DATA ENTRY OUTSOURCING FAQ
Data Entry Outsourcing FAQ: The Questions Buyers Actually Ask
Here are 23 common questions operations, finance and compliance teams ask before outsourcing data entry. The answers explain what to expect from a provider, what to check before choosing one and how YesAssistant handles key parts of the process.
01
Where Data Entry Projects Go Wrong
3 questions
Most of the time the issue is the denominator, not dishonesty. A provider quoting character-level accuracy on a ten-field record can report 99% and still leave a visible error in roughly one record in five. Correct at the character level and wrong at the record level can both be true.
So before you accuse anyone of overstating their numbers, check three things. First, which denominator the number uses. Second, whether it is measured on a sample or on the full batch. Third, whether you are looking at a keying error or a rule error. A rule error means two operators read your field definition differently. That is a specification problem, not a quality problem.
Two causes, and both are structural rather than personal.
- Team turnover: The operators who learned your rules left, and their replacements were trained on the previous output instead of on a document. Ask any provider for the monthly attrition rate on your account. Almost nobody volunteers.
- Rule drift: Edge cases got resolved verbally in the first weeks and never written back into the specification, so every new person invents a slightly different answer to the same question.
The fix is identical for both: a field dictionary the provider maintains as a living document, with a change log you are allowed to read.
Find out where the hours are going before you change providers. The cause may be the provider, the contract or your own field definitions.
- Correcting the same field repeatedly: The field definition is ambiguous. That one is yours to fix.
- Handling items the provider could not resolve: There is no exception queue, so exceptions land on you by default. That belongs in the contract.
- Checking everything because nobody trusts the output: You are paying for QA twice. That is a measurement problem, and it is solved by an agreed denominator, not by more reviewing.
02
Accuracy and Proof
6 questions
Nothing at all, until you know the denominator. The same batch can be reported three ways.
- Field level accuracy: Correct fields divided by total fields keyed.
- Record level accuracy: Records with zero errors divided by total records.
- Character level accuracy: Correct characters divided by total characters.
The gap is not academic. A ten-field record delivered at 99.5% field level accuracy leaves roughly one record in twenty carrying at least one error. The same 99.5% quoted at record level is a far stronger promise. Character level is the weakest of the three and the one most often quoted in sales decks, because it produces the biggest number.
Get the denominator written into the contract, then ask whether it is measured on a sample or on 100% of output and who does the measuring. Some providers report DPMO, defects per million opportunities, which is the same math scale for high-volume work. For a deeper breakdown, see data entry accuracy standards.
Three methods, in ascending order of cost.
- Automated validation. Format validation, range checks and lookup validation against a reference table. Cheap, catches structural errors, misses a plausible wrong value.
- Acceptance sampling. A QA analyst re-keys or reviews a statistical sample, commonly under ISO 2859-1 acceptance sampling plans. Cost scales with sample size rather than total volume.
- Double key verification. Two operators key the same record independently and the system flags every mismatch. Highest catch rate and the highest unit cost.
Production runs usually mix all three at different stages. Ask which method applies to which fields, because a provider may double-key a five-field header and only sample the line items underneath it. That distinction is where most accuracy disputes start.
When a batch misses the contracted accuracy target, the provider should rework it at no extra cost and verify it again before acceptance. A usable rework clause should state four things:
- The trigger threshold
- Who pays for the rework
- The correction turnaround
- Whether repeated misses trigger a service credit or an exit right
“We guarantee accuracy” is only a claim. “Batches below the contracted field level accuracy are re-keyed at our cost within one business day and three consecutive misses release you from the notice period” is a measurable commitment. Get that commitment in writing.
Field-level accuracy, written into the scope of work before the first batch runs. Correct fields divided by total fields keyed, measured on your fields rather than on a character count.
Verification is layered rather than uniform. Automated format validation and range validation run on every record, acceptance sampling runs on the batch, and double key verification runs on the fields you flag as critical. You tell us which fields carry the risk, and we price the verification method against those fields instead of charging the highest method across everything. Batches that fall below the contracted level are re-keyed at our cost, not rebilled as new work.
Run a paid pilot against a golden set. Build a sample of 100 to 200 records your own team has already keyed and verified. Include the ugly documents on purpose: poor scans, handwriting, the layouts that break your current process, the edge cases nobody documented.
