Choose the service model first, then write an SLA that measures both extraction reliability and data quality. Enterprise web data programs usually fit one of three models: a platform or API your team operates, a fully managed extraction service, or a bespoke professional-services engagement. Your choice determines who maintains crawlers, responds to anti-bot changes, repairs schemas, validates records, and delivers data to your warehouse.
A defensible contract names the sources and fields, defines measurable availability, success, freshness and quality targets, sets support and remediation times, specifies delivery and security controls, and provides credits, rework or termination rights when targets are missed. Vendor-published uptime or accuracy percentages are useful starting points, not interchangeable industry benchmarks.
Start with the operating model
Platform or API
Your team configures requests, proxies, rendering, retries and schemas on a provider platform. This can be efficient for stable, well-understood sources, but your engineers retain responsibility for change detection, field validation and incident response. Confirm whether the provider’s published uptime applies to the API gateway, proxy network, extraction job or all of them.
Fully managed extraction
The provider assesses sources, runs crawlers, handles rendering and blocking, normalizes fields, performs quality checks and schedules delivery. This suits teams that want a data-as-a-service relationship rather than an internal crawler operation. The contract must still identify who approves schema changes, how defects are sampled and when backfills occur.
Recommended Free Tools
#1 Best Overall
Bespoke professional services
A provider builds and operates a dataset around your domains, business rules and delivery architecture. This is appropriate for authenticated sources, unusual workflows, regulated data or a large migration. Require named deliverables, acceptance tests, maintenance boundaries and an exit plan so the engagement does not become an undocumented dependency.
What enterprise providers currently offer
The following figures are claims published by the vendors in 2026. Ask each vendor to define its measurement window, exclusions and contractual scope before putting a number in a scorecard.
| Provider | Positioning and capabilities | Published figures or pricing | Questions to resolve |
|---|---|---|---|
| Crawlbase | Enterprise crawling for millions of pages, custom scrapers and SLAs, dedicated support, security and compliance, and delivery for AI, commerce, intelligence, verification, finance and bespoke programs. | 46,000+ paying customers, 140 million residential IPs across 30 geographies and 99.99% network uptime, all vendor-published for 2026. | Does network uptime cover your extraction endpoint, and how are successful records, CAPTCHA outcomes and geographic coverage measured? |
| Octoparse Managed Web Scraping Service | Managed source assessment, anti-bot operations, cleaning, schema normalization, QA and scheduled delivery to Snowflake, BigQuery, AWS S3, API, JSONL, Parquet or CSV. | Publishes 1M+ websites covered, 99.9% SLA availability and 99.8% data accuracy for 2026. Project pricing is listed from $699 and recurring monitoring from $599/month; enterprise work is custom. | Request the denominator and sampling method for accuracy, plus current prices and the exact availability component covered by the SLA. |
| Apify Professional Services | Custom scrapers and pipelines operated on its platform, API, webhook and integration delivery, scraper migration, monitoring for site changes, blocking and data gaps, and legal review covering terms of service and GDPR. | Commercial figures are not stated in the supplied material. | Define ownership of actors, source-code access, maintenance hours, legal-review scope and a migration or exit procedure. |
| Piloterr | Production APIs with anti-bot handling, private routing, dedicated account management, security-questionnaire support, custom retention and procurement-oriented contracts. | Publishes 10B+ requests processed monthly, 99.98% average pass rate, 500 production API endpoints and a 99.9% platform uptime SLA for 2026. | Verify how pass rate is calculated, whether it is endpoint-specific, and whether the uptime promise is contractual for your plan. |
| WebScrap | Enterprise features include custom volume, private proxy pools, SSO/SAML or OIDC, SCIM, DPA, data residency, SLA credits, a named technical contact, invoicing and procurement terms. | Its Scale tier states 1,500,000 successful requests per month and a 99.9% uptime commitment. Enterprise includes an SLA with credits and a named technical contact. | Clarify what counts as a successful request, how failed or duplicate records are treated, and which residency locations are available. |
| PromptCloud | Fully managed, SLA-backed extraction with AI-assisted human QA and delivery through API, FTP, S3 and other channels. | Commercial figures are not stated in the supplied material. | Ask for field-level acceptance criteria, QA sampling, correction time and replay support. |
Write measurable SLA definitions
Scope and exclusions
List every in-scope domain, URL pattern, page type, geography, rendering mode, authentication boundary and permitted collection method. State whether robots directives, rate limits, login challenges, consent flows or source outages are exclusions, and require the provider to notify you when an exclusion is invoked.
