Answer for the first workflow you have in mind.
Your roadmap
What this measures: whether the company is ready to put a first AI workflow into weekly use, with an owner, real data and a way to check the results.
Where you stand
Foundations first.
Ready to scope a first pilot.
Ready to build.
Foundations firstReady to scope a pilotReady to build
of 12 readiness points
Held at this stage by the weakest link, whatever the points say.
The preconditions for a first pilot are not in place yet, which is a common starting point. The fastest route is an owner, one workflow and access to real records, before anyone chooses a tool.
Some of the preconditions are in place. What is missing is specific, and closing it turns a good intention into a pilot you can measure.
The preconditions are in place. Choose the workflow, measure it as it runs today and build the first version with the people who will use it.
Readiness profile
Six conditions for a first pilot. The weakest one sets how far a pilot can go, whatever the average says.
- Owner and budgetWeakest link
- One repeated workflowWeakest link
- Real data and accessWeakest link
- Useful output and testsWeakest link
- Time to practiseWeakest link
- Approvals and oversightWeakest link
Your answers as a list
AI adoption today
- Little or none
- Individuals
- One team
- Built in
With little AI use today, build practice time into the first weeks. People adopt what they have used on their own cases.
Individual use is a start. The next step is one shared way of working on one workflow.
Start with the people who already use it most: write down what they do, and turn it into the shared workflow.
One team already shares a tool. Measure what it changed before extending it.
AI is already part of your workflows, so the question is which one to extend next and how you will measure it.
Your next three moves
- Name one owner with a business outcome and a budget line. Without that person, pilots stall between teams.
- Pick one repeated workflow and write down how it runs today: the steps, the hand-offs, the time and the errors. That is the baseline you will measure against.
- Get approval for a representative set of real records and a route to reach them. Most first pilots stall on data access long before the model matters.
- Define what a useful output looks like and collect the common failures, so the first version can be tested against real cases.
- Give a small group time to practise in their normal work. Adoption starts there, on their own cases.
- Decide who approves data use, human review and ongoing operation before anything reaches production.
- Measure the chosen workflow as it runs today, so the before and after are real numbers.
- Build the first version with the people who will use it, and test it on past cases with known answers.
- Agree who owns and maintains it after launch, and document it so the team can run it without outside help.
Where to start
With demand as the constraint, start where selling time goes: account research, proposals and follow-up.
With delivery as the constraint, start in operations: documents, reporting and reconciliations, where rules do most of the work.
With technology as the constraint, pick a workflow the business can own without joining the IT queue, with the approvals agreed from day one.
Reads only
Account brief before each call
AI reads your CRM notes, past emails and public sources and drafts a one-page brief. The seller checks it before the call.
- First version
- 2 to 3 weeks
- Measure
- Preparation time per call
See how it works ↗Drafts, a person acts
Follow-up that sets the next step
From the call notes, AI drafts the follow-up with the agreed next step and who owns it. The seller edits and sends it the same day.
- First version
- 2 to 3 weeks
- Measure
- Days from call to an agreed next step
See how it works ↗Drafts, a person acts
Proposal first drafts
AI assembles a first draft from your catalogue, past proposals and the call notes. Prices come from your rules, and a person approves.
- First version
- 3 to 4 weeks
- Measure
- Days from call to proposal
Reads only
Find the handoff worth fixing
Follow one order from sale to delivery and time each handoff. The longest wait is usually the first place to automate.
- First version
- 2 to 3 weeks
- Measure
- Days an order waits between teams
See how it works ↗Drafts, a person acts
Document intake with an exception queue
AI reads incoming documents and extracts the fields, rules check them, and people handle only the exceptions.
- First version
- 3 to 4 weeks
- Measure
- Handling time and errors per document
Drafts, a person acts
Recurring reports from existing data
Rules calculate the numbers from your systems, AI drafts the commentary, and a person edits before anything is sent.
- First version
- 3 to 4 weeks
- Measure
- Hours per reporting cycle
Reads only
Answers from your own documents
People ask in plain language and get an answer from your documents, with the source beside it. Nothing in your systems changes.
- First version
- 2 to 3 weeks
- Measure
- Minutes to a verified answer
Drafts, a person acts
Triage for incoming requests
AI classifies each request and routes it to the right person with a suggested reply. People keep the decision.
- First version
- 2 to 3 weeks
- Measure
- Time to first response
Writes, with a log
A workflow the business owns
Built on approved tools within IT’s guardrails and run by the team that uses it, so improvements stop waiting in the IT queue.
- First version
- 4 to 6 weeks
- Measure
- Requests closed without an IT ticket
See how it works ↗Read first, write later
- ReadAI reads, searches and summarises. Nothing in your systems changes.
- DraftAI prepares the draft, and a person checks it and acts.
- WriteAI updates records itself, once the checks have held for a few weeks.
Before AI writes anything, agree who holds the authority, log every action, plan how a person corrects a mistake, keep the data in Europe and check the use case against the EU AI Act risk categories.
The first 90 days
- Day 30
- An owner, one workflow with its baseline, and approved access to real records.
- Day 60
- The first version defined and tested on real cases, with the failures written down.
- Day 90
- A small group using it in normal work, with the before and after measured.
- Day 30
- The open gaps closed and the workflow measured as it runs today.
- Day 60
- A first version built with its users and tested on past cases.
- Day 90
- In daily use by a small group, with the result measured and a decision on what comes next.
- Day 30
- The first version built and tested against cases with known answers.
- Day 60
- In daily use by a small group, with the before and after measured.
- Day 90
- Documented and handed to its owner, with the second workflow chosen from the results.
For your case
With several business units, start in one and make the result easy to copy.
For a portfolio, start in one company or in the management company, then turn what works into a playbook for the rest.
At your size, much of this can run in-house with the templates in the resource library.
At your size, start with one business unit and one workflow before anything company-wide.
Explore the resource library ↗
Generated on this page from your answers with fixed rules, so the same answers always give the same roadmap. Nothing is stored unless you send it.