Methodology: How the LeadChange Score Works

The LeadChange Score summarizes how well a product covers the needs of a software or service category, based only on facts we have verified. This page publishes the full model. The same model is implemented in the system that builds our pages. This is version 1.5; every change to the model and its rules is dated under "Version history" below, and our rankings, comparisons, cost guides and product pages name the version they were built with.

• Scale: 0 to 10, shown with one decimal and a confidence label (High, Medium or Low).
• Published only with enough evidence: a product gets a number when at least 60% of the items we score are verified and its confidence is Medium or High. Otherwise it shows "Not scored yet" with how many of its facts are verified.
• Scope: computed per category. A product listed in two categories can have two different scores, and a product built for another job scores lower than in its own category.
• Inputs: verified facts only. No opinions, no user ratings, no commercial data.
• What it is: an editorial assessment based on public information. Every score is shown with that label, on a page that gives the date its facts were last checked ("Facts checked").

A fact is verified when all of the following are true:

• It is specific. A price, a plan limit, a certification, a supported integration or a capability, not an opinion or a marketing claim such as "easy to use".
• It is on a public page we can cite. We keep the exact address of the page the fact was checked against, the date of the check and the value we read.
• The source is the right kind. A vendor's own pricing page, documentation, help center or trust center counts as primary. Anything else counts as secondary and lowers the product's confidence (see "Sources hierarchy" below).
• It is current. A newer value replaces an older one, and the older value is kept in the fact's history with the date it changed.

A capability counts as present only with evidence that the product offers it, and as absent only with evidence that it does not. Silence on a vendor's site proves nothing either way, so it is never scored.

Every sub-criterion scores between 0 and 1, and a dimension is the average of its verified sub-criteria. A sub-criterion we could not verify is left out of the average; it never counts as zero (see "Missing data" below).

• Capability coverage: the share of the category's core capabilities the product offers, among the capabilities verified either way. Our research records the capabilities a product verifiably offers; where we have not yet verified that a capability is absent, this dimension has no verified sub-criterion and is left out, and the verified capabilities count toward coverage instead.
• Pricing transparency & value: three sub-criteria. Transparency scores 1 for fully published prices, 0.5 for partly published prices and 0 for quote-only pricing. A verified free trial scores 1 and a verified absence of one scores 0. Value compares the list cost per month for 25 employees, computed from the published prices of plans sold to a company that size, with the category median: 1 at or below the median, otherwise the median divided by the product's cost. Value is left out when the cost cannot be computed.
• Integrations & API: five sub-criteria. Open API, webhooks, SSO and SCIM score 1 each when verified present and 0 when verified absent. Native integrations score the product's count divided by the highest count in the category.
• Security & compliance: SOC 2 Type II, ISO 27001 and a GDPR program score 1 each when verified present and 0 when verified absent. HIPAA is added as a fourth sub-criterion only in categories that handle health data: benefits administration, employee wellness, ACA compliance, PEO services and COBRA administration.
• Fit & support: company sizes served (small business, mid-market, enterprise) divided by 3, support channels (up to 3) divided by 3, and mobile apps.
• Data confidence: critical facts verified on primary sources, divided by the critical facts that apply to the category.

When a whole dimension has no verified sub-criterion, it is left out and the weights of the other dimensions are scaled up in proportion, so they still add up to 100%.

Some categories are outsourced services run by the provider's own staff and systems, not software you operate, so several software checks say nothing useful about them. This applies to PEO services, ASO services, COBRA administration, HSA and FSA administration, ICHRA administration, unemployment claims management and relocation management. Payroll services, employers of record and background check services are used through software the buyer operates, so they keep the software checks. Service categories keep the same six dimensions, the same weights and the same rules for missing data, but score service facts instead:

• Integrations & data feeds: an open API and electronic file feeds score 1 each when verified present and 0 when verified absent. Named integrations score the number of platforms the provider names, up to 3, divided by 3.
• Security & compliance: a SOC 1 Type II or SOC 2 Type II report scores 1; a verified statement that the provider has neither scores 0. HIPAA is scored as in other health-data categories. A public registration scores 1 when the provider is on the register and 0 when we checked the register and it is not; we check the IRS list of certified professional employer organizations and ESAC's list of accredited members.
• Fit & support: company sizes and support channels are scored as in other categories. Published support hours and a published implementation time each score 1 when the provider publishes them, and are left out when it does not.
• Critical facts: pricing model, price transparency, plans, company sizes served, services offered and support channels, plus the facts that define the category (for PEOs: payroll tax filing, workers' compensation, benefits offered and IRS certification; for COBRA administrators: HIPAA and support hours). Starting prices are not critical here: a provider that prices by quote has none to verify, and its quote-only pricing is itself the verified fact.
• Comparison tables in a service category compare the published price and the facts that decide the choice in that category (for PEOs: IRS certification, ESAC accreditation, workers' compensation, pricing model, states served and minimum headcount; for COBRA administrators: state continuation, online elections, premium collection, carrier updates, participant support hours, the SOC report, HIPAA, file feeds and implementation time). A column appears when at least half the providers have a sourced value. When a provider's public pages say nothing about it, the table shows "Not published" instead of guessing.

