Once you know where AI can add value, redesign your workflow so AI supports productivity while people stay in control of key decisions. This works best when you use the strengths of both AI and people.
This stage uses the pain point mapping you created in Stage 2: Identify opportunities.
For each task you plan to improve, be clear about what AI will do, what people will do, and how outputs will be checked (Table 1). Start small and be specific.
For example:
Any workflow that uses AI needs checkpoints. Choose an oversight pattern that suits the task and decide:
| Oversight pattern | When to use | How it works |
|---|---|---|
| Check before acting (human-in-the-loop) | High business impact | A person approves every AI output before using it |
| Watch while acting (human-on-the-loop) | Medium business impact, high volume | AI acts automatically; a person checks and steps in when needed |
| Spot-check | Low business impact, very high volume | A person reviews a sample of AI outputs |
| Exception handling | Predictable edge cases | AI handles routine cases and flags exceptions for a person’s review |
Clear roles and careful checks by people helps prevent errors in AI outputs from scaling. It also makes it clear who’s responsible for the final outcome at each step.
Learn what’s working and what needs adjusting in your workflow. For example:
Feedback loops let you continually improve the workflow based on real use. They help you catch issues early and avoid repeat mistakes.
Create a short workflow specification for your redesigned process. Include:
If the workflow isn’t documented, it’s harder to test and run consistently. Clear documentation also helps people understand their role, what’s changing, and why.
Fill out Table 6 and 7 in the business process mapping template.
Sarah redesigned the client onboarding process to improve speed and consistency. In the redesigned workflow, AI helps with repeat tasks and her team stay in control of key decisions.
| Process name | New client onboarding (redesigned) |
|---|---|
| Current state | Manual drafting, ad-hoc follow-ups, and an inconsistent welcome process. |
| Target state | AI-assisted drafting, consistent follow-ups, and a standardised welcome sequence. |
| Expected benefits | Reduce onboarding time from 5–7 hours to 2–3 hours, reduce document chasing from 3 rounds to 1, and avoid welcome email errors. |
Sarah then recorded who does what and how outputs are checked.
| Task | AI responsibility | Person responsibility | Oversight pattern |
|---|---|---|---|
| Quote creation | Draft a quote based on client details and service type. | Review, edit, send the quote. | Check before acting (human-in-the-loop) |
| Document request email | Draft email listing required documents based on client type. | Review for accuracy and send the email. | Check before acting (human-in-the-loop) |
| Follow-up reminders | Send reminders on day 3 and day 7 if documents are missing. | Review the reminder list weekly and call the client if documents are still missing after day 7. | Watch while acting (human-in-the-loop) |
| Welcome email | Draft an email with correct portal link. | Review and send the email. | Check before acting (human-in-the-loop) |
Key design decisions
Before investing in building your new process, use one or more of these low‑cost checks:
Document the results and your decision to proceed.
Testing before building AI helps you catch design flaws early. It also supports safer pilots by making sure checks and escalation paths work as intended.
Use this template across all stages to prepare your business.
You should now have a redesigned workflow and a documented specification that is ready to pilot. Come back and complete this stage for as many workflows as you need.