Send the identical set to every shortlisted provider and score what comes back against your known-correct version. You now hold a measured accuracy number per vendor instead of a claimed one. You also get a live read on communication speed and on how each team handles an exception it cannot resolve.
Keep the golden set afterward and re-run it quarterly against live production. For the full evaluation scorecard, see how to choose a data entry service provider.
Yes, and it is free. Send 100 to 200 records you have already keyed and verified. We process them at no charge and sign an NDA before any file transfer. We run them through the same process, the same team and the same QA that production would use. Then we return the finished batch in your own output format alongside the measured accuracy against your verified version.
You get a real number to compare against every other provider on your shortlist. We get a calibrated rule set before volume starts. No minimum commitment until that result is in front of you.
Free Data Audit
Not sure which of these steps your process is missing?
Send a 100 to 200-record sample and a description of the workload. We come back with your exception rate, a cost per thousand finished records, and the gaps in your current handoff.
03
Pricing and contracts
4 questions
Five models are in common use, and the right one depends on the shape of the work.
- Per record or per document: Best when volume moves around but documents look alike.
- Per field or per keystroke: Best when documents vary widely in length or field count.
- Per hour: Best for messy, exploratory, or research-heavy work where output cannot be counted cleanly.
- Per FTE (full-time equivalent), billed monthly: Best for steady recurring volume where you want the same trained people every day.
- Fixed project fee: Best for one-time backfile conversion with a defined endpoint.
Match the model to the risk you want to hold. A per-hour rate on rules-stable repeating work hands every efficiency gain to the provider. A per-record rate on unpredictable documents hands the variability risk back to you.
Six drivers, roughly in order of impact:
- Source document quality and legibility
- Number of fields per record
- The required accuracy level and the verification method that goes with it
- Turnaround window
- Volume commitment
- Compliance overhead for regulated data
Two of these compound. A rush turnaround and a double key requirement on the same job move a quote by more than either does alone. If a price comes back high, ask which driver is carrying it before you assume the provider is expensive.
Five, and none of them are the headline rate.
- Minimum volume commitment and what you pay if you fall short of it.
- Ramp schedule, so you are not paying the full rate during calibration weeks.
- The rework clause and the accuracy denominator it is measured against.
- Notice period and rate lock duration, which should be stated separately.
- Data return and destruction at exit.
Quarter-to-quarter terms with a mutual notice period are normal in this market. An annual lock with no performance out is not something you have to accept.
By the shape of your work rather than a package tier. Repeatable volume with a stable field map runs per record or per entry. A one-time backfile conversion, data migration or database cleanup runs as a fixed project price.
Four things stay the same whichever model you land on. The accuracy denominator is written into the scope of work. Rework on a failed batch is at our cost. There is no minimum commitment until the sample batch result is in front of you. And the ramp period is priced separately from steady state, so you are not paying full rate while we calibrate. How we scope and quote a project is set out on our data entry pricing page.
04
Security, compliance, and where the work happens
4 questions
Outsourcing sensitive records is safer when the controls are documented and verifiable. Do not rely only on what a provider says in a sales call. Ask for written proof of the controls that apply to your data.
- A signed NDA covering the provider and the named individual operators who touch your files.
- A DPA (data processing agreement) setting out the purpose, retention period and every sub-processor in the chain.
- SOC 2 Type II or ISO 27001. Type II shows whether controls operated over a period, while Type I evaluates whether they were designed appropriately. Ask which one the provider can show you.
- The framework that matches your record type. For PHI (protected health information), confirm whether the provider is acting as a HIPAA business associate and whether a signed BAA (business associate agreement) and appropriate safeguards are required. PCI DSS applies to payment card data, while PII (personally identifiable information) requires controls that match the applicable privacy rules.
- Personal data rules. EU GDPR and UK GDPR may require an appropriate lawful basis and transfer mechanism when personal data moves between jurisdictions.
- Named technical controls. Look for encryption in transit and at rest, role-based access, disabled local storage, restricted removable media and session logging.
Relevance matters more than badge count. Multiple certifications or audit reports do not replace the specific agreements, safeguards and technical controls required for the type of data you are outsourcing.
Whether the work is onshore, nearshore or offshore matters for more than the rate. Three things in particular.
- Data residency and which country's law governs the agreement.
- Overlap hours with your team, which sets how fast an ambiguous record gets a ruling.
- Overnight turnaround, which an offshore or nearshore delivery center may provide depending on its location and overlap hours.
