Double, called Keeper until October 2025, charges $10, $25 or $50 per connected client a month on its Core, Plus and Scale plans, with unlimited users. Core includes AI Bank Feeds and Ask Double, Plus adds AI summaries and receipt management, and Scale adds AI journal entries, bills, accruals and flux analysis. Every tier carries a pooled AI credit allowance.
Those are standard month-to-month list prices. A firm that signs after a sales demo may land on negotiated rates and an auto-renewing contract, and the AI is metered: when the firm's shared credits run out, the AI features pause for every client until the cycle resets or someone approves overage spending. The real cost per client is the tier price, plus a share of any extras, plus your team's credit habits, and each of those three parts can be estimated before you sign.
Same company, new name: reading old Keeper reviews safely
Keeper announced its change of name to Double on 23 October 2025 in a note from its chief executive: same team, same product, new name and website, with weekly product updates and support carrying on as before. keeper.app now answers with a permanent redirect to doublehq.com, so old bookmarks still land in the right place. What does not carry over is old advice. Reviews and comparison tables written about Keeper may quote earlier plan names and prices, and some help articles still mention a legacy "Lite" plan. Use the current Core, Plus and Scale prices below, and treat any review that never mentions Scale or AI credits as probably out of date.
The pricing page has two tabs. The one for internal finance teams is an annual flat rate quoted on request, so ignore it if you run a practice. The other tab, per-client pricing for accounting and bookkeeping firms, is the one every figure here uses. Its "Get started" buttons lead to a demo booking rather than a card form, which matters for the contract questions further down.
What each price tier switches on, feature by feature
Double sits on top of a client's ledger rather than replacing it. A client is "connected" once you import their QuickBooks Online or Xero ledger, and from then on the tier you choose for that client decides which workflow and AI features appear. This is the page's own breakdown, as of November 2026, with the AI credits each client adds to the firm's shared pool alongside it.
| Tier | Price per connected client a month | Features it adds | AI credits it adds to the pool |
|---|---|---|---|
| Core | $10 | Two-way ledger sync, branded client portal, AI Bank Feeds, task management, firm-wide reporting and time tracking, vendor tax-form requests, integrated email, AI Fix this Rec, Ask Double | $1 |
| Plus | $25 | Everything in Core, plus one-click AI financial summaries and receipt management | $5 |
| Scale | $50 | Everything in Plus, plus accruals, allocations, AI journal entries, AI bills and AI flux analysis | $10 |
| Enterprise | Quoted on request | Everything in Scale, plus custom service terms, custom API endpoints and a dedicated Slack channel | Not stated |
The tier names tell you little, so here is what the AI features actually do, using the vendor's help-centre descriptions rather than the marketing page.
- AI Bank Feeds (Core). Bank-feed coding inside Double, described as assisting the bookkeeper rather than replacing the review. Each line lands in one of four buckets: Matches (against existing ledger entries), Rules (your if-then-else rules), Auto Classifications (vendors the ledger has always coded the same way, with a high or medium confidence label) and Needs Review. The last bucket splits into Uncertain, where a recognised vendor has inconsistent history, and New Vendors, where the AI searches the web and suggests who the vendor is.
- AI Fix this Rec (Core). The pricing page's name for AI help when a reconciliation does not tie; the feature pages call the same idea AI Reconciliations. Treat it as a second pair of eyes on a stubborn difference, not an automatic sign-off.
- Ask Double (Core). A chat panel inside Double that answers questions about a client's ledger, tasks and documents in plain language and can create and assign tasks. It is available for QuickBooks clients only, and you can pick a Standard, Auto or Expert model. Drafting and posting transactions from the chat is a Scale feature, so on a Core client Ask Double reads and organises but does not post.
- AI summaries (Plus). A drafted narrative for the monthly report built from the client's last 12 months of financials, including vendor-level spend and year-on-year comparisons, with professional or casual prompt styles you can edit per client.
