Notion for Accountants: Firm Wiki, Client Trackers and Notion AI

Yes, a small accounting firm can run its procedures wiki and client tracker in Notion on the Plus plan ($10 per member a month billed yearly), with no AI at all. Notion AI is worth paying for only if staff will use it daily: it comes with Business at $20 per member, and every member pays it.

Most firms get the order wrong. They buy the AI tier, then discover nobody has written the procedures it is supposed to answer questions about. An assistant that searches a half-empty wiki gives half-empty answers. So this tutorial builds the two foundations first, a wiki that people trust and a tracker that tells you what is overdue, and only then puts a price on the AI layer.

Follow me on Instagram@sagnikteaches

What goes in Notion and what stays out

Notion is a flexible workspace of pages and databases (tables whose rows are pages in their own right). That flexibility is also the risk: a firm that stores everything there ends up with a second filing system nobody can police. Draw the line before you build anything.

Connect on LinkedInSagnik Bhattacharya

The right-hand column is where a leak does real damage, and it is also the column Notion does least well. A tracker that holds only a client code, a service and a date can be shared with every member of staff without a second thought. If you want a view of how engagement terms should be drafted, the engagement letters tutorial covers that side separately.

Subscribe on YouTube@sagnikteaches

The firm wiki, built in an afternoon

A wiki fails when it is a dumping ground, so give it a shape with four top-level pages and one rule: every procedure has an owner and a review date.

1. Procedures. One page per repeatable job: year-end accounts, payroll run, new client onboarding, bank feed reconciliation, filing-deadline escalation.
2. Checklists. Short, tickable versions of the procedures, kept separate so a preparer can copy one per job.
3. Templates. Email wording for chasing records, engagement-change notices and the client-facing "what we need from you" list.
4. Firm policies. Working from home, file naming, retention, who may approve what.

Make Procedures a database rather than a folder of loose pages. Give it these properties: Name, Service line (select), Owner (person), Last reviewed (date), Status (Draft, Current, Under review) and Review due (a formula that adds 12 months to Last reviewed). Then create a view called "Overdue for review" filtered to Review due before today. That single view is what stops the wiki rotting. The two formulas you need, using functions from Notion's formula help page:

Review due:        dateAdd(prop("Last reviewed"), 12, "months")
Days to deadline:  dateBetween(prop("External deadline"), now(), "days")   (used in Jobs, below)

Notion also has a built-in wiki. On a standalone page, open the three-dot menu, then More, then Turn into wiki; the page gains Home, All pages and Pages I own views, and each page in it has an owner and a verification that can run until a date you choose. When a verification expires, Notion notifies the owner in their inbox and by email, which does the same job as the review formula without anyone opening a view. Verifying pages that sit outside a wiki is a Business and Enterprise feature (the pricing page calls it "Verify any page"); the help page does not say which plans include verification inside a wiki, so on Plus check the Verification option on one wiki page before relying on it, and keep the Status and Last reviewed properties above as the fallback.

A filled-in procedure page

Here is how one page might read for a six-person firm (illustrative wording; adapt it to your own method).

Procedure: Onboarding a new limited-company client
Service line: Compliance | Owner: Practice manager | Status: Current
Last reviewed: 2026-09-14 | Review due: 2027-09-14

Trigger: engagement letter signed and returned.
1. Create the client row in Clients (code, service lines, partner, preparer).
2. Send the "what we need from you" list (template: Records request, new client).
3. Request prior-year accounts and the previous accountant's handover letter.
4. Run identity and risk checks in the compliance system; record the date here, not the result.
5. Add the first three deadlines to Jobs with owner and internal target date.
6. Book the 30-minute kick-off; add notes to the client row.

Done when: all six steps ticked, partner has initialled the risk assessment.
Common stall: handover letter not received after 10 working days. Escalate to the partner.

Note that step 4 records the date of the check and never the result. That habit keeps sensitive detail out of Notion and still lets a reviewer see that the step happened.

The client tracker: three linked databases

One giant table with fifty columns becomes unreadable by the second month. Use three small databases connected by relations (a property that points a row at a row in another database).

The Jobs database does the work. Its Status property should have a small, honest set of values: Not started, In progress, With client, With reviewer, Ready to file, Filed, Closed. Resist adding more. Every extra status is a decision someone must make and nobody will make it consistently.