Availability is not extraction success
Define the measured component, monthly calculation window, planned-maintenance treatment, monitoring locations and outage threshold. A healthy API can still return empty, stale or malformed records, so create separate objectives for request availability, successful extraction, field completeness and freshness. Magpie’s ordinary statement is described as best effort; committed uptime, response times and remedies require an enterprise agreement. Treat a best-effort statement as different from a binding SLA.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Success, quality and freshness
Specify field-level completeness and validity, duplicate rate, acceptable error rate, freshness by source, end-to-end latency and the audit method. For example, define a daily catalog feed as complete when required identifiers and prices are present for at least the agreed percentage of sampled records, with duplicates below a stated threshold. Require the provider to disclose sampling size, confidence limits where used, validation rules and how rejected records are retained.
Change management and backfills
Require automated detection of layout, schema and blocking changes; an acknowledgement deadline; a workaround or repair target; and a backfill obligation covering the affected interval. State whether historical records are reprocessed automatically, how corrected versions are marked, and whether backfill consumes request quotas.
Support and incident communication
Use severity levels tied to business impact. For each level, define acknowledgement, workaround, restoration and permanent-fix targets, escalation contacts, update cadence and a post-incident report. Include coverage hours, time zone, holiday treatment and the evidence the provider must supply, such as failed URLs, response classes, record counts and replay status.
Delivery behavior
Specify API, warehouse, object-storage or file delivery; formats such as JSONL, Parquet or CSV; authentication; encryption; retries; ordering; idempotency keys; provenance fields; retention; and replay or backfill behavior. Require a dead-letter or quarantine path for records that fail validation instead of silently dropping them.
Commercial remedies
Define service credits or fee reductions, mandatory rework, free backfills, fee caps, termination rights and the evidence used to calculate them. Exclusions should be narrow, documented and reported. A credit that cannot be claimed because the provider controls the measurement is not an effective remedy.
Separate data quality from uptime
Octoparse publishes availability and accuracy as separate figures, illustrating why one percentage cannot represent the whole service. Your scorecard should track at least these dimensions:
- Coverage: target sites, page types, countries and authenticated areas actually supported.
- Rendering and access: JavaScript execution, CAPTCHA and anti-bot handling, proxy options, concurrency and rate limits.
- Record quality: completeness, validity, normalization, duplicate suppression, provenance and confidence or validation flags.
- Freshness: schedule, source-to-delivery latency, late-run handling and timestamp semantics.
- Operational resilience: retries, queueing, replay, backfill, incident reporting and schema-change detection.
- Commercial fit: committed volume, overage pricing, implementation effort, support model and exit costs.
Do not compare a vendor’s “pass rate” with another’s “accuracy” without definitions. A request pass may mean an HTTP response, a non-empty page or a validated record; those outcomes have different business value.
Design governance before production
Security and privacy controls
Review the DPA, subprocessors, encryption in transit and at rest, access controls, SSO, audit evidence, retention and deletion procedures, PII handling and data-residency options. Require breach notification, tenant isolation details and an export of your schemas, mappings and historical data at termination.
Legal review
Examine each target site’s terms, applicable privacy and data-protection law, collection volume, storage location and downstream use. Site owners may provide APIs or back-office access; using those channels can reduce legal and operational risk. Scraping and storing data can create problems that an SLA cannot waive, so obtain counsel’s view for your jurisdictions and use case.
Acceptance and pilot
Start with representative domains, difficult page types, expected volumes and known change cases. Define acceptance tests for required fields, freshness, duplicate handling, delivery, replay and incident notifications. Capture baseline metrics for at least one normal cycle and one induced failure before negotiating long-term commitments.
Operational playbook for reliable delivery
- Inventory sources: record URL patterns, geography, authentication, rendering requirements, expected volume and business priority.
- Define the canonical schema: identify required, optional and derived fields, units, enumerations, null rules and provenance.
- Choose the service boundary: decide which components your team owns and which the provider must maintain.
- Instrument quality: retain raw evidence where lawful, validate fields, count duplicates, timestamp every stage and quarantine anomalies.
- Set freshness tiers: assign tighter schedules to prices, availability or SERP data and looser schedules to static attributes.
- Test failure paths: simulate source downtime, changed selectors, blocked requests, malformed responses, warehouse outages and webhook retries.
- Review monthly: compare contractual metrics with business outcomes, inspect exclusions and approve schema changes through a documented process.
