Choose a job your team repeats every week. It should follow rules you can write down, use information already in your tools, and fail in ways you can spot and fix. Start with one trigger and one output. Measure how it runs today, and keep a person approving anything that needs judgment. The biggest job is rarely the best first project. Pick the one you can prove.
The workflow at a glance
- The buyer question: A business owner with five or six repetitive jobs and no fair way to pick one: what should we change first, and how will we know it worked?
- The current process: Candidates are chosen by whoever complains loudest. Nobody has written the steps down, so there is no baseline and no way to tell later whether anything changed.
- What can improve: A short, scored list of candidates, each with evidence (volume, error cost, data location), and one chosen workflow with a single trigger, a single output, and a written acceptance test.
- Tools and access needed: Read access to the systems the candidate workflows touch (inbox, forms, CRM, spreadsheets, file storage), one week of task logs, and an owner who can describe each process. No new software is needed to choose.
- Where people still approve decisions: The owner signs off on which workflow is chosen and on the list of decisions that will stay human (pricing, refunds, anything legal or relationship-sensitive).
- How to measure whether it works: Record the baseline before building: time per item, items per week, error or rework rate, and how long items wait. Re-measure on the same items after launch.
Examples marked “Hypothetical example” are illustrations, not client results.
Most owners already know which tasks annoy them. The hard part is deciding fairly which one to fix first. The method below scores each candidate, counts the cost of mistakes as well as time, and settles what stays human before anything is built.
Why the first choice matters more than the tool
In McKinsey’s survey of organizations using generative AI, redesigning workflows had the biggest effect on whether companies saw a bottom-line (EBIT) impact, out of 25 attributes tested. Only 21% of respondents said their organizations had fundamentally redesigned at least some workflows (McKinsey, The State of AI, 2025). McKinsey’s later 2025 survey found that AI “high performers” were nearly three times as likely as other respondents to have fundamentally redesigned individual workflows (McKinsey, 2025).
The gain comes from changing how the work runs. Software added to a messy process keeps the mess. As one practitioner guide puts it, if a process is unclear or inconsistent, “automation will repeat the problem at speed” (Waqar Syar).
Step 1: List the jobs that repeat
Spend one week writing down every task that happens more than once a week and involves moving information from one place to another. For each one, note:
- what starts it (an email, a form, a signed contract, a date),
- what “done” looks like,
- which tools it touches,
- who owns it today.
Look for the usual signs: someone copies information between systems, documents wait in an inbox, managers chase the same updates, or customers wait while someone prepares a predictable reply (OpSmith).
Hypothetical example (not a client case): A small gallery lists three candidates. Logging collector inquiries from the website into a spreadsheet, preparing the monthly sales summary, and chasing artists for artwork details before a show.
Step 2: Score each candidate on six factors
Score every candidate from 1 to 5. Write the evidence next to each score, such as “about 40 inquiries a month.” The evidence matters more than the total (OpSmith).
| Factor | Question to ask | What a 5 looks like |
|---|---|---|
| Frequency | How often does it run? | Daily or many times a week |
| Rule clarity | Can you write the steps without saying “it depends”? | Clear steps, few exceptions |
| Data access | Is the information already in a tool you can connect? | Structured fields, export or integration available |
| Time cost | How long does each run take, including waiting? | Long, repetitive handling |
| Cost of error | If it goes wrong, how bad is it? | Mistakes are visible and easy to reverse |
| Ownership | Is there one person who can approve and maintain it? | A named owner who cares about it |
These factors overlap with published scoring frameworks that use frequency, time cost, error cost, rule clarity, data availability and risk (Waqar Syar; CompileSquad). Treat the total as a starting point for discussion, not a verdict.
Step 3: Price the mistakes as well as the minutes
Time saved is half the sum. Also ask what a slow or wrong run costs. Speed shows why. In a study of 1.25 million sales leads, firms that tried to contact an online lead within an hour were nearly seven times as likely to qualify it as firms that waited even one hour longer (Harvard Business Review, 2011). In an inquiry workflow, a late reply can cost the sale as well as staff time.
For each candidate, ask what a delay costs, what a wrong output costs, and who would notice. A task with modest time savings but a high cost of delay can still be the right place to start.
Step 4: Decide what stays human before you build
Write down where a person must approve before anything goes out. NIST’s Generative AI Profile warns that AI systems can “confabulate,” meaning they state false information confidently, and it recommends defined human oversight roles (NIST AI 600-1). Relationships, sensitive pricing, unusual commitments and other high-stakes judgment calls should sit behind explicit approval.
Steer clear of these as a first project: work nobody owns, processes where two teams disagree on the rules, payroll or regulated approvals, and anything running on a system you’re about to replace (Automataize).
Step 5: Shrink the scope to one trigger and one output
Define the workflow from one starting event to one useful result. For example, “from a website inquiry to a draft reply and a logged contact,” not “the whole sales process” (OpSmith). Automate the common path first and send exceptions to a person (Automataize).
Step 6: Measure the baseline, then set the acceptance test
Before anything changes, record how many runs happen in a set period, how long each takes, how many go wrong, and how long the customer waits. Mark which numbers you measured and which you estimated (CompileSquad). Then agree on the pass mark the new version has to meet on real cases. Without a baseline, you can’t show whether the change worked.
Checklist: is this a good first workflow?
- It runs at least weekly, and you know the rough volume
- The trigger and the finished output can each be named in one sentence
- The information it needs is already in a tool you can access
- A wrong output would be noticed and could be reversed
- One named person owns it and will review exceptions
- You’ve written down which decisions stay with a human
- You’ve recorded a baseline (volume, time, errors, wait time)
- You’ve agreed an acceptance test before building
- There’s a manual fallback if the automation stops
Where to go next
Once you’ve picked a workflow, the next posts in this series cover the most common first choices. See connecting inquiries, CRM updates and sales follow-up, automating client onboarding and document collection, and using AI for proposals, reporting and admin. When you’re ready to roll this out more widely, read implementing AI across your team’s existing tools. For budgeting, see what business automation costs and how to scope it.
Grownup Business work starts with this choice. We agree the job, the approval points and the success measure before work begins, then test against real operating cases. A written scope comes before any payment. See Grownup Business, then request a demo and tell us which repeated job costs your team the most time.
If Grownup also manages your Instagram, the same workflow can pick up inquiries that start there. Instagram Management is a separate, optional service. You do not need it to start any business workflow.