Views that answer one question each

• This week by deadline. Board or table, filtered to External deadline within seven days and Status not Filed or Closed, sorted ascending.
• My jobs. Filter Owner to "Me" so each person opens one page and sees only theirs.
• Waiting on client. Status is With client, sorted by how long it has sat there. This view is the one partners actually read.
• Review queue. Status is With reviewer, grouped by the reviewing partner.

Add the Days to deadline formula shown earlier, plus a second formula that turns short deadlines red: if(prop("Days to deadline") < 10, style("Late", "red"), "") (Notion's style() function accepts colour names). A preparer who sees a red "Late" next to "4" acts; a preparer who sees "14 March" in grey does not.

Worked example: a six-person firm and 140 clients

Say a six-person firm has 140 active clients: 90 small companies, 30 sole traders and 20 landlords. Each company has three recurring jobs a year (accounts, an annual filing, a payroll or bookkeeping cycle), sole traders have two and landlords one. That gives roughly 270 + 60 + 20 = 350 recurring jobs a year, plus ad hoc work.

Setting it up takes one person about a day: half a day to import the client list from a spreadsheet into Clients (Notion imports CSV files), and half a day to generate the recurring jobs by duplicating a Job template per client. Maintenance afterwards is the part that matters. If preparers update Status as they work, it costs seconds per job. If they do it weekly from memory, the tracker is wrong by Wednesday. Agree that Status is changed at the moment the job moves, and have the practice manager check the Waiting on client view every Monday for ten minutes.

After one filing season, the useful measure is simple: how many jobs reached their external deadline still in "With client"? Count them from the view. That number tells you whether your chasing process works, which is a separate question from whether the tracker works. The real examples tutorial shows what firms do with that kind of finding.

Access settings that matter for a firm

Notion's help centre says Notion AI respects existing permissions: it cannot read a page the person asking has no access to. That makes your permission set-up the real control, so get it right before any AI is switched on.

• Teamspaces. The pricing page lists open and closed teamspaces on the lower plans and private teamspaces on Business. Put partner-only material (fee reviews, staff matters) in a private teamspace.
• Database permissions. Business adds granular database permissions, which let you limit who may edit the Clients database while everyone can read it.
• Guests. Keep clients out of the workspace. If you do invite a guest, share a single page, not a database that contains other clients' codes.
• Page history. Pricing shows 7 days on Free, 30 on Plus and 90 on Business. A wrongly deleted procedure is recoverable only inside that window.

Notion's help centre also states that on Free, Plus and Business the AI provider keeps data for 30 days or fewer and then deletes it, while Enterprise offers zero data retention for most models. If your engagement letters promise no third-party retention at all, that sentence is the one to show your partners before you decide. For wider questions about what firms should check before any AI tool touches client information, see the AI risk and error checks tutorial.

Pricing Notion AI against what it replaces

Notion's pricing page and its credits help pages set out the AI items below (as of November 2026). The plan boundaries are what matter for a firm.

Standard and Select credits are per-member allowances: they reset monthly, do not roll over and cannot be pooled or moved between people. Notion's own estimates put a simple task (polishing an email, finding decisions in meeting notes) at about 5 Standard credits on Auto, its model router, a medium task (tracking task progress, writing a work report) at about 10 and a complex one at about 25, with hand-picked top models costing several times more. One catch for firms: Business is a workspace plan, so every member pays the Business seat, not only the people who will use the AI. Seat prices move, so confirm them on the pricing page on the day you decide, then run the sum.

Extra cost of AI per month = (Business seat - Plus seat) x every member of the workspace
Break-even hours = Extra cost / hourly cost of the people using it

Illustration (yearly-billing list prices, six-person firm):
  ($20 - $10) x 6 members          = $60 a month
  Cost of a preparer's hour        = $35
  Break-even = 60 / 35             = about 1.7 hours saved per month across the firm

Credit check for one heavy user:
  40 medium tasks a week x 10 credits x 4.3 weeks = about 1,720 Standard credits
  Allowance 1,500, so this person runs short in the last days of the month
  unless an owner sets a Notion credit limit

Put that way, the seat cost is small against the time. The real question is whether six people will each genuinely save a quarter of an hour a month, and the honest answer depends on what you ask the AI to do. The credit check matters less for a firm than the seat sum: most preparers will use the agent a handful of times a day, well inside 1,500 credits, but a partner who runs every meeting through AI notes after 1 December adds about 10 credits an hour. The jobs below are where a firm with a decent wiki tends to see value.

