AI Readiness Assessment for Law Firms and RIAs
Most AI readiness assessments measure how ready people feel. The question that decides whether AI pays off is simpler: where is the repetitive work, and how many hours does it cost? A scored checklist, a way to find use cases from evidence, and a 30/60/90 plan.
An AI readiness assessment should tell you where to start, not how you score. Most assessments are questionnaires or consulting engagements. They ask leaders how ready the firm is, run workshops to list use cases, and produce a maturity score. That is useful for alignment, but it measures opinions. The answer that decides whether AI returns anything is a work inventory: which repetitive tasks exist, how often they happen, how long each one takes, and which systems they touch. Get that right and the rest of the assessment is a short checklist.
What an AI readiness assessment should answer
For a law firm or an RIA, five questions cover it. The first matters more than the other four combined.
- The work. Which repetitive work exists across the firm, how many hours a month it costs, and which team owns it.
- Access. Whether AI can reach the systems that work lives in: the CRM, the document management system, the practice management or portfolio system, email, e-signature, the custodian portals.
- Policy. Whether there is a written rule for which tools are approved, what client data may go into them, and what has to be reviewed by a person.
- People. Who owns each automation once it runs, and who decides what gets automated next.
- Measurement. How the firm will know what AI delivered, in units leadership will accept.
The AI readiness checklist
Score one point for each yes. Do it in one working session with operations, IT and compliance in the room, because each group will answer a different row. The score is a starting point for the conversation, not a benchmark.
| Area | Ask | Yes if |
|---|---|---|
| Work | Do we have a list of our repetitive processes? | It names each process, its team and its owner |
| Do we know how often each one happens? | Counts come from system activity, not memory | |
| Do we know how long one occurrence takes? | Minutes per item are measured or timed, not guessed | |
| Can we rank them by hours a month? | There is one ranked list leadership has seen | |
| Access | Can AI reach the systems the work touches? | Each system has an API, a supported connector or a permissioned browser session |
| Can access be scoped and revoked centrally? | Credentials are not living in personal scripts | |
| Policy | Is there a written AI acceptable use policy? | It lists approved tools, client data rules and review rules |
| Does every agent have to be registered? | Each one has an owner, a team and a place it runs | |
| People | Is there an owner for the AI program? | One person answers for what is running and what is next |
| Can operations staff build, not just request? | Building does not require an IT project for every change | |
| Measurement | Do we know our baseline hours? | The hours each process costs today are written down |
| Will we report work done, not usage? | Reports count items, documents and hours, not logins |
Reading the score. Nine or more: you are ready to automate the top of your list now. Five to eight: fix the gaps in access or policy alongside the first one or two automations rather than before them. Four or fewer: start with the work rows, because without a ranked list every other answer is guesswork.
If the policy rows are the gap, our AI acceptable use policy template covers approved tools, client data, human review and agent registration. If measurement is the gap, see AI usage tracking for the five measures leadership asks for and how to count them honestly.
How to identify AI use cases from evidence, not workshops
The usual way to find AI use cases is a workshop: people list the tasks that annoy them, and the loudest ones win. The list is real, but it is skewed toward what people remember and what they dislike, not toward what costs the firm the most time. Shared inboxes, end-of-month tie-outs and form reviews that nobody owns end up undercounted.
The evidence-based version has three steps.
1. Measure frequency and duration
For each repetitive task, count how often it happens over an observed window, and how long one occurrence takes. Frequency comes from activity already in your systems: records created, documents filed, emails handled, forms submitted. Duration comes from the same activity or from timing a handful of real occurrences.
2. Rank by hours a month
Frequency times minutes per occurrence gives hours a month. A task that happens 400 times a month at 10 minutes each costs about 67 hours; one that happens 20 times at 45 minutes costs 15. The first is almost always the better place to start, even if the second is more annoying. Read the figure as the cost of the process, including everyone who touches it, and as a way to rank, not a forecast.
3. Route each one to the right tool
Not everything at the top of the list should be handled the same way.
- High-volume, high-accuracy work such as intake, filing, new account paperwork, form review and reconciliation should run the same way every time. It suits automation that runs as deterministic code, with any model step scoped and logged.
- Individual, judgment-heavy work such as drafting a client update or summarising a file suits an assistant such as Claude or Microsoft 365 Copilot, used by the person whose judgment it is.
Law firms can see how this plays out by process in legal automation: what to automate first. RIAs will recognise most of the list in RIA operations challenges.
