Accounting data entry services are the narrowest thing a firm can buy. A batch of source documents goes out as images, coded rows come back, and nothing in between involves a decision about a client's books. What decides whether those rows are usable is the specification, and writing it is your job rather than the provider's. Federal and state buyers of the same service write theirs down, and their terms are the ones a firm buying a bookkeeping engagement rarely sees: the unit the work is priced in, the accuracy standard and who measures it, and what has to be true about the images before the quoted price applies at all.
What Moves Out, and What Has to Come Back
You are buying transcription, and the occupational description of the job says so in plain terms.
The federal profile for the role covers reading source documents such as canceled checks, sales reports, or bills, and entering data in specific data fields or onto tapes or disks for subsequent entry, using keyboards or scanners (O*NET, Data Entry Keyers, 43-9021.00). The verification tasks in the same profile stop at the source document: compare data with source documents, or re-enter data in verification format to detect errors, then locate and correct data entry errors or report them to supervisors.
Read that list and the shape of the purchase is clear. What goes out is images, meaning bank and card statements, vendor invoices, receipts, checks and expense reports. What comes back is rows keyed against those images, in whatever layout you asked for. Nothing in the description reaches an opinion about which account a row belongs in.
Two things decide whether that return leg is worth anything, and both belong in writing before the first batch leaves.
- The format the rows and images arrive in. In the price ranges the National Archives catalogued for federal agencies buying document conversion, TIFF and PDF are the most common file formats and are provided at no additional cost, while some vendors provide images by other media or by file transfer at no additional fee and others charge for it (National Archives, Sample Digitization Pricing). Name the layout your ledger will accept, and test one small file end to end before a real batch depends on it.
- The condition of the images you send out. The same chart notes that data entry and indexing prices often assume indexed fields can be easily located within a document and are legible, and that scanning prices often assume records are in good quality and require little to no prep or special handling, with more extensive prep liable to move the rate (National Archives, Sample Digitization Pricing). A quote written against clean statements is not a quote for a client who photographs receipts on a car dashboard.
One more thing has to travel with the batch, and it is the one firms leave out. A keyer can only put a row where your list of valid destinations allows. Your chart of accounts, your vendor list and your class or job codes are that list, so if they are stale on the day the batch goes out, the rows come back tidy and wrong.
Two Turnaround Dates and a Route Back for the Unreadable
A keyed batch is only worth its date, and a standard turnaround is not a date.
Name the day the batch has to be with the provider and the day the coded rows have to be back, then set both against the close you are trying to protect. The close calendar version of that discipline, with an internal deadline sitting behind the client-facing one, is worked through in record to report services.
The second thing to date is the exception. Some documents in every batch are unreadable, incomplete or duplicated, and with no named route back they simply sit. A state scope of work for keying staff writes that route into the job itself, requiring the clerk to process source documents by reviewing data for deficiencies, resolving discrepancies using standard procedures, or returning incomplete documents to the supervisor for resolution (State of Hawaii, Office of Enterprise Technology Services, Scope of Work, Data Preparation Unit Data Entry Clerk).
The physical leg is also longer than a proposal makes it sound. In the Internal Revenue Service pilots that moved paper return processing to outside contractors, returns were sent to contractor sites where they were scanned, and the contractors extracted the return data from the scanned images and transmitted the data to the IRS for processing (TIGTA, The IRS Has Made Limited Progress Achieving Paperless Processing, Report Number 2026-408-003, February 6, 2026). Each of those handoffs adds time to the loop, and the loop is what you are buying.
How Accounting Data Entry Services Are Priced, and What a Heavy Month Costs
On the federal side this work is usually quoted per unit of keying rather than per seat, and the unit is small.
The federal pricing chart that shows it is one of three products the Federal Records Management Council released through the National Archives in June 2019, alongside a white paper and a calculation tool, to help agencies estimate costs and build a business case for digitization projects (National Archives, Records Express, FRMC Digitization Cost Benefit Analysis Tools).
For data entry and indexing, that chart records task-based pricing of $0.005 to $0.019 per keystroke alongside hourly pricing of $26 to $40 an hour for a data entry specialist, and notes the work is typically priced on a per keystroke basis or per 1000 characters, occasionally priced by hourly labor only. Those ranges are what the chart observed on various vendors' GSA Schedule 36 for document conversion, rather than a survey of what accounting firms pay (National Archives, Sample Digitization Pricing).
Scanning is a separate line. Standard black and white pages ran $0.04 to $0.07 per image, and the chart notes vendors may impose a minimum image volume requirement and may also offer discounted pricing for higher volumes (National Archives, Sample Digitization Pricing).