- AI Transactions (Scale, behind "AI journal entries" and "AI bills"). You upload a document or describe the entry in plain words, and an agent drafts and posts it. The help article names payment-processor exports, loan documents and payroll journals as typical sources, and says one file can produce several transactions. Saved skills let the agent reuse your choices next time.
- AI Flux Analysis (Scale). Plain-language explanations for variances in the profit and loss and balance sheet, with materiality thresholds you set (the help article's examples are $1,000 or 20%, combined with AND or OR).
- AI Agents (grouped under Scale in the help centre). Instructions, a schedule or trigger, and a list of clients the agent may touch. The pricing page does not list agents as a separate line, so ask where they sit.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe pooled AI credit allowance sitting under the price
The billing FAQ is where the second meter appears. Every plan includes AI credits, pooled across the firm: $1 per Core client, $5 per Plus client and $10 per Scale client each month. The vendor's own example, two clients on each tier, gives a $32 pool. Credits reset at the start of each billing cycle. Notice the proportions: the allowance is 10% of a Core fee but 20% of a Plus or Scale fee, so a firm heavy on Core clients has a thin pool.
Two kinds of usage draw on it. Ask Double is recorded at firm level even when you are working on one client's file. Bank feeds, executive summaries, flux analysis, reconciliations, transactions and workpapers are recorded against the client concerned. Because the pool is shared, one enthusiastic user running long chats on a big workbook can use up what forty other clients contributed.
The limit behaves in a way worth knowing before month-end. At 80% of the limit, if overages are not enabled, Double shows a banner and emails the admins. At 100% the AI features pause across all clients, not only the heavy one, until a Super Admin raises the usage budget or the cycle resets. Overage is opt-in: a Super Admin sets an overage budget and an alert threshold (the budget has to be higher than the alert), the extra is billed on top of the plan, and it lands on the next monthly invoice. The help article does not state a price per extra credit, so that goes on your question list.
Double's own tips for stretching the pool are practical and free. Keep each Ask Double chat to one task, because the full history and any uploaded file are re-sent with every new message. Upload only the worksheet you need, since file processing is described as the most credit-intensive thing the chat does. Pick the model deliberately: the help article says Standard costs about 7% of Expert, and that Auto typically spends about half what always-on Expert would.
Costs that sit outside the per-client price
The three tier prices are not the whole bill. Check each of these against your own set-up.
- Autoclose, $500 a month, firm-wide. Agents fetch client bank and card statements, code bank feeds using your rules and history, reconcile every month and flag what they cannot resolve. The same page marks accruals, AI Transactions, flux analysis and the agentic browsing that pulls statements from sites with no integration as Scale-plan features, so ask how much of Autoclose works for clients on Core or Plus. In the vendor's words, you review the agents' work, resolve the flagged items and sign off. Spread over 40 clients it is $12.50 each. At an internal cost of $35 an hour, that is 14 hours of work a month it must remove before it pays for itself.
- The Tax Suite. Listed as included with a $200-a-month annual commitment. Ask the sales team to say in plain words whether that is a spend commitment or an extra charge on top of your client fees.
- Email inboxes. One practice inbox is free with Core. Each extra practice inbox and each personal inbox costs $10 a month.
- The ledger subscriptions themselves. Double connects to a client's QuickBooks Online or Xero file, so those subscriptions stay a separate cost, paid by the client or by you.
- A billing trap on leaving. Archiving a client cuts the ledger connection and removes them from future billing, and reactivating is charged pro rata from that date. But the billing FAQ warns that removing the Double app from inside QuickBooks or Xero does not make the client "unconnected", so you can keep paying for a client who has already left.
Plan changes follow the usual proration rules. Upgrading a client mid-cycle adds a pro-rated charge to the next invoice. Downgrading takes effect on the next invoice date, so the client keeps the higher tier until then. QuickBooks-connected clients can also start a 14-day free trial of Scale from the client plan cell; at the end, the subscription reverts. That trial is the cheapest way to test AI journal entries and flux analysis on a real month.
