Studio
A Chief AI Officer's First 90 Days: A 30/60/90 Day Plan
By the llcloud team 6 min read
Quick answer
In the first 30 days a Chief AI Officer audits every AI tool, data flow and idea in the company. Days 31 to 60 bring quick wins and a written AI usage policy. Days 61 to 90 deliver a ranked roadmap and the first AI project shipped to real users. llcloud offers fractional CAIO services that follow this plan, part time and sized to your company.
A new Chief AI Officer arrives to the same pressure everywhere: leadership wants results, staff want to know what is allowed, and someone has already signed up for three AI tools. The first 90 days decide whether the role becomes a working program or another set of slides. This is the plan we follow, in the order that works, whether the CAIO is full time or fractional.
What does a Chief AI Officer do in the first 90 days?
They learn what AI the company already uses, make it safer and more useful straight away, then set a direction and prove it with one shipped project. In short:
| Phase | Focus | Ends with |
|---|---|---|
| Days 1 to 30 | Audit tools, data, ideas and people | An AI inventory and a findings summary |
| Days 31 to 60 | Quick wins and a written policy | Approved tools, an AI usage policy, first training |
| Days 61 to 90 | Roadmap and first project | A ranked roadmap and one AI project live with real users |
Days 1 to 30: what should the AI audit cover?
The audit should cover four things: tools, data, ideas and people. The goal is a complete, honest picture before any big decision.
Tools
List every AI tool in use, paid or free, approved or not. Ask each team, check expense reports and software subscriptions, and look at which browser extensions and integrations are switched on. Expect to find more than anyone thought. US federal agencies are required to keep an annual inventory of their AI use cases for the same reason (OMB M-24-10). You cannot govern what you have not listed.
Data
For each tool, find out what data goes in, where it is stored, whether the vendor may use it for training, and who can see it. Pay special attention to customer data, personal data and anything confidential. This is often where the most urgent problems sit.
Ideas and past attempts
Collect the AI ideas people already have, and ask what has been tried before and why it stalled. Old pilots are useful evidence. They show where the data is messy, where buy-in was missing, and which problems people still care about.
People and goals
Talk to leadership about what they want from AI, and to staff about their most repetitive work. Meet the CTO or lead engineer early. Our guide on CAIO vs CTO covers how to split the work cleanly.
The phase ends with a short written summary: what is in use, what is risky, what is promising and what leadership wants.
Days 31 to 60: which quick wins and rules come first?
Fix the biggest risks, approve a small set of tools, and deliver two or three quick wins people notice. Then write it all down as a policy.
Close the obvious risks
If the audit found customer data going into an unapproved tool, deal with that first. Move the work to an approved tool with acceptable data terms, or stop it. Do it calmly and explain why, since people were usually just trying to work faster.
Pick quick wins
Choose a few small improvements that save real time within weeks and carry little risk. Good candidates are repetitive writing, summarising long documents, and first drafts that a person always reviews. Measure the time saved before and after, even roughly, so you have evidence.
Write the AI usage policy
Put the rules in plain language: which tools are approved, what data may never go into them, when a human must review AI output, and who to ask when unsure. Keep it short enough that people read it. Our guide to an AI usage policy for a small business walks through what to include. The voluntary NIST AI Risk Management Framework is a good structure for the risk side as the program grows.
Start training
Run short, practical sessions built around the approved tools and the quick wins. In the European Union, companies that use AI systems also have a legal duty to take measures for a sufficient level of AI literacy among the staff who operate them (EU AI Act, Article 4), so training is more than a nice extra there.
Days 61 to 90: how do you build the roadmap and ship the first project?
Rank the opportunities from the audit by value and effort, pick one to ship, and get it in front of real users before day 90.
Build the roadmap
For each opportunity, write down the problem it solves, the expected benefit, the data it needs, a build or buy decision and a rough effort. Rank them and share the list with leadership. Be explicit about what you are not doing and why. A roadmap with ten top priorities has none.
Choose the first project well
The first project should be narrow, useful and measurable. Pick a problem people already feel, with data you can use safely, that can ship in a few weeks. Avoid the most ambitious idea on the list. Its job is to prove the program delivers.
Ship and measure
Design it, build it, test the output quality with real examples, and release it to a small group first. Decide up front how you will know it helps: time saved, fewer errors, or how many people keep using it after the first week. Then report the result to leadership honestly, including what did not work.
Set the rhythm
By day 90, the company should know how AI work runs from here: a regular review of the roadmap, a named owner for each project, a policy that gets updated, and a channel for new ideas. If you are still deciding whether this calls for a full-time hire, see when to hire a Chief AI Officer.
What mistakes should a new CAIO avoid?
The common ones come from moving out of order:
- Picking tools before the audit. You end up adding a fourth overlapping subscription.
- Banning AI outright. People keep using it, just where you cannot see it.
- Writing a long policy nobody reads. One or two pages that people follow beat twenty they ignore.
- Starting with the biggest project. A large first project that slips can sink trust in the whole program.
- Skipping measurement. Without a before and after, a good result is only an opinion.
How can llcloud run your first 90 days?
llcloud is an independent studio that builds apps, and we also work with companies as a fractional Chief AI Officer. Our engagements start with this plan, part time and sized to your company:
- AI strategy and roadmap: the audit, the priorities and the build or buy calls
- Building AI products: designing and shipping the first project, not just specifying it
- AI governance and data privacy: approved tools, the usage policy and data rules
- Team training: practical sessions built around your approved tools
We build and ship our own apps, with no ads, no account and data kept on the phone wherever we can. Asking where the data goes before anything else is how we approach every audit.
To start, write to support@llcloud.app with a few lines about your company and the AI tools you already use.
Sources
Questions
What should a Chief AI Officer do first?
Listen and take stock. The first weeks go into an audit: which AI tools staff already use, what data goes into them, what has been tried before, and where leadership wants help. Every later decision, from the policy to the roadmap, depends on knowing what is already happening, and most companies find more AI use than they expected.
Can a fractional CAIO follow a 90-day plan?
Yes. The plan is about order, not hours. A fractional CAIO works part time, so each phase may involve fewer days on site, but the sequence holds: audit, quick wins and policy, then roadmap and a first shipped project. The audit often moves faster with a fractional CAIO, because staff talk openly to someone from outside.
What is an AI quick win?
A small, low-risk improvement that saves real time within weeks, using tools the company can safely approve. Examples include a reviewed template for drafting support replies, summarising long internal documents, or cleaning up meeting notes. A quick win builds trust in the program and gives the CAIO evidence about what staff actually find useful.
What should be shipped by day 90?
One real AI project in front of real users, with a way to measure whether it helps. It does not need to be large. A narrow feature that works and is used teaches more than a big one still in testing, and it proves the roadmap is a delivery plan rather than a wish list.
Does llcloud run this 90-day plan for companies?
Yes. llcloud offers fractional CAIO services and starts engagements with this sequence: audit, quick wins, AI policy, roadmap and a first shipped project. We design and ship AI features ourselves, so day 90 ends with something working. Write to support@llcloud.app to discuss your company.