Quality control is a third line, priced apart from the keying it checks. The chart puts advanced clean-up and additional quality control measures at $0.01 to $0.04 per image, or $35 to $55 an hour for a quality assurance specialist (National Archives, Sample Digitization Pricing).
That chart opens with a warning worth reading twice. Pricing structures vary significantly from vendor to vendor, which can make a precise comparison difficult, because vendors often have their own definitions for specific tasks and labor categories: one vendor's price for scanning may include basic prep and image clean-up, another's may add basic indexing, and a third may simply quote an hourly rate for the same task (National Archives, Sample Digitization Pricing).
So the bill tracks the documents. A month when a client's volume doubles is a month when this line doubles, with no contracted band to absorb it and no idle seat to soak it up. That is the opposite of the volume-band arrangements that price a whole function against a forecast, worked through in finance and accounting BPO, and it sits outside the structures a bookkeeping quote normally uses, covered in bookkeeping services pricing.
Buying the hour instead moves the risk back to you and makes the output yours to measure. One state office took that route for a data preparation unit, buying roughly 1,968 hours of a temporary full time equivalent data entry clerk at an hourly rate for a term commencing July 1, 2016, and then wrote its own output floor into the scope (State of Hawaii, Office of Enterprise Technology Services, Scope of Work, Data Preparation Unit Data Entry Clerk).
The Accuracy Standard, Because the Default Is a Single Pass
What the provider measures by default is whatever comes bundled with its standard price, and that default is weaker than a quote makes it sound.
The National Archives chart names what a data entry and indexing price typically includes: 99% accuracy, single pass (National Archives, Sample Digitization Pricing). Single pass means one operator keys the field once, and nothing compares that keystroke against a second independent reading of the same document.
Double keying is the standard alternative. Two operators key the same field independently, and every disagreement between them is flagged for a person to settle.
One study of patient-reported outcome questionnaires measured what that buys, and the answer depended on the field. On checkbox fields read by optical mark recognition, single-key entry produced 0.370 errors per 1000 fields against 0.046 per 1000 fields for double-key entry, a difference the authors reported as statistically significant at p = 0.020. On handwritten fields read by intelligent character recognition, the same two methods ran 6.734 against 3.367 errors per 1000 fields, and there the authors found no statistically significant difference, at p = 0.565 (Paulsen, Overgaard and Lauritsen, PLoS One, 2012).
Read the first comparison as the measured one and the second as untested at that sample size. Both ran on structured questionnaires rather than crumpled receipts, so the gap between the two methods is a direction to reason from, not a rate to quote back to a provider.
The denominator is where a specification quietly fails. A percentage measured across keystrokes, a percentage measured across a sampled clerk's day, and the question you care about, whether the rows in this batch are right, are three different measurements, and a provider will report whichever one it chose.
A buyer who wants a stronger method writes it into the scope, and a usable standard names four things: the unit the accuracy is measured in, the floor, who samples the output and how, and the remedy when the floor is missed. The state scope of work above names three of them. It sets the floors and attaches a remedy to them, stating that if averages of fewer than 200 records per day are input and/or less than 95% accuracy is found for a data entry clerk, the office will request that the clerk be replaced and the vendor shall provide a new data entry clerk within 2 days. It names the measurer and the method too, since the office periodically conducts random counts and a random sample of each clerk's work (State of Hawaii, Office of Enterprise Technology Services, Scope of Work, Data Preparation Unit Data Entry Clerk). What it never says is what the 95% is measured across, which is the line to write into yours.
Name all four, and accept that a floor with no remedy behind it is a preference.
What Is Still on Your Desk When the Rows Come Back
Everything that required a decision.
One federal definition names the safe side of the same line, treating an individual providing only typing, reproduction, or other mechanical assistance in the preparation of a return as someone who is not a tax return preparer (26 CFR 301.7701-15(f)(1)(viii)). The scoping consequences of that boundary are worked through in what a virtual assistant for accounting firms covers.
A second definition, the one the confidentiality rules for client tax return information use, draws its line further out. A person engaged in the business of providing auxiliary services in connection with the preparation of tax returns is a tax return preparer for the purposes of section 7216, and the regulation's own example says a secretary who types the information on the returns into a computer is one for that section even though she is not one under section 7701(a)(36) (26 CFR 301.7216-1(b)(2)). Keying does not sit outside those rules because it is mechanical.
So the rows land and the work starts. Classification against the client's real facts, the reconciliation that proves the coding, the accruals, the close. Which of those move to a provider and which do not is sorted in accounting tasks to outsource, and the preparer and approver split for a ledger engagement is set out in record to report services.