Five jobs to test during the trial

1. Ask the wiki a question. A new starter types: "What do we do if a client's records are more than ten working days late for a year-end?" The agent should answer from the procedure and cite the page. Test it by asking something whose answer you know. If it answers from general knowledge instead of your page, your wiki is missing that rule. That is a useful discovery rather than a failure.

2. Turn meeting notes into tasks. Paste a kick-off call's notes onto a client page and prompt:

From these meeting notes, list every action with an owner and a date.
Mark anything without a clear owner as "Unassigned".
Do not invent dates. Output a table: Action | Owner | Date | Source line.

A realistic output (illustrative) is a five-row table with two rows marked Unassigned. What you fix is the dates: the agent will correctly refuse to invent them, and you add the real ones. The Unassigned rows are the value, because those are the actions that usually fall through.

3. Autofill a property. The pricing page lists autofill among the Notion Agent features. On the Jobs database, an autofill property can summarise a long Notes field into a one-line "Next action". Check twenty rows by hand before trusting it. Where notes are thin, the summary will be confidently vague.

4. Draft the chaser email. From the Waiting on client view, ask for a polite reminder that lists the specific records outstanding. Use "[first name]" in the draft and have a person check the list against the job before sending.

5. A weekly Custom Agent. Ask a Custom Agent to post, every Monday, a summary of jobs due in the next 14 days that are still in With client. Custom Agents need Business or Enterprise. By default an agent spends its owner's Select credits, even when the whole team uses it, and the owner can set a spending cap per agent; a workspace owner can instead fund it from the workspace's Notion credits under Credits, then Credit source, then Workspace budget. Credits per run depend on how much the agent reads, so run it for two weeks and read the figure under the agent's Credits menu before you commit. A Monday summary is a medium task by Notion's own table, roughly 10 credits a run on Auto, so about 43 credits a month against a 500-credit Select allowance. One agent that reads a single filtered database costs far less than one that searches the whole workspace.

Moving the old procedures in without copying their mess

Most firms already have procedures, scattered across Word files, a shared drive and one senior person's head. The temptation is to paste them all in on day one. Don't. A wiki that opens with 80 stale pages teaches staff that the wiki is stale.

Start with the five jobs that cause the most interruptions. Ask the practice manager which questions they answered most often last month: where does the signed letter go, who approves a fee change, what is the naming rule for working papers. Those five answers are your first five pages, and because they answer real questions people will visit them on day one.

For the rest, a short cleaning prompt helps once you have a procedure in plain text with no client details in it:

Rewrite this procedure as numbered steps, one action per step, each starting with a verb.
Keep the order. Do not add steps that are not in the text.
After the steps, list anything ambiguous as questions for the owner, for example
"Who approves in step 4?" or "What counts as late in step 6?"

The second half of that prompt is the valuable part. Illustratively, a draft that came back from a payroll procedure might end with: "Step 3 says 'check the totals'. Against what? Prior month, or the payroll report?" A person who has run payroll for years will not notice the gap, because they know the answer. A new starter will, and the question is exactly what the new starter would have interrupted someone to ask. Put the answers into the page and the wiki gets better than the habits it replaced.

Recurring jobs without retyping them every year

A tracker that someone has to repopulate by hand each year will be abandoned by the second year. Two Notion features help, and neither needs AI.

• Database templates. Create a Job template per job type with the checklist already inside the page, for example "Year-end accounts: 14 steps". New jobs are created from the template, so every job of that type carries the same checklist.
• Repeating database templates. Notion's help centre describes templates that create a fresh copy in the database on a schedule: daily, weekly, monthly or yearly, with a gap you choose (every 3 months, say) and a start date and time. Set the Year-end accounts template for each client to repeat yearly from the month before their year end, and the job appears without anyone remembering to create it. Test one client first: a repeat that fires on the wrong date creates a duplicate every cycle until someone notices.

