Dext vs AutoEntry for a Small Practice: Costs, Accuracy Claims and How to Check Them

Neither vendor publishes a test method: Dext claims over 99% accuracy and AutoEntry up to 99%. On list prices and the assumptions below, AutoEntry is cheaper until clients send more than roughly 35 to 60 documents a month each, depending on your client count; above that, Dext usually wins. Run 30 of your own documents through both before deciding.

Accuracy claims on vendor pages are marketing, not measurements: neither says which documents, which fields or how many. The test below takes an afternoon and costs about $13. A word on ownership, since it comes up: AutoEntry is sold as AutoEntry by Sage and its about page says Sage acquired it in 2019, while Dext presents itself as formerly Receipt Bank. Neither fact is a reason to choose one, but both are reasons to check how well your own ledger is supported.

Follow me on Instagram@sagnikteaches

What the two vendors claim about accuracy, and what is missing

Dext's Return on Dext page says it turns invoices, receipts and bank statements into structured data with over 99% accuracy before it reaches your ledger. AutoEntry's accountants page says AutoEntry is up to 99% accurate, with always-on data quality checks, and describes optical character recognition plus machine learning. Neither page defines accuracy, names the documents tested, or says whether a person checks the extraction.

Connect on LinkedInSagnik Bhattacharya

The definition matters more than the percentage. A tool that is 99% accurate per field can still need an edit on more documents than that suggests, because a document is only clean if every field is right. With six fields per document and independent errors, the share of documents with no error is 0.99 to the sixth power, about 94%, so roughly six in a hundred need an edit. At 98% per field it is about 89%, at 95% about 74%, and at 90% only 53%. A practice posting 400 documents a month is looking at about 23 edited documents at 99% and about 106 at 95%. That is the number to test, and it is why the sheet below scores documents as well as fields.

Subscribe on YouTube@sagnikteaches

Pricing logic: per client against per credit

The two vendors charge for different things. Dext's practice plans charge per client regardless of volume, with credits for heavy extraction. AutoEntry charges per credit, with unlimited clients and users, and each document type uses a different number of credits. For Dext's own figures at your client count, see Dext pricing for accountants; the Essentials totals on annual billing used below are $207.99 at 10 clients, $353.99 at 20, $469.99 at 30 and $701.99 at 50, excluding tax and before any extra credits.

AutoEntry's USD list prices are monthly, excluding tax, and every plan includes unlimited users, unlimited cloud storage and all features.

Other terms from the page: unused credits roll over for up to 90 days; going over your allowance is allowed up to 200% of the plan at a higher rate per credit that the page does not print; above 2,500 credits you book a demo for a custom package; payment is by card only; and if you cancel, data stays accessible for a further 13 months. Accountants and bookkeepers can join a partner programme that lets them offer clients 50% off the first six months.

Build the 30-document sample

Use real documents your practice has already processed, so the posted ledger entry can serve as the answer key, and draw them from at least three clients. If any of those clients limit third-party processing, use your own firm's paperwork or ask permission for a small sample. Do not let either vendor choose the documents for you.

The answer key and the six fields

Create a spreadsheet with one row per document and six answer columns: supplier, date, document number, net amount, tax amount and gross total. Fill them in from the originals, not from either tool's output, and have a second person check the key. Add the currency if you included foreign invoices.

Fix the matching rules before you see any results. A supplier name matches if it is the same legal entity once case, punctuation and suffixes such as Ltd or LLC are ignored. Dates, document numbers and amounts must match exactly, to the cent. Then give each error a severity: critical for a wrong amount or a lost negative sign, major for a wrong date, tax figure or supplier, minor for a wrong document number or currency.

Here is how rows from the key might look, with outputs from two unnamed tools. The documents and errors are invented to show the scoring; the vendors' real output will differ.

Run both tools on the same files and log the minutes

Upload the same 30 files to each tool on the same day, through the same route (web upload is the most neutral), and do not correct anything until both runs are finished. Time two things for each tool: how long from upload until all 30 are ready to review, and how long your reviewer needs to check them against the key.

Dext's 14-day trial needs no card and gives full access to the core features. AutoEntry's trial includes 25 free credits, which is not enough. Count the credits your sample needs: if 22 documents are simple invoices or receipts at one credit each, six invoices have line items at two credits each and two are credit notes at one each, that is 22 plus 12 plus 2, or 36 credits. The Bronze plan at $13 for 50 credits covers it, and the page says you can downgrade or pause at any time and cancel up to 24 hours before renewal.