Pricing a 40-client book: the full monthly bill
Take a four-person accountancy practice with 40 monthly bookkeeping clients, all connected. After sorting them (the method comes in a later section), 22 sit on Core, 12 on Plus and 6 on Scale. Three staff connect personal inboxes; the practice inbox is already free. All figures are illustrative arithmetic on the list prices above.
| Line | Quantity | Unit price | Monthly cost |
|---|---|---|---|
| Core clients | 22 | $10 | $220 |
| Plus clients | 12 | $25 | $300 |
| Scale clients | 6 | $50 | $300 |
| Extra personal inboxes | 3 | $10 | $30 |
| Total without add-ons | 40 clients | $850 | |
| Average per client | $21.25 | ||
| Autoclose add-on | 1 | $500 | $500 |
| Total with Autoclose | $1,350 | ||
| Average per client | $33.75 |
Three things fall out of that sum. First, the yearly figure without add-ons is $10,200, or $16,200 with Autoclose. Second, mixing tiers is worth real money: putting all 40 clients on Scale would cost $2,000 a month before inboxes, against $820 for the mix, a gap of $1,180. Third, the shared credit pool for this firm is 22 x $1 + 12 x $5 + 6 x $10 = $142 a month. That sounds generous until you remember that the 22 Core clients contribute only $22 of it.
Passing the cost on is a pricing decision rather than a software one. If your fixed fees already include a margin for tools, $21.25 a client may be absorbed; if they do not, the tutorial on pricing AI bookkeeping as a service line shows how firms build tool costs into a fee.
How many minutes a month each tier must save
A price only means something against the time it removes. Using an internal cost of $35 an hour (substitute your own loaded figure), the break-even minutes per client per month are simple division: the extra price divided by 35, multiplied by 60.
| Decision | Extra cost a month | Minutes saved to break even |
|---|---|---|
| Core against doing without Double | $10 | 17 |
| Plus against doing without Double | $25 | 43 |
| Scale against doing without Double | $50 | 86 |
| Upgrade a client from Core to Plus | $15 | 26 |
| Upgrade a client from Plus to Scale | $25 | 43 |
| Upgrade a client from Core to Scale | $40 | 69 |
| Autoclose spread over 40 clients | $12.50 per client | 21 |
The upgrade rows are the ones that decide tiers, because the client is already connected and paying $10. A Scale upgrade must save about 69 minutes on that one client every month. That is plausible for a client with 120 supplier bills to post, and implausible for a freelancer with 35 bank lines.
Sorting clients into tiers with an eight-row test
Instead of guessing, read each client's month against the features. Find the rows that describe real work in that client's ledger, and take the highest tier any of them needs.
| Something that happens in the client's month | Feature that answers it | Lowest tier |
|---|---|---|
| Bank lines to code and reconcile | AI Bank Feeds, AI Fix this Rec | Core |
| You want to question the ledger in plain words | Ask Double | Core |
| The monthly report needs a written note | AI summaries | Plus |
| Receipts arrive as photos or forwarded emails | Receipt management | Plus |
| Prepayments, accruals or costs split across classes | Accruals, allocations | Scale |
| Entries built from loan papers, payroll or processor exports | AI journal entries | Scale |
| Supplier bills arrive as PDFs to be posted | AI bills | Scale |
| Month-on-month variances need explaining | AI flux analysis | Scale |
Then run the minutes test from the previous section. If a client matches a Scale row but the work is small, stay lower. Here are six clients, all generic and illustrative, worked through.
- A café with daily card payouts. Rows one and two always; row six (the payout journals) only twice a month, about 10 minutes of work. Verdict: Core at $10. Going to Scale adds $40 to save 10 minutes, and the break-even is 69.
- A two-van trades business. The owner photographs receipts and wants a short note each month: rows one, three and four. Verdict: Plus at $25, needing 26 minutes of saving over Core. A weekly receipt-matching chore of 10 minutes plus a 20-minute report note clears it.