Questionnaire, consultant or evidence
There are three common ways to run an AI readiness assessment. They are not exclusive, but they answer different questions.
| Questionnaire | Consulting engagement | Evidence-based inventory | |
|---|---|---|---|
| What it measures | How ready leaders say the firm is | What staff describe in interviews and workshops | What the work actually is, from activity in your systems |
| Output | A maturity score | A report and a roadmap | A ranked list of repetitive work, by hours a month, with evidence |
| Time | An afternoon | Weeks to months | A few weeks of observed activity |
| Cost | Usually free | Quotes in the six figures are common | Software, often with a free tier |
| Blind spot | Opinions, not work | Work people forget to mention | Work that leaves no trace in a connected system |
| Best for | Starting the conversation | Change management and executive alignment | Deciding what to automate first |
A 30/60/90-day AI implementation plan
Days 1 to 30: find the work
- Complete the checklist above and name an owner for the program.
- Build the work inventory from evidence and rank it by hours a month.
- Write or adopt an acceptable use policy, including agent registration.
- Record the baseline hours for the top five processes.
Days 31 to 60: run the first automations
- Automate the top two or three high-volume processes end to end.
- Hand the judgment-heavy items on the list to the assistant your team already uses.
- Report weekly: items completed, documents handled, hours returned.
Days 61 to 90: make it a program
- Add the next processes from the ranked list, team by team.
- Show leadership the split between work fully automated, in progress and still manual.
- Measure hours returned against the baseline, and re-rank the list.
Where Caddi fits
Caddi turns the readiness assessment from a document into a working list.
- Caddi Discover reads the tools you connect, read only and under the access you already have. It maps the tasks, hours and documents behind the work, and ranks the repetitive work by hours a month, with the evidence behind each finding so a skeptic can open one and check the counts. It is a way to rank findings and start a conversation, not a forecast.
- Automate or export. Each finding goes to the best tool for the job: high-volume, high-accuracy work runs in Caddi Automate as deterministic code, and individual, judgment-heavy work can be exported to Claude as a skill or to Microsoft Copilot as an agent.
- Caddi Govern measures every Caddi run and the work Caddi hands to Claude and Copilot, and reports the hours returned against the baseline Discover estimated. It does not see a person's ad hoc chats or automations built elsewhere without Caddi.
If you would rather talk it through first, a free AI consultation walks through your firm's list with you. To turn the ranked list into a plan for next year, see plan your 2027 AI adoption roadmap.
Caddi
See how Caddi AI Agents can find your most repetitive back-office work and automate it across every tool you use
Frequently asked questions
What is an AI readiness assessment?
An AI readiness assessment tells a firm whether it can put AI to work, and where to start. A useful one answers five things: which repetitive work exists and how many hours it costs, whether AI can reach the systems that work lives in, whether there is a written policy for client data and review, who will own each automation, and how results will be measured. Most of the value is in the first answer, the work inventory.
How do law firms and RIAs know where to start with AI?
Start from the work, not the tools. Measure how often each repetitive task happens and how long one occurrence takes, multiply to get hours a month, and rank by that number. Then route each item: high-volume, high-accuracy work such as intake, filing, form review and reconciliation suits automation that runs the same way every time; individual, judgment-heavy work suits an assistant such as Claude or Microsoft 365 Copilot.
What is the difference between an AI readiness assessment and an AI maturity assessment?
A readiness assessment asks whether the firm can start and where: the work inventory, access to systems, policy and ownership. A maturity assessment scores how far an existing AI program has come, usually across stages from experimentation to firm-wide production. Firms early in their AI program get more from a readiness assessment; firms with live automations get more from measuring what those automations deliver.
Do we need an AI strategy consultant for a readiness assessment?
Not necessarily. Consultants are useful for change management and executive alignment, and quotes in the six figures are common for a process review. But a work inventory built from interviews and workshops records what people remember doing. An evidence-based assessment reads the activity already in your systems, which is faster, cheaper and harder to argue with. Many firms use the evidence first and bring in advisers for the parts that need people.
How long does an AI readiness assessment take?
A questionnaire takes an afternoon and a consulting engagement often takes weeks to months. An evidence-based work inventory depends on how long a window of activity you observe; a few weeks of real activity is usually enough to rank the repetitive work by hours and pick the first processes. The checklist in this guide can be completed in a working session with operations, IT and compliance in the room.
How does Caddi Discover help with an AI readiness assessment?
Caddi Discover reads the tools you connect, read only and under the access you already have, maps the tasks, hours and documents behind the work, and ranks the repetitive work by hours a month with the evidence behind each finding. Each finding can be automated in Caddi or exported to Claude as a skill or to Microsoft Copilot as an agent. The estimate is a way to rank findings and start a conversation, not a forecast; Caddi Govern then measures the hours actually returned against it.