Sample report for an invented accounting firm. See the leader brief. Get your own free AI assessment. Other samples: Wealth management practice · Law firm · Trades business · Engineering firm
Sidekick Orchestration Sidekick Orchestration
Free AI assessment

Sample report · invented firm

Where AI fitsSalal Creek Accounting

// AI assessment Accounting firm

At a glance

Where to start

Client service: missing documents chased for you in busy season.

What it costs today

About $12,100 a month.

Book 30 minutes with Chase

Book 30 minutes with Chase: cal.com/chase-sidekick/intro

The two-minute version

Where you stand
A few people use AI on their own. No approved tool or written rule yet.
The main risk
Information about your clients, such as clients' financial and account information, going into personal AI accounts your firm can't see.
A sensible first move
Partners first: one approved tool your firm controls, one page of rules, and a month on client service.
What you'd report at day 90
Hours on chasers, and days from first request to documents in, before and after.

Top opportunities, in your order

Opportunity 1

Missing documents chased for you in busy season

You ranked client service first. Seniors stop writing chasers one by one. Each Monday, notices for every file missing items are drafted from your system's status fields, for one person to approve.

Seniors stop writing chasers one by one.

Draws from
Standard item names and status fields from your practice-management system. The email gives only a count.
Reviewed by
The tax manager approves the Monday batch; sensitive files go to the partner.

Opportunity 2

Engagement letters drafted for partner review

You ranked proposals and quotes second. Renewal letters are mostly last year's, and the small changes are where mistakes hide. A flagged first draft leaves the partner free for fees, scope and acceptance.

Renewal letters are mostly last year's, and the small changes are where mistakes hide.

Draft letterExample

FIRST DRAFT FOR MARION LEUNG. NOT SENT.

Engagement letter first draft: note to the partner

Client
Kestrel Bay Landscaping Ltd.
Year end
June 30

Same as last year:

  • Compile the annual financial statements
  • Prepare and file the T2 corporate return

Changes since last year:

  • New: quarterly GST returns, added to the client's services in your system on August 12
  • New template paragraph: how the firm uses AI tools, with a line for the client's consent

Flagged for you:

  • Fee: blank for you to set. Last year's fee and hours attached.
  • Scope: GST returns in this letter or a separate one?
  • Client continuance check: yours to complete before signing.

Standard wording, including that a compilation is not an audit or review, is unchanged.

Draws from
Your approved template and the client's fields in your practice-management system (services, year end, last fee and hours).
Reviewed by
The engagement partner signs before anything is sent.

Opportunity 3

Partner numbers on the first business day

You ranked leadership reporting third. A one-page summary shows the partners what needs a decision.

A one-page summary shows the partners what needs a decision.

BriefExample

Monthly partner summary

PARTNER SUMMARY · October · Partners only
From your practice-management export on the first business day. Each figure links to its row.

Needs A Decision

  1. Write-offs proposed: $14,800 across 6 files. Largest: Castillo personal tax, $4,100. Approve, bill, or talk to the client?
  2. WIP over 90 days: $61,300, of which Kestrel Bay's year end is $18,900.

Billing

  • Billed in October: $412,000. Collected: $377,000.
  • Receivables over 60 days: $52,600 across 11 clients. Largest: Harbour Signs, $9,200.

Capacity

  • Files waiting on a manager's review: 23. Oldest: 16 days.
  • Staff overtime in October: 41 hours, mostly the corporate group.
Draws from
A scheduled export from your practice-management system in one agreed layout.
Reviewed by
Your finance lead checks it against the system; partners decide write-offs and follow-ups.

Where else AI fits

  1. Operations and projects

    Today
    Status checked file by file.
    Set up once
    Read-only status fields and due dates.
    Then runs
    A Monday page: where files are, what's stuck.
    You still decide
    Partners decide any reassignment.

  2. Finding information

    Today
    Seniors answer the same questions all season.
    Set up once
    Your manual and checklists in one shared library.
    Then runs
    Staff ask and get answers with the section linked.
    You still decide
    Managers keep the manual current and settle anything unclear.

What this time costs today

You told us 20 to 50 hours a week across your team on client service; we used 35. That's time spent today, not a promised saving.

  1. 35 hours × $80your numbers
  2. $2,800a week
  3. $12,100a month
  4. $146,000a year

35 hours × $80 = $2,800 a week ≈ $12,100 a month ≈ $146,000 a year

Your data and AI

You hold clients' financial and account information, confidential client business information and personal information about your clients. You're not early. Others who hold this kind of information handle it in layers: limit what AI can reach, limit what it can do, write the rules down, and have a person read everything that leaves.

In this order

  1. Step 1Status: Done

    Accounts first

    Everyone uses two-step sign-in, you told us. Worth a quick check with IT. Keep it on for every account AI connects to.

    Everyone uses two-step sign-in, you told us. Worth a quick check with IT. Keep it on for every account AI connects to, keep passwords in a password manager, and keep reminding people what phishing looks like.

  2. Step 2

    Know what's sensitive and where it lives

    List what's restricted, then keep it out of reach.

    List what's restricted, in plain words, and where each item lives. Start from the list you gave us, above. Many AI connections can reach everything your login can open, so decide one system at a time rather than file by file. Keep restricted files out of what AI can reach, or behind a separate account, and confirm it with a quick test. IT can check who can open what before anything is connected.

  3. Step 3

    Then choose the tool

    You run Microsoft 365, so Copilot is the natural place to start: it works inside the sign-in and permissions IT already manages. Check your agreements first; one that bans outside processing means a private model. A ChatGPT or Claude business plan is the alternative.