Log the capture channel too. Dext's page lists mobile and web apps, multi-channel capture and WhatsApp integration. AutoEntry lists web upload, forwarding to a dedicated email address and a phone app. If your clients send documents through a channel only one tool offers, that is a finding even if accuracy is equal.

Scoring: fields, clean documents and severity

Four measures cover it. Field accuracy is correct fields divided by 180. The clean-document rate is documents with zero edits divided by 30. Severity points add up the errors using weights of 3 for critical, 2 for major and 1 for minor. And review time is the minutes your reviewer needed. In the spreadsheet, put the tool's output in columns I to N, add a helper column O that counts mismatches per row, and total the results below.

Field accuracy     = SUMPRODUCT(--(B2:G31=I2:N31)) / (30 * 6)
Row mismatches (O) = SUMPRODUCT(--(B2:G2<>I2:N2))      (fill down to row 31)
Clean documents    = COUNTIF(O2:O31, 0)
Clean-document rate = COUNTIF(O2:O31, 0) / 30

Text comparisons in a spreadsheet ignore case, which suits the supplier rule, but strip suffixes in a helper column first. Keep each tool on its own sheet so that nothing in one run can leak into the other.

Reading the result before you look at the price

Fill in one results sheet per tool, and agree how you will read each line before any numbers arrive, so that a favourite cannot creep in once you see which tool is ahead. The right-hand column is the reading rule; the two tool columns stay empty until both runs are scored.

The critical row deserves the most attention because of what one error does downstream. A lost minus sign on a $96 credit note posts a refund as a charge, so the supplier balance is out by $192, and the mistake survives until a reconciliation catches it. One of those in 30 documents is a finding even if every other line is a tie.

Review time converts to money with one sum. If one tool's 30 documents take 8 minutes less to check than the other's, that is about 0.27 minutes a document. A practice posting 400 documents a month saves about 107 minutes, roughly 1.8 hours, which is worth about $71 a month at a $40-an-hour staff cost. Compare that figure with the monthly price difference in the cost table further down before calling either tool cheaper.

If every line is inside the noise, the accuracy test has still done its job: it has told you that accuracy will not decide this choice, and the decision moves to cost, capture routes and integration. A gap of seven or more fields, or a difference in critical errors, is a different matter and is worth repeating on a fresh batch before you commit.

Log each critical or major error in a form the vendor's support team can act on, because a vague complaint gets a vague answer. A filled-in example, invented to show the level of detail rather than taken from either tool:

Document:     23, supplier credit note, PDF from email
Field:        Gross total
Answer key:   -96.00
Extracted:    96.00 (sign lost)
Severity:     Critical
Repeatable?   Yes, two of three credit notes in the sample
Question:     Is there a setting or supplier rule that keeps
              credit notes negative, and does a correction now
              apply to this supplier's future credit notes?

Send the same kind of note to both vendors for their worst errors. How quickly and how specifically each one answers is part of what you are buying, and the replies often reveal a setting that changes the result.

The warm round: do they learn from your corrections?

Both vendors say their tools learn. Dext describes supplier rules and, as a paid add-on, AI Assist guidance; AutoEntry lists auto-suggested suppliers and customers and custom categorisation rules. After you have corrected the first 30 documents in each tool, send 15 new documents: ten from suppliers in the first batch and five from new ones. The comparison that matters is the clean-document rate on the ten repeat-supplier documents against the first round.

Score the warm round with the same sheet, then compare two rates: clean documents among the ten repeat-supplier documents (clean divided by 10), and the clean rate those same suppliers achieved in round one. A tool that learns shows a clear rise on the repeats while the five new suppliers look much like round one. That pattern says corrections are sticking, as promised, and that your first month in production will be rougher than your fifth. If the repeat-supplier rate is no better than round one, the corrections are not sticking, and that is worth raising with the vendor before you roll out to every client. For the supplier-rule side of this, see supplier rules and an auto-publish review routine.

Cost check at your own volumes

The cost comparison needs two inputs for each client: documents a month and bank statement pages a month. The table makes two assumptions you should replace with your own: 25% of documents need line items, which makes the average 1.25 credits a document, and each client sends two statement pages a month, which is six credits. The Dext column is Essentials on annual billing before any credit top-ups; the AutoEntry column is the smallest plan that covers the credits, before overage.

