you do every week that feels like copying, reformatting or summarizing?
- What takes senior people’s time that a capable junior could draft?
- Where do people wait on information from someone else?
- What reports or updates are produced on a regular schedule?
- What do new joiners find hardest to learn?
Expect a list of perhaps twenty to fifty items across the firm. That is plenty.
Step 2: Score each workflow
Score each item from 1 (low) to 3 (high) on five criteria.
| Criterion | What a high score means |
|---|---|
| Frequency | Happens weekly or more often |
| Effort | Takes meaningful time, especially from senior or expensive people |
| Text and document weight | Mostly reading, writing, summarizing or restructuring information |
| Checkability | A person can quickly tell whether the output is right |
| Low risk of harm | A mistake would be caught before it reached a client, investor or regulator |
Add the scores. Then apply two filters before ranking:
- Data sensitivity. If the workflow involves confidential client or deal data, note which tools and settings would be acceptable. This doesn’t rule it out, but it may change the order.
- Ownership. Is there a named person who wants this to improve and will spend time on it? A high-scoring workflow with no owner is usually a poor first choice.
Step 3: Pick two or three to start
From the top of the list, choose a small set that:
- spans at least one quick, visible win and one more substantial workflow
- has an owner in each case
- can be tried with tools you already have, or with minimal setup
Resist the urge to start ten things. A few well-run starts give you evidence, internal examples and trained people. Many thin starts tend to give you noise.
Step 4: Decide what each one needs
For each chosen workflow, ask which kind of help it needs:
- Training if the tools you have could do the job and people just need the skill and the habit.
- A policy or process decision if the blocker is uncertainty about data, approval or responsibility.
- Engineering if the workflow repeats often, spans several systems and would need re-prompting by hand every time.
Often, first workflows land in the training column. That is a good sign: it means you can start quickly and cheaply.
Step 5: Run it, review it, decide
Give each workflow a short trial period with a clear before and after, such as time taken, quality, or how often rework is needed. Review honestly at the end: scale it, adjust it or stop. Then return to the inventory and pick the next candidates. The list becomes a living plan rather than a one-off exercise.
Common mistakes
- Starting with the most ambitious idea. Complex, high-risk workflows make poor first projects. Earn trust on simpler ones.
- Asking only leadership. The people doing the work usually know where time is lost.
- Ignoring checkability. If nobody can quickly tell whether the output is right, errors slip through and trust erodes.
- Choosing workflows without owners. Change needs someone who cares about the result.
- Treating the first list as final. Priorities will shift as people learn what AI does well.
When you do not need us
If your team leads can spend an hour each on the inventory, and someone senior can run the scoring conversation without it turning political, you can do this entirely in-house with the method above. Outside help is most useful when you want a neutral view across teams, need a recommendation you can take to partners or a board, or want the training and building done alongside.
How Ruby helps
- AI strategy: running the inventory and scoring with your teams and giving a clear, honest recommendation on where to start, including what to leave alone for now.
- AI training: hands-on sessions on the chosen workflows, so the first wins actually happen.
- AI engineering: building the workflows that deserve a system, once the evidence is in.
A useful first step is one practical hour of AI training for the team that owns your top workflow.