    Choose the tool last, after accounts and your restricted list. What decides it: your agreements with clients, any rule on where data is stored, the systems you already run, and what your people will actually use. Once nothing rules a tool out, a decent tool that gets used every day beats a perfect tool that never gets picked up. Mandate one people dislike and they drift back to personal accounts with no oversight.

Your tool options

You run Microsoft 365, so Copilot is the natural place to start: it works inside the sign-in and permissions IT already manages. Check your agreements first; one that bans outside processing means a private model. A ChatGPT or Claude business plan is the alternative.

  • Copilot in Microsoft 365. Works inside the sign-in and permissions IT already manages; Microsoft says it only shows people what they can already open.
  • A ChatGPT or Claude business plan. OpenAI and Anthropic each say they don't train on business data by default. Both connect to Outlook and SharePoint with admin approval.
  • If data has to stay in Canada. Microsoft stores Copilot prompts and responses in Canada when your Microsoft 365 data location is Canada, which is set at sign-up. OpenAI offers Canadian storage at rest for eligible new ChatGPT Enterprise workspaces. For Claude, Canadian residency runs through cloud platforms; ask Anthropic about its own plans. Storage in Canada doesn't always mean processing in Canada, so get each commitment in writing.
  • A private model on your own hardware. Open models can run on an office computer, so data can stay in the building. But they're usually less capable, someone must patch and secure the machine, and connecting email, calendar and other systems is slow work. Rarely the right first step; the right one when an agreement or rule bans outside processing outright.

Settings to ask for

  • Training off, in a business workspace the firm owns. Use a business workspace so the firm owns the accounts and data, training is off by default, and you control which apps people connect and can remove access the day someone leaves.
  • Connections need approval. In Microsoft 365, IT can require admin approval before anyone connects an AI app to email or files. That way every connection is a decision someone made.
  • Web search off for sensitive work. Turn off Copilot web search for the people who handle work for clients. It's one admin policy, set per user or group, and people can't turn it back on.
  • Drafts only, to start. Where the tool allows it, leave send and edit permissions off to start. AI prepares the email and a person sends it. That also limits the damage if an email or web page hides instructions aimed at the AI. Check the actual connection, because instructions alone aren't a lock.
  • Keep sensitive work out of chat memory. Many AI tools remember past chats. For sensitive work, use a temporary chat or turn memory off. On some personal plans, rating a reply can still send that conversation for training, even with training off.

CPABC's review of its Code of Professional Conduct and AI tools (June 2025) asks registrants to keep confidential information out of AI queries, to document the tool, the input and the output, and says supervising staff extends to AI tools.

What stays with you

  • Tax positions and advice. Salary or dividends stays your professional judgment.
  • Assurance conclusions. Evidence, judgments and the opinion stay with the engagement team.

What could go wrong, and how to prevent it

  1. Information your clients trust you with

    How it's prevented
    Put written rules in the AI's settings from day one, and widen what it can reach one system at a time as it earns trust.
    Do this week
    Ask IT two questions: who can open what in your shared files, and whether anyone uses a personal AI account for work.

  2. Staff using unapproved tools

    How it's prevented
    Give people an approved tool that beats the free one, and block unapproved AI apps on firm devices.
    Do this week
    Send a two-line note: no information about clients in personal AI accounts. Then ask, with no penalty, what people have been using.

  3. Wrong answers going to clients

    How it's prevented
    The responsible person stays accountable, so check citations, figures and sources: a public tracker lists over 2,000 court decisions worldwide involving AI-invented content.
    Do this week
    At your next team or partners' meeting, agree: nothing AI-drafted leaves until the responsible person has read all of it.

Rules here were checked against their sources in September 2026. Your partners confirm how they apply to your firm.

What it takes

  • Partners first

    You and up to five partners use it first: about 3 hours of yours over two weeks, then 30 minutes a week. Others follow.

  • Turning it on

    IT does it, with Chase or from his checklist: partners first, web search off for client work, data storage checked, file permissions tidied.

  • A policy that fits your quality system

    Your one-page AI policy fits CSQM 1, the quality-management standard for your assurance and compilation work: which tools, what goes in, who reviews.

A first project for you

Try this yourself, before your team does.

Use the firm's business AI account, on templates, past work or your inbox, with names taken out.

Starting instruction

Here are three updates our firm has sent before, with names and details taken out. Write one reusable update template in the same voice: what's done, what's coming up with dates, and what we need from them. Mark every blank with [FILL]. [paste three past updates, names and details removed]

How to check it

Read the template against the three originals. If you'd send it with only the blanks filled in, it's ready to show the other leaders.

If it saves you time, that's the case for your partners.

How Chase would help

  • Solo for Execs

    Start with yourself. AI leadership in 30 days: AI set up around your own work, plus what you need to lead it in your firm and answer the questions your partners, team and clients ask.

    CAD 2,900 once

  • Fractional leadership

    Chase leads AI adoption with your leadership team for 90 days: first two weeks meeting your leaders and showing AI on your firm's own templates and past work (not client files), then agreed measures reviewed every 30 days. Teams come in once leaders use it.

    From CAD 6,000 a month

    Scaled to how many leaders and teams are involved.

Who sets it up
Chase works with your IT provider directly, or gives them a checklist, your choice. Your IT provider turns it on.
If it isn't working
Fractional: measures agreed up front and reviewed every 30 days; if a milestone slips because of us, we keep working at no extra fee until it's delivered.
You keep everything that's built.

Chase signs your confidentiality agreement and never logs in to your systems or asks for client files. He doesn't give accounting, tax or professional advice.

30 minutes with Chase to walk through your report and decide where to start.

You ranked client service first and worry about client data. We'd start there.

Book 30 minutes with Chase

Book 30 minutes with Chase: cal.com/chase-sidekick/intro