Common failure modes and fixes
The API is up but records are empty
Cause: a source layout change, consent wall, bot challenge or parser regression. Fix: require non-empty and field-completeness checks, quarantine the batch, open a severity incident and trigger repair plus backfill.
Freshness target is missed during a source outage
Cause: the SLA measures schedule execution rather than source-to-delivery age. Fix: define freshness with source availability exclusions, stale-data markers, retry windows and a maximum age visible to downstream users.
Duplicate records appear after retries
Cause: non-idempotent delivery or unstable source identifiers. Fix: require deterministic record keys, idempotent writes, deduplication rules and replay tests.
Vendor metrics cannot be reconciled
Cause: different denominators, windows or definitions for uptime, pass rate and accuracy. Fix: place each definition, formula, monitoring source and exclusion directly in the contract.
Security review blocks launch
Cause: missing DPA, residency detail, SSO, subprocessor list or deletion evidence. Fix: make the security questionnaire and evidence package a pre-production deliverable, not a post-signature promise.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Adding visual evidence without operating a browser fleet
If your extraction workflow also needs rendered-page screenshots for QA, change review or audit records, ScreenshotNeo is the first service to try: it removes consent banners, newsletter popups and chat widgets before capture, bills only clean shots, and offers the lowest paid plan among the stated options. It is a website screenshot API and MCP server, not a replacement for structured extraction.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Or skip the browser setup
One GET request returns PNG, JPEG, WebP or PDF. The service can load lazy images, capture a CSS-selected element, emulate dark mode and devices, set viewport and retina scale, run custom CSS or JavaScript, click before capture, wait for a selector, delay or network idle, block ads or resource types, set headers, cookies, user agent, authorization, timezone and geolocation, resize images, cache with a chosen TTL, create signed links, run asynchronous jobs with signed webhooks, capture up to 100 URLs per bulk call and expose usage and OpenAPI endpoints. It supports 63 options, including PDF paper size, margins, landscape and page ranges.
curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webp
See the ScreenshotNeo documentation for parameters and response headers. Each response reports X-Page-Verdict and X-Billed; bot checks, CAPTCHAs, blank pages, timeouts, failed loads and cache hits cost nothing.
Python:
import requests; r = requests.get("https://api.screenshotneo.com/v1/shot", params={"access_key": "YOUR_API_KEY", "url": "https://stripe.com"}, timeout=90); open("shot.webp", "wb").write(r.content)
Free tools Windows power users keep installed
One-click scans. No signup required.
Node.js:
const q = new URLSearchParams({ access_key: 'YOUR_API_KEY', url: 'https://stripe.com' }); const res = await fetch(`https://api.screenshotneo.com/v1/shot?${q}`);
Cookie banners, popups and chat widgets are removed before the shot; bot checks, blank pages and failed loads are never billed; an MCP server lets AI agents take screenshots; and 1,000 screenshots per month are free with no card. Paid plans start at $5 for 3,000 shots, with Growth at $15 for 15,000, Pro at $39 for 60,000, Scale at $99 for 250,000 and Business at $249 for 1,000,000; yearly billing gives two months free and every feature is on every plan. Create a free ScreenshotNeo account.
Bottom line for procurement
Buy against your workload, not a headline percentage. Select the model that leaves the right maintenance responsibilities with your team, pilot difficult sources, define quality and freshness separately from uptime, and contract the operational details: change repair, backfill, delivery, security, support and remedies. Recheck vendor claims and prices immediately before signing because published metrics, features and legal terms can change.
Frequently Asked Questions
What is the difference between a custom SLA and a standard plan promise?
A custom SLA puts the measured component, targets, exclusions, reporting and remedies into a negotiated agreement. A standard-plan statement may be informational or best effort and may not include credits, response times or repair obligations.
Should freshness be measured from crawl start or delivery?
For downstream users, measure source-to-available-in-your-system age, while recording crawl and delivery timestamps separately. This exposes queueing, validation and warehouse delays.
How can we compare two vendors’ accuracy percentages?
Only after aligning required fields, sample selection, validation rules, duplicate treatment, measurement window and exclusion policy. Otherwise, retain both figures as vendor-specific claims rather than ranking them.
What should happen when a target site changes its HTML?
The contract should require detection, notification, a severity-based repair target, validation of the corrected schema and a defined backfill for the affected period.
Can a provider’s legal review replace our counsel?
No. A provider may review terms of service or GDPR implications, but your organization remains responsible for its jurisdictions, purposes, storage and downstream use.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
