If you lead a practice, the AI conversation usually lands in one of two extremes. "This will change everything." Or "this is too risky to touch." Both are useless on a Monday morning.
The real question is: which workflow should we ship first so we get capacity back without creating compliance debt?
If you can't ship the first workflow in three days, you picked the wrong first workflow.
Not because three days is magic. Because the first win needs to be narrow enough to control, safe enough to defend, and valuable enough that people actually notice when it works.
The first workflow isn't your biggest workflow
Most firms instinctively pick the biggest pain. "Let's automate advice." "Let's replace half the back office." "Let's redesign the whole client experience." That is how you end up in a six-month project with nothing deployed.
The first workflow is the one that proves you can run AI safely with real inputs, real constraints, a review gate, and an evidence trail. Once you have shipped one, the second becomes easier. The third becomes repeatable. That is how you build AI muscle.
Six criteria for a great first workflow
Use this as a quick scoring model. The best first workflow scores high on most of these.
Immediate time back
It should remove repeated admin, not create new steps. If the workflow saves 15 minutes once a month, it won't change behaviour and it won't earn its place.
Clear inputs and a clear output
If you can't define what goes in and what comes out, you can't supervise it. Good first workflows have bounded inputs (a document pack, a template, a checklist) and bounded outputs (a draft email, a completeness report, a structured extraction).
A natural review gate already exists
The best first workflows already have a human review point. That means you aren't inventing governance. You are formalising what people already do. Draft, review, send.
Low integration dependency
If your first workflow requires three vendor integrations, you will stall. Start with workflows that can run with controlled inputs, controlled outputs, and minimal systems touch. Mature into deeper integrations after you have shipped.
Low "hallucination harm"
Early workflows should tolerate uncertainty by design. The workflow should flag exceptions, ask for missing inputs, and stop when unsure. It shouldn't be making irreversible decisions on day one.
Easy to measure
If you can't measure it, you can't improve it. Pick a workflow where you can track time saved, rework reduced, exceptions raised, and quality improvements.
First-workflow shortlist
Here are the ones that tend to work first in financial planning and adjacent professional services.
Client comms drafting with approvals
Input: client email thread, required disclosure insert, tone guide. Output: draft email, missing-info flags, required disclosure check.
Why it is a good first workflow: immediate time back, a natural review gate, and the client experience improves without anyone hitting "auto-send".
File note and record completeness check
Input: file note, checklist, required fields. Output: completeness report, flagged gaps.
Why it is good: it reduces remediation, supports supervision and consistency, and carries low risk if the system only flags and never invents.
Document pack triage and extraction
Input: client pack (PDFs and emails), extraction schema. Output: structured summary, missing-doc list, exception flags.
Why it is good: it cuts admin time, creates consistency for downstream steps, and is easy to measure.
Template-driven meeting follow-up pack
Input: meeting transcript or notes, template, required prompts. Output: follow-up email draft, task list, compliance reminders.
Why it is good: it reduces the "after meeting" drag and keeps outputs templated and reviewable.
What not to pick first
If you want to avoid pain, avoid these as your first workflow.
- Anything that claims "fully automated advice"
- Anything with undefined data access ("it can see everything")
- Anything that requires a brand-new operating model to review
- Anything that relies on "trust the model, it is accurate now"
These might be future workflows. They aren't your first.
Why the three-day pilot pattern works
A three-day pilot forces discipline. You define the boundaries. You define the gates. You define the evidence. You ship something people can actually use.
It also forces the right behaviour. You stop talking about AI and start operating with it. For leaders, that matters, because adoption follows shipped outcomes, not strategy memos.
A simple exercise you can run with your team
If you want to pick your first workflow quickly, run this five-step exercise.
- List the top ten repeated tasks your team hates.
- Circle the three that already have a review step.
- Pick the one with the clearest inputs and outputs.
- Define "done" in one paragraph.
- Decide the review checklist.
If you can't do steps three to five, the workflow is too fuzzy to be first.
Where workshops fit
The first workflow is a behaviour change as much as a build. That is why role-based workshops matter. Employees learn safe, repeatable patterns. Builders learn how to implement gates, boundaries, and logs. Champions learn how to keep it consistent as tools change underneath them.
This is how you turn one pilot into a culture.
Over to you
If you had to ship one workflow in three days that you could stand behind in front of compliance, what would it be?
If you aren't sure, that is a good sign you need a tighter selection process, not a bigger strategy deck. Bring your shortlist to a AI Fitness Review and we will pressure-test it with you.