A sample AI roadmap: what you get after a consultation

What you will take away

  • How one workflow is written up: where the hours go today, three ranked changes, what not to do yet
  • How each change shows its benefit, effort and risk
  • What a concrete first step looks like
Sample

See the format before you pay. The sample AI roadmap was made for a fictional cleaning company of 32 people and is labelled as a sample; it shows how I describe each task, what it saves, what it needs and what it risks.

Download the sample as PDF

Sample document. The company, the people and every number below are fictional. It shows the format and depth of the written roadmap you receive within 24 hours after our 45-minute consultation on one of your workflows.
AI roadmap · one workflow · after our call

Company A: from client request to paid invoice

Commercial cleaning · 32 people: 24 cleaners and technicians on client sites, 8 in the office · about 60 business clients on monthly contracts, plus one-off jobs · United Kingdom · October 2026 · Prepared by Lidia Sorokina, Elivance Tmi

In short

What hurts
Almost two full-time office roles, about 76 hours a week, go into retyping, phone calls and chasing payment.
Where to start
Send each invoice the moment a job is marked done, and let reminders chase the payment.
What it gives
By our estimate, the three changes give back 45 to 50 hours a week, more than one full-time role, so a third coordinator is most likely not needed.

How it works today

RoleWhere the time goesHours a week
Coordinators
two people, 40 h each
Reading requests and writing repliesabout 10 each
Retyping requests into the scheduling toolabout 6 each
Quotes for one-off jobs, written from scratchabout 3 each
Visit changes and sick calls, phoned one by oneabout 8 each
27 of 40 each, about 54 for both54
BookkeeperInvoices made by handabout 4
Phoning late payersabout 10
14 of 4014
Senior or managerChecking that visits happened, phoning cleaners, collecting reports8
TotalAlmost two of the eight office rolesabout 76

Hours are your estimates from the call. Two weeks of simple time tracking will confirm or correct them. Quotes stay as they are for now; they get easier once the inbox drafts replies.

Three improvements, ranked

Ranked by benefit against effort, best first.

1Invoices and paymentEffort: low · bookkeeper

Now
Money arrives weeks after the work, and about 10 hours a week go into phone calls about it.
After
The invoice goes out as soon as a job is marked done. Reminders go by email and messenger before the due date and after it. A client who still has not paid gets a call from an AI assistant and can switch to your bookkeeper during the call.
Benefit: about 10-12 hours a week, almost all of the phoning and the manual invoices, and invoices paid sooner.
Effort: settings in the accounting software you already use.
Risk: a reminder or a call to a client who has paid or is in a dispute. Reminders pause for disputed invoices, and your bookkeeper checks the list before the first run. The AI assistant says who it is in its first sentence.

2Request inboxEffort: medium · coordinators

Now
Two people do the same reading and retyping for every email.
After
Common questions (prices, visit times, what a clean includes) get an automatic answer. Everything else the AI sorts (extra clean, complaint, schedule change, invoice question) and drafts the reply and the job entry. A coordinator checks and sends.
Benefit: about 20-25 hours a week of coordinator time, across both coordinators.
Effort: the drafts learn from your past replies, then run side by side with the team for a week. Confirm with your scheduling vendor that jobs can be created by import or a connector.
Risk, client data: emails hold client addresses and building access codes. Use a business AI plan that does not train on your data, and keep access codes in the scheduling tool, not in email.
Risk, wrong answers: automatic answers only for questions with one fixed answer. Everything else waits for a coordinator.

3Client, office and cleaners in one loopEffort: medium · coordinators and senior

Now
The office only knows a visit happened, or moved, when someone picks up the phone.
After
Built on your current scheduling tool. After each visit the cleaner taps one "done" button. If nobody taps by the expected time, an AI assistant calls the cleaner, who can switch to a live operator in the same call. The client gets "cleaning complete" automatically. When a visit moves, both the cleaner and the client get a message. No new app, no written reports, no phoning round.
Benefit: about 12-15 hours a week: phoning about visit changes and sick calls for the coordinators, checking visits and collecting reports for the senior. Clients also stop calling to ask "did anyone come?"
Effort: check that your scheduling tool supports a button and messages, and set an expected finish time for each site.
Risk: false alarms when a visit runs long, and messages to old phone numbers. Allow a margin before the call goes out, and clean the contact list first.

What not to do yet

  • A new scheduling platform. Moving 24 people on client sites to a new app costs more than it returns.
  • A third coordinator. The three changes free more hours than a new hire would add. Measure office hours for four weeks after improvement 1 goes live, then decide.

First step

Improvement 1, in two stages

List last quarter's invoices with the days each took to be paid. Then switch on invoices on completion and email reminders.

Estimated effort: about 4 to 6 hours of setup. Cost is these hours at the rate of whoever builds it: your team, a freelancer or me. No running cost: it is part of your current plan.

Stage two, a month later: messenger reminders and AI calls to clients who still have not paid. They add a small running cost per message and call, estimated before you decide.

Implementation is quoted separately. You decide who builds it: your team, a developer or me.

When this works, next steps

  • Time from the "done" button goes straight into timesheets for payroll.
  • A photo after cleaning goes with the "done" button, and the client rates the visit in one click; a low rating alerts the manager.
  • A cleaner reports sick by button or in the call, and the system suggests a replacement from those who are free.
  • A weekly dashboard for you: visits, late arrivals, complaints, unpaid invoices, margin per client.