Ask for the delivery location by name rather than the head office address. The two are frequently different, which is not a problem by itself as long as you know which is which and your data processing agreement names it.
If a record type cannot legally leave a jurisdiction, establish that before you build a shortlist, not after.
You own it. Say so in the contract anyway, in a clause that also names derived files and intermediate working copies, not only the final deliverable.
At exit, you want three things in writing:
- Return of all source and output files in an agreed format,
- A defined destruction window covering every copy including backups,
- A certificate of destruction issued afterward.
Ask one more question that buyers usually miss. How long do working copies live on production machines during the engagement, not just after it ends?
Signed NDAs covering the company and every named operator on your account. Role-based access, so each person sees only the fields their task requires. Encrypted transfer and encrypted storage. Handling practices that match the framework governing your record type. That includes HIPAA rules for protected health information, and EU GDPR and UK GDPR rules for personal data.
We work to your security questionnaire rather than asking you to accept ours. If your records need a business associate agreement or a data processing agreement, we sign it before the first file moves. Ask for the written control list and we will send it. The full detail sits on our data entry accuracy standards page.
05
Scope, fit, and automation
3 questions
The label covers four different jobs, and providers price them differently.
- Typing values from a source into a target field. Invoices, applications, order forms.
- Data capture and data extraction. Pulling fields off scanned documents or PDFs. Optical character recognition (OCR) or intelligent character recognition (ICR) reads first, a person verifies second.
- Data cleansing. Deduplication, format standardization, validation against a reference source, data appending for missing fields.
- Indexing and conversion. Tagging documents with searchable metadata, or backfile conversion between formats.
Decide which of the four your project is before you compare quotes. A per-record price for keying and a per-record price for data extraction are not the same number and should never be compared side by side.
For most buyers, the answer is both, with automation first and people on what it cannot read. Optical character recognition and intelligent document processing handle structured, high-volume, low-variation documents well. They degrade on handwriting, poor scans, non-standard layouts, and any field that needs judgment rather than reading.
The model most teams settle on is human-in-the-loop. Automation extracts, confidence scores route the low-confidence items into an exception queue, and a person resolves them.
If you already own an IDP tool and production quality is inconsistent, you probably do not need a different tool. You need an exception queue with a named owner and a turnaround SLA (service level agreement) attached to it.
Four cases where the math does not work.
- Volume is low and irregular. Below a few hundred records a month, coordination costs more than the labor it replaces.
- The task is a decision, not a transcription. Approving an invoice is not data entry. Entering the invoice is.
- Your rules are undocumented. If the logic lives in one person's head, write the field dictionary first. No provider can infer it.
- The records cannot leave a jurisdiction. Check the data residency requirement before you build a shortlist.
06
Working together
3 questions
Three artifacts. Prepare them and setup takes days instead of weeks.
- A field dictionary. Every field, its format, its allowed values, and the rule for what to do when the source is blank, illegible, or contradictory.
- A golden set. Correctly keyed records used for calibration at the start and for QA every quarter after.
- Defined access. Which systems, which credentials, which approval path, and who revokes them at exit.
If your own team cannot write the field dictionary, that is the real project. Two analysts interpreting the same field differently is the most common root cause of accuracy problems, and no external provider can fix it from the outside.
The first month should be a controlled ramp, not an immediate handoff of your full queue.
Start with the field dictionary, golden set and access rules. Run the pilot first and use the results to resolve unclear field definitions. Then agree on the production workflow, exception queue, escalation path, handoff window and turnaround SLA before volume increases.
During the ramp, compare live output against the golden set and record every new edge case in the field dictionary. Do not let verbal rulings become permanent rules. For the full preparation and handoff process, see how to outsource data entry.
Three differences, and all three show up in month two rather than week one.
- A documented field dictionary, built with you during onboarding, so the rules survive any one person leaving.
- A QA layer separate from the operator. A freelancer checks their own work, which is the same pair of eyes twice.
- Continuity cover. Sick days, holidays, and volume spikes are our staffing problem, not a gap in your queue.
A freelancer is cheaper per hour. A managed team is cheaper per corrected record once volume is steady, which is the number that actually hits your operating cost.
Free Data Audit
Not sure which of these steps your process is missing?
Send a 100 to 200-record sample and a description of the workload. We come back with your exception rate, a cost per thousand finished records, and the gaps in your current handoff.
Useful Links
