Short answer: these are four typical situations in service businesses of 5 to 50 people, each with the workflow I would build, where a person approves, the sprint that fits and what I would not promise. Find the closest one to yours and bring it to the free 15-minute call.
These are scenarios, not client case studies. They are based on the kinds of situations owners ask me about. They are not results from a client engagement, the numbers are assumptions you can replace with your own, and nothing here is a promise. When a client agrees, I will publish a real case note with measured before and after figures.
A 12-person agency where leads sit and discounts wait on the founder
Single AI workflow sprint · enquiries and first replies
The situation
Enquiries arrive by web form, email and LinkedIn. Someone checks the inbox when they can. Quotes are built by hand, and any discount or non-standard scope waits until the founder is free. Good leads go cold while the answer sits in a queue.
What is actually breaking
Not the speed of typing. Three things: nobody owns the first reply, the quote is rebuilt from scratch each time, and the discount limit is only in the founder's head, so every exception comes to them.
What I would fix before automating anything
Write down the discount limit and the scope the team can approve alone, and name who owns the first reply. Automating an undecided limit would just send the same exception to the founder faster.
The workflow I would build
- New enquiry arrives
- Enriched and scored
- Reply and quote drafted
- Sign-off above the limit
- Project and onboarding start
Where a person approves
Anything within the written limits goes out under the account manager's name. Only a discount or scope above the limit reaches the founder's phone for a yes or no. The owner sets the limits and can change them at any time.
The sprint that fits
Built in the agency's own CRM and inbox (for example HubSpot and Microsoft 365 or Google Workspace), with a recorded walkthrough, a one-page runbook and 30 days of tuning.
Illustrative arithmetic. Assumed inputs, not a finding: 4 people, 2 hours a week each on first replies and quotes, at $45 an hour. That is 8 hours a week, 368 hours over 46 working weeks and about $16,560 of payroll time a year. Your own numbers go into the calculator on the homepage.
What I would not promise. No promise of more leads, a higher close rate or hours saved. I guarantee the build against written acceptance criteria. AI drafts can be wrong, which is why the approval step is built in.
A 25-person professional services firm where onboarding is copy and paste
Single AI workflow sprint · client onboarding and handover
The situation
When a client signs, someone creates folders, sets up accounts in several tools, sends the welcome pack and books the kick-off. It takes the same steps every time, but it depends on one coordinator remembering them all. When that person is away, onboarding slips.
What is actually breaking
A repeated, rule-based process spread across four tools, with the steps held in one person's head and no written place where a non-standard client is handled.
What I would fix before automating anything
Write the steps as they really run, including what happens with a non-standard engagement. That written map is the product of the first 2-hour mapping session, and you keep it whether or not anything is built.
The workflow I would build
- Client signs
- Accounts and folders created
- Welcome pack sent
- Non-standard terms flagged
- Kick-off booked
Where a person approves
Standard clients flow through with no one touching them. A client with non-standard terms, or a missing piece of information, is flagged to a named person before anything is sent to the client.
The sprint that fits
Built in the firm's own accounts, with credentials the firm grants and can revoke. Ownership transfers at final payment, and there is no lock-in.
Illustrative arithmetic. Assumed inputs, not a finding: 3 hours of coordination per new client, 6 new clients a month, at $50 an hour. That is 18 hours a month, 216 hours a year and about $10,800 of payroll time a year.
What I would not promise. No promise of faster onboarding times or fewer errors until the workflow has run and been measured. I would add a before and after measurement to the written scope if the owner wants one.
An 8-person firm with automations that half work and tools nobody remembers buying
AI cleanup sprint
The situation
Over the years someone set up a handful of Zaps and a couple of Make scenarios. Some fail silently, two do the same job, and the firm pays for overlapping subscriptions. Nobody can say what is still running or what it does.
What is actually breaking
No map. Without a list of what is live, who owns it and what it touches, fixing one thing breaks another, and nobody dares switch anything off.
What I would fix before automating anything
The first deliverable is the written map of every live automation. The fix follows the map, not the other way round.
The workflow I would build
- Audit every live automation
- Repair failing flows
- Remove duplicate subscriptions
- Owner approves what is switched off
- Priority list handed over
Where a person approves
Nothing is switched off without the owner confirming it. Anything that touches client data or money is listed and approved before it is changed.
The sprint that fits
Failing flows repaired, duplicate tool subscriptions removed where I find them, a written map of every live automation and a priority list. 30 days of support follow.
Illustrative arithmetic. Nothing to calculate in advance, because the starting point is unknown. The map shows what the firm is paying for, and that figure is the real before.
What I would not promise. No saving is promised. I remove duplicate subscriptions where I find them, and I do not know in advance how many there are.
A 40-person services firm where sales, delivery and finance all run through the owner
End-to-end AI ops engine
The situation
Enquiries, project handover, weekly reporting and invoicing each run in a different tool. The owner is the link between them: checking, approving, chasing and reconciling. A two-week absence is not something the business could survive.
What is actually breaking
The owner is the integration layer. Decisions, data and exceptions all route through one person because nothing connects the tools and nothing is written down about who can decide what.
What I would fix before automating anything
Decision rights first: who may approve what, up to which limit, and what must be escalated. At this size I would usually suggest starting with one pipeline before connecting all three.
The workflow I would build
- Sales to delivery handover
- Delivery status pulled together
- Invoice raised on completion
- Exceptions reach the owner
- Weekly exceptions report
Where a person approves
The automation enforces the written limits, so routine work moves without the owner. The owner receives an exceptions report and approves only what is above their limits.
The sprint that fits
Connected sales, delivery and finance on a shared data layer in the firm's own tools, an exceptions report and full handover, with 90 days of tuning.
Illustrative arithmetic. Assumed inputs, not a finding: 10 people, 3 hours a week each on status chasing and reconciling, at $55 an hour. That is 30 hours a week, 1,380 hours over 46 working weeks and about $75,900 of payroll time a year.
What I would not promise. No promise that the business runs without the owner, or of savings. The aim is to reduce what depends on the owner. The price depends on the number of systems and the monthly volume, and I confirm it in writing before work starts.
Which one is closest to yours?
If none fits exactly, that is normal. The pattern is usually the same: a repeated process, spread across several tools, with limits that live in one person's head. The free AI Workflow Candidate Map scores your own workflows in about ten minutes, and nothing is sent anywhere.
Bring your situation to a 15-minute call
No pitch and no slides. I will tell you which sprint, if any, fits, and what I would fix first.
Book a free 15-minute workflow call Or take the 2-minute diagnostic