- A property-management firm. Prepaid insurance, held deposits and a monthly question about why costs moved: rows five and eight. Verdict: Scale at $50.
- A freelance designer. About 35 bank lines a month and nothing else. Verdict: Core at $10, with no case for anything higher.
- A small manufacturer. Around 120 supplier bills a month arrive as PDFs: row seven. If posting one by hand takes two minutes, that is 240 minutes. Scale's extra $40 pays back once AI bills removes 69 of them, about 29%. Verdict: Scale at $50, subject to the pilot below.
- A family garage. Receipt photos from the workshop but no commentary wanted: row four only. Verdict: Plus at $25.
Those six clients cost $170 a month together, an average of $28.33, and only two of them needed Scale. In this illustration most clients sit near the bottom of the table, and the expensive tier is reserved for the ledgers heavy with accruals or documents.
Limits the help centre states that change the value
Several limits are documented in the help articles rather than on the pricing page, and some change what you are really buying.
- AI Bank Feeds supports one regional edition of QuickBooks Online. The help article says other regions and Xero are "coming soon". If most of your clients are on Xero or on a different edition, the headline Core feature may not work for them yet. Check this before anything else.
- Ask Double is QuickBooks-only, according to its help article.
- Double's bank feeds are separate from the ledger's own feeds. A line cleared in one still shows as pending in the other, and categorised transactions do not appear as cleared in QuickBooks reconciliations. The article puts this down to a QuickBooks API limitation. Your reconciliation routine has to cope with that.
- The pool pauses AI for every client at once, as covered above.
There is also a limit that is not stated anywhere and is worth testing. The Auto Classifications bucket is built on how the ledger has coded a vendor before, with a confidence label. That is a sound idea when the history is right and a trap when it is consistently wrong. Picture a client whose last two years of fuel and equipment purchases from one supplier were all coded to "office supplies". The history is perfectly consistent, so the lines come back as high-confidence matches, and the mistake repeats at speed. That is my reading of how the bucket is described, not a vendor warning. At onboarding, sort a client's largest ten vendors by annual spend and check their accounts against your chart-of-accounts policy before trusting the label.
The review boundary matters just as much with the agent features. The vendor's own wording is that you review the work, resolve the flagged items and sign off, so the sign-off stays yours. The tutorial on what bookkeeping AI agents do alone and what you must review sets out a review routine that applies here.
A two-cycle pilot that settles the price on your own clients
A pilot beats a demo because it uses your clients' untidy ledgers. Pick six: two per tier by the eight-row test, with at least one whose coding history you already distrust. Before switching anything on, time the last close for each, task by task, from your current time-tracking. Then run two billing cycles.
- On day one, a Super Admin sets the usage alert threshold and a modest overage budget, so a surprise pause cannot stall a close.
- Record minutes per close for each pilot client against the old figure.
- Record corrections: how many AI-suggested codings you changed per 100 bank lines, split by bucket.
- Record credits used per client from the usage screen, and note any pause.
- At the end of cycle two, compare minutes saved with the break-even minutes from the earlier table.
The table below shows how a result might read. The numbers are invented to show the method, not measured on any real firm.
| Pilot client | Tier | Minutes before | Minutes after | Saved | Break-even | Verdict |
|---|---|---|---|---|---|---|
| Café | Core | 95 | 70 | 25 | 17 | Keep |
| Two-van trades business | Plus | 140 | 115 | 25 | 43 | Drop to Core |
| Property-management firm | Scale | 260 | 150 | 110 | 86 | Keep |
| Manufacturer | Scale | 310 | 215 | 95 | 86 | Keep, margin thin |
In that illustration, two of the four tier choices were clearly right, one was marginal and one should step down. That is the value of a pilot: it corrects the tiers before they are priced into a contract. For a fuller method of testing an AI bookkeeping tool on real files, with a scoring sheet, see how to test a tool on your own client files.
Questions to put to the sales team before you sign
The pricing page says the first step is a personalised demo, and a signed pilot contract comes with a customer-success contact. The billing FAQ also says contracts typically carry negotiated pricing and renew automatically. These questions turn the standard list price into a binding one.