On those assumptions AutoEntry stays cheaper up to roughly 35 documents per client at 10 clients, 55 at 20 clients and 60 at 30 clients. The breakpoints are lumpy because the credit ladder is: a practice at 520 credits either pays overage on 20 credits, which AutoEntry charges at a higher per-credit rate it does not print, or moves up to the 1,500-credit plan at $298 instead of $103, a $195 step for a small rise in volume. Sit just under a plan boundary and you gain; sit just over and you pay for credits you will not use, although unused credits roll over for 90 days.

Work one client's month through to see how the credits stack up. Twenty-four invoices, eight of them with line items, use 24 plus 8, or 32 credits. Ten receipts add 10. Three pages of bank statement add 9, and one supplier statement adds 2. That is 53 credits for one client, which at the Platinum rate of about $0.21 per credit is roughly $11 a month. Dext's per-client price at 20 clients is $17.70, and it does not rise with documents, but heavy bank statement and line-item work consumes Dext's credit allowance, so run your heaviest client through the trial and watch the counter.

Capture routes, integrations and what else each includes

The rows that usually decide it are the integration and the capture route. If your ledger is one of the packages AutoEntry lists by name, it is a candidate. If clients already message receipts on WhatsApp, Dext's route may save you chasing. For a built-in alternative that might make both unnecessary, whether Xero Smart Document Capture is enough without Dext is worth reading, and Dext against Hubdoc runs a similar comparison on time saved.

Choosing from the numbers in three lines

Put the decision in three lines so you can defend it to a partner or a client. First, if one tool has at least four points more field accuracy, or two or more fewer critical errors, choose it unless its cost is more than double. Second, if accuracy is a tie, choose the cheaper one at your volumes, using the table above, provided the saving exceeds about $50 a month. Third, if cost is also close, choose on integration with your ledger and on the capture routes your clients actually use. Revisit the choice after three months of real use, when your clean-document rate has settled.

Two special cases bend the rule. A practice with many very small clients, a dozen documents a month each, leans towards AutoEntry because it charges for documents and not for clients. A practice that wants rules learned across its whole book leans towards Dext with AI Assist, which adds $7 per client on top, so work out that line first using a test on your own client files before it goes in the budget.

Four ways a capture test goes wrong

• Testing only clean PDFs. Both tools do well on them, and your real trouble comes from faded receipts and multi-line invoices.
• Reusing suppliers one tool already knows. If you have used Dext on a client for a year, its learned rules flatter it, so include a client that neither tool has seen.
• Scoring against the other tool's output. The answer key comes from the original documents or the posted ledger.
• Mixing extraction with coding. Whether a tool picks the right ledger account is a separate test from whether it reads the total correctly, and a tool that is good at one may be average at the other.

Treat the first test as a screening round. If the numbers are close, repeat it in a quarter on a fresh batch of 30 and see whether the ranking holds. The cost of the second test is another afternoon, and the cost of a wrong choice is a year's subscription and a migration.

Questions practices ask before running the test

Is AutoEntry still its own product?

Yes. It is sold under its own name with its own pricing page, trial and partner programme. Its site carries the line AutoEntry by Sage, and its about page says Sage acquired it in 2019. That affects which packages each vendor treats as home ground, so check the integration for your own ledger first.

Can I test both with the free trials?

Dext offers a 14-day trial with no card required. AutoEntry's trial includes 25 free credits, which will not cover 30 documents: a sample with six line-item invoices needs about 36 credits. Buying the Bronze plan, $13 for 50 credits, covers the test, and the page says you can downgrade or pause at any time and cancel up to 24 hours before renewal.

Do I need client permission to test on their documents?

Check your engagement terms and any data requirements the client has set before you upload their documents to either vendor. If in doubt, use your own firm's receipts and invoices or ask the client to approve a small sample. This is a practical checklist, not legal advice, so ask a qualified adviser if your clients have strict rules.

How many documents make a reliable result?

Thirty documents with six fields each give 180 fields. That is enough to catch a tool that struggles with your document mix, not enough to separate two tools within a few points of each other. Treat a gap of under about four percentage points as a tie and decide on cost, integration and capture routes instead.

Further reads

• Bank Statement PDF to Excel With AI: Check Every Line Before Import — Bank statements are the costliest document type in both tools.
• Accountants AI Risk Checklist: Error Points, Each With a Named Check — Error points to check across an accounting workflow.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Dext practice pricing calculator and pricing page, Dext AI Assist page, Dext help centre and Return on Dext page; AutoEntry pricing, accountants and bookkeepers, and about-us pages. Prices are USD list prices read in early November 2026.