The arithmetic that matters is not the price per document. It is the cost per usable ledger, and rework gets charged to the wrong budget. The provider bills the keystrokes either way, while the fixing lands on your reviewer, in the week you bought the service to protect.
Software changes the unit rather than the review. The same pricing chart notes vendors may offer different methods, such as optical character recognition, for capturing data entry fields to reduce the manual keystrokes required (National Archives, Sample Digitization Pricing). Fewer keystrokes is a smaller invoice, and what an extracted field means in a particular client's chart of accounts is still a question somebody has to answer. Where that line sits for accounting work generally is covered in AI versus offshoring.
A volume commitment can fail on volume and on quality at once, which is worth knowing before you agree one with no quality clause beside it. The IRS declined to extend a historical scanning contract, citing reasons such as the contractor's inability to achieve the desired volumes and quality of scanning, on a contract that required the contractor to scan 100 million pages per year. The contractor cited difficulty obtaining the required security clearances, and the IRS had cleared 245 of the 470 contract personnel as of April 2025 (TIGTA, The IRS Has Made Limited Progress Achieving Paperless Processing, Report Number 2026-408-003).
The Exposure Attaches to the Image, Not to the Coded Row
Raw source documents are the most identifying material in the whole stack, and they move first.
A bank statement carries the account number. A vendor file carries a taxpayer identification number. The federal profile for the keying role lists canceled checks among the source documents read (O*NET, Data Entry Keyers, 43-9021.00). None of it has been classified, redacted or summarized at the point it is uploaded, so the security duty attaches to the cheapest work in the stack, before anyone has produced anything of value.
The federal government buys this service at a scale no accounting firm will match, and it treats it accordingly. All individuals who work under the IRS Zero Paper Initiative contracts must pass an IRS background clearance, and the clearance queue became the capacity constraint. As of August 2025 the IRS Human Capital Office had approved 250 of the 496 individuals the interim contractor had submitted, only 184 of those approved were personnel working in the processing pipeline, and the contractor estimated it needed approximately 600 individuals in that pipeline to achieve the contracted volumes (TIGTA, The IRS Has Made Limited Progress Achieving Paperless Processing, Report Number 2026-408-003).
The wait is built into that process rather than being bad luck. Workers often need a job immediately and cannot wait the estimated four to five weeks it takes to be cleared to work on the contract (TIGTA, The IRS Has Made Limited Progress Achieving Paperless Processing, Report Number 2026-408-003). Screening is what sets that clock, so ask how long your provider's screening takes and who is waiting on it.
A state buyer of a single keying seat asked for less and still asked in writing, requiring the vendor to complete background screening of candidates before work commenced, covering education and qualification verification, employment verification with the last two employers, two professional references, and a check that the candidate has no conviction of a violation of any law (State of Hawaii, Office of Enterprise Technology Services, Scope of Work, Data Preparation Unit Data Entry Clerk).
Two duties you already carry reach this work whether or not the proposal mentions them. The service provider oversight the FTC Safeguards Rule requires of your firm is set out in cybersecurity for CPA firms, and the consent and masking rules that apply once client tax return information crosses a border are set out in back office support for CPA firms. Neither of them waits for the work to become interesting.
The uncomfortable part is that limiting access does not help much here. A reviewer can be handed one schedule. A keyer needs the whole image, because the image is the input.
Decide Whether You Are Buying Keying or Judgment
Data entry is a real purchase and a small one. It is worth buying when your bottleneck is the volume of documents rather than the number of decisions, when your chart of accounts and vendor list are current enough to constrain the keying, and when someone in the firm has time to check what comes back against a standard you wrote.
Write that standard first. Name the unit the price is quoted in, the accuracy measure and its denominator, the floor that measure has to clear, who samples the output and how often, the remedy when a batch misses the floor, the day the rows are due, and the route back for a document nobody can read. Seven lines, and they are the difference between a batch you can post and a batch you have to redo.
Then take last month's batch and count two things: how many rows came back needing a correction, and how many needed a decision your provider was never going to make. The first number tells you whether you specified the work properly. The second tells you whether keying was the right purchase at all.
If the second number is the big one, the constraint is judgment rather than volume, and no keying rate touches it. Accountably places trained offshore accountants and tax preparers inside US CPA, EA and accounting firms, on your software and your SOPs, in about 3 to 4 weeks, with the review chain and the signature staying where they already are. Since 2022 that has meant 20+ US firms and 30+ placements. Don't trust us, test us: the Free 40-Hour Proof Pilot puts a fixed block of your own representative work through full multi-layer review, so your reviewer grades real output before a client file depends on it, and if someone is not the right fit in the first 30 days the 30-Day Fit Guarantee replaces them free.