Here is a filled-in Jobs row, so you can see what "small and honest" looks like (illustrative):

Client: C-0417 | Job type: Year-end accounts | Period: Year to 31 Mar 2027
Status: With client | Waiting on: Bank statements, Feb and Mar
Internal due: 2027-06-30 | External deadline: 2027-12-31
Owner: Preparer 2 | Reviewer: Partner A
Next action: Chase on 14 Apr; escalate to partner call if still missing 28 Apr

Notice that the internal due date sits six months before the external deadline. Firms that only track the legal deadline find every job in the red at the same moment. An internal date that you set by working backwards (review time, client queries, one slip) is what keeps the Waiting on client view meaningful.

A fifteen-minute weekly routine

1. Open "This week by deadline" and confirm every row has an owner. Unowned rows are the real risk.
2. Open "Waiting on client" and pick the three longest-waiting rows. Decide who chases and by when.
3. Open "Overdue for review" in the wiki and assign one page to refresh.
4. Ask each person for one thing that confused them last week and add it to the relevant procedure.

That last step is how a wiki grows without anyone being told to write documentation. It also feeds the AI: every confusing moment turned into a wiki line is one more answer the assistant can give correctly. If staff push back on the routine, the adoption tutorial is worth reading before you add anything more.

When the AI lists closed clients as late filers

This one is easy to make (illustrative). A firm turns on the AI and asks it to "list all clients with late filings". It returns eleven clients. The partner then discovers that four of them are closed clients whose rows still have old open jobs. The AI has done its job; the data is untidy. The fix is a rule in the wiki, "closing a client means setting every open job to Closed", and an "Active" filter on every view. Expect your first AI answers to expose data hygiene problems, and budget an hour to fix them.

How to tell it is working after 90 days

Judge the build on four plain measures, taken from your own views.

• Overdue procedures: the "Overdue for review" view should be empty or close to it.
• Status accuracy: pick ten jobs at random on a Friday and ask the owners whether Status is right. Nine out of ten is a healthy bar.
• Questions answered without a partner: count how often new starters used the wiki instead of interrupting someone. A tally on a sticky note is enough.
• AI usage: ask each paid-seat holder how many times they used it last week. Seats that show zero get downgraded at renewal.

If the AI measures disappoint but the wiki and tracker measures are good, you have still built something worth having, and you can drop back to Plus. For a broader take on whether the AI layer pays for itself, the Notion AI value tutorial covers it. If you would rather map this with someone, my AI implementation consultation covers that on a 1:1 call.

Questions accountants ask before moving into Notion

Can Notion replace practice management software?

Not fully. Notion tracks jobs, deadlines and procedures well, but it has no built-in time billing, engagement invoicing or client portal with e-signature. Most small firms keep their existing billing and filing tools and use Notion as the internal wiki and status board. Test it for a quarter before cancelling anything.

Should client financial data live in Notion?

Keep the tracker to status, dates, owners and client codes, and leave ledgers, returns and bank details in your accounting and document systems. A tracker that holds no sensitive figures is far easier to share with staff, and far safer if a page is shared by mistake.

Does Notion train its AI on my firm's pages?

Notion's help centre says that by default neither Notion nor its AI providers use customer data to train models, and that Notion AI respects each person's existing page permissions. Read the current subprocessor list and retention terms before you put client material in.

Is the free plan enough to test this?

For one person sketching the wiki and a small tracker, yes. Notion limits pages and blocks on Free once a workspace has two or more members, so a firm will outgrow it quickly and Plus removes that limit. Free and Plus include only a limited trial of Notion AI, so build everything first and judge the AI separately. Page history is 7 days on Free.

Further reads

• Accounting Firm Standard Operating Procedures With AI: Record to Publish — Turning recorded walk-throughs into procedure drafts before they reach the wiki.
• How Much Does AI Cost a Small Accounting Firm? — Where a Notion AI seat sits against the rest of the AI bill.
• Is Copilot Worth It for a Small Accounting Firm? Tasks and Seat Cost — The same seat-cost sum for Microsoft 365 Copilot.
• Accountant Offering AI-Enabled Automation: What to Ask Before Hiring — Questions to ask when software promises automation.
• AI for Accounting Teams in a Small Firm: Roles, Prompts and Sign-Off — Role-by-role prompts once the wiki exists.
• HubSpot for Accounting Firms: Lead Tracking and Onboarding Pipelines — A lead and onboarding pipeline to sit alongside the job tracker.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Notion pricing page; Notion help centre pages on AI security practices, managing AI credits, how credits work for Custom Agents, wikis and verified pages, database templates, formula syntax and importing data; all checked 4 November 2026.