- What is the contract term, and when does auto-renewal take effect? What notice do I need to give?
- Do the standard month-to-month prices above apply, or are my rates different, and for how long are they fixed?
- What does one extra AI credit cost beyond the pool, and can that rate be written into the contract?
- Does Autoclose use the same credit pool, or is it metered separately, and which of its agents work for clients on Core or Plus?
- Where do AI Agents sit: in Scale, in Autoclose, or somewhere else?
- Which QuickBooks Online editions and which Xero features does each AI feature support today, and is there a date for the "coming soon" items?
- If I archive a client, when does billing stop, and what exactly must I do inside QuickBooks or Xero?
- Is the Tax Suite a spend commitment or an extra?
- What can I export if I leave, and how are client portal links handled afterwards?
Write the answers into a one-page summary and keep it with the contract.
When a general AI assistant does enough for a fraction of the price
Double is a workflow product. You pay for the ledger connection, client portal, task tracking and AI inside one place. If your real need is smaller, say help categorising an exported bank file or drafting a client note, a general assistant covers it. ChatGPT Business and Claude Team both list Standard seats at $25 a user a month on monthly billing ($20 on annual), with a two-seat minimum. Four seats at $25 is $100 a month, about $2.50 per client across a 40-client book, against $21.25 for the Double mix above.
The gap is not the whole story, because the assistant writes nothing back to the ledger, tracks no close, and gives your clients no portal. It also means client bank data leaves the ledger and goes into a chat, which needs a plan that does not train on business content by default and a clear note to clients. The tutorial on testing Claude and ChatGPT on bank data gives you a repeatable way to measure what these assistants get right. Many practices end up with both, using the assistant for one-off analysis and the platform for the monthly close. For two other price structures to hold against Double's, read the breakdowns of Dext's plans for accountants and of Canopy's power-ups and client credits.
Double pricing questions from practice owners
Is Keeper still sold, or is it only Double now?
Only Double now. Keeper announced its change of name to Double in October 2025, and keeper.app redirects permanently to doublehq.com. Old reviews and price lists that mention Keeper may use outdated plan names, so rely on the current Core, Plus and Scale pricing page and ask the sales team to confirm what applies to you.
Does Double charge per user or per client?
Per connected client. A client counts as connected once you import their QuickBooks Online or Xero ledger, and you are billed for connected clients at the start of each monthly cycle. Team members, client portal contacts and unconnected clients are free, apart from $10 for each extra connected email inbox.
What happens when my firm uses up its AI credits?
By default the AI features pause for every client until the cycle resets or a Super Admin raises the usage budget. If you enable an overage budget, extra usage is billed on top of your plan and added to the next monthly invoice. Set an alert threshold on day one so the pause never surprises a close.
Can I put different clients on different plans?
Yes. Admins choose Core, Plus or Scale for each client under the client plan setting, so a simple client can stay at $10 while a complex one sits at $50. Upgrades are charged pro rata for the rest of the cycle, and downgrades take effect from the next invoice date.
Further reads
- AI Powered Bookkeeping Platform: How a Practice Chooses One — How a practice weighs platforms before committing to one.
- Best AI Accounting Agents: QuickBooks, Xero JAX and Sage Copilot — What the ledger vendors' own AI agents already do.
- AI Bookkeeping Software vs a Human Bookkeeper: Accuracy and Cost — Accuracy and cost set against a human bookkeeper.
- AI for Chartered Accountants: What to Tell Clients About Your AI Use — What to tell clients once AI touches their books.
- AI Bookkeeping Software Challenges: Where It Goes Wrong and the Fix — Where AI bookkeeping goes wrong and how to catch it.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Double's pricing page and help-centre articles on billing, AI credits, AI Bank Feeds, Ask Double, AI Transactions, AI Flux Analysis and the executive summary; the keeper.app redirect; Double's own 23 October 2025 announcement of the rename.