Many products are listed in several categories, but each was built for one main job. A module of a product built for another job should not outrank the products built for this one, so the category score is multiplied by a category fit:

Products with a secondary fit show the reason next to their score, for example "Primarily an applicant tracking system; HRIS module is secondary." Awards such as "Best overall," "Best value" and the best pick for a company size go only to products with a fit of 0.85 or more. The best pick for a company size also needs the product's own "Best for" description to name that size, or no size at all. In payroll software, a product whose listed plan does not file payroll taxes stays in the ranking with the label "Doesn't file payroll taxes" and its score unchanged, but is never featured: no award or top pick, no place in a hub's top list or the home page's rankings, and no recommendation in the matching flow.

The category score is 10 times the category fit times the weighted sum of the six dimensions, rounded to one decimal. Products with equal scores are listed alphabetically. Products that are not scored yet are listed after every scored product, the best-documented first.

When a vendor sells plans for different jobs, a category page shows only the prices of the plans for that category's job. For example, an employer of record page shows a vendor's EOR plan, not its cheaper recruiting plan. A product listed in several categories shows no price in a category for which it publishes no plan.

• "See plans" and "Quote only": in a price table, "Quote only" means the vendor publishes no prices at all; "See plans" means it publishes prices that are not a monthly starting price for that table (another plan line, a yearly or per-check price).
• Who may buy a plan: a plan the vendor sells only to some company sizes, for example "businesses with fewer than 20 employees" or a plan for the owner alone, has no cost outside those sizes: cost tables show "Not offered" there, and it is left out of the cheapest-price answers.
• Flat fees: a flat monthly fee up to a headcount is that fee alone up to it, never the fee plus the per-employee rate.

Pages for a specific segment, such as "for small business" or "for enterprise," list only products with a verified fact showing they serve that segment. They also explain the requirements that matter most to the segment, and they reweight the dimensions toward what it needs:

Enterprise pages use their own weights instead of the company-size row. Every segment page lists the weights it uses.

A product without a number on a segment page shows "Not scored for this segment": it may have a score in its category, on its own page.

Missing data is never assumed, in either direction. An item we could not verify is not counted as present and not counted as absent: it is left out of the score and lowers the product's coverage, the share of the items we score (core capabilities and sub-criteria) that are verified. Every score is shown with its confidence label:

• High: at least 80% of the critical facts are verified on primary sources.
• Medium: at least 50% are; the rest are missing or rely on secondary sources.
• Low: fewer than 50% could be verified on primary sources.

A product gets a number only when its coverage is at least 60% and its confidence is Medium or High. Otherwise it shows "Not scored yet" with how many of its facts are verified, it is listed after the scored products and it receives no award. This keeps a score about the product, not about how much of it we have been able to research.

Payments, affiliate commissions and lead fees are never inputs to the score, and this is enforced in the system that builds our pages, not only promised:

• The score reads a fixed list of facts. The scoring code takes the verified facts listed on this page and nothing else. An automated test fails our build if that code ever refers to offers, affiliate links, commissions, leads or buyers.
• Rankings sort by score, then by name. Products with equal scores are listed alphabetically. Nothing else changes the order.
• The publishing system cannot see lead data. The database role that builds our pages has no access to requests sent through our forms, to the providers that receive them or to what they pay.
• Commercial links are labeled. The only place commercial data appears on a page is a "Visit" button that may earn us a fee or commission, and every such button carries a visible "Sponsored" label.

Whether a vendor pays us has no effect on its score or its position in rankings. See How we make money.

• Scores update with their facts. When an underlying fact changes, for example when a vendor changes its pricing or earns a certification, every score and page that uses the fact is rebuilt. The change appears in the change log of the product's own page, and price and plan changes also on its pricing page, each with the date.
• One "Facts checked" date per page: the most recent date the facts the page shows were checked against their sources. A page also shows "Updated" (the last time its content, a fact value or the set of ranked products actually changed; a reordered ranking is not a change) and, when different, "Facts checked".
• When facts are re-checked: every week, a scheduled check re-reads the sources of the facts checked longest ago (prices first) and records, for each fact, whether its source still says the same; also when we update a product's research and when we act on a correction request. A re-check that finds no change moves "Facts checked", never "Updated"; a fact whose source changed is corrected through the same sourced process before the page changes.
• Model changes are dated. Any change to the dimensions, weights or rules on this page is listed under "Version history" below.

Before a page is marked as finished, an independent automated review checks it. The reviewer did not write the page, has no access to how it was written and cannot change it; it reads the published page, the facts behind it with their source quotes, how the same products appear on their own pages, the sources the text cites and the top search results for the page's question. It scores six criteria: whether the page answers the question better than those results, accuracy against the facts, consistency, freshness, readability and honest wording. Each problem it finds is quoted, classed and given a fix.

• Veto: a problem a buyer could act on (a wrong or unsupported price, legal or security claim, a contradiction, a promise the site does not keep) vetoes the page, and it cannot be marked as finished until it is fixed and passes a new review.
• Confirmed before it counts: a serious problem the reviewer raises is checked again by a second, focused review of the same evidence. It counts when that check confirms it; it is set aside only when the check quotes the fact or the page words that answer it. A confirmed false or unsupported statement always counts.
• Two levels: a page is Verified, and can be marked as finished, when no confirmed serious problem remains. It is Top when it is also judged to answer its question better than the top three search results, with no important gap left. Smaller gaps go to the page's list of improvements and do not reopen it.
• Follow-up reviews: after a page is corrected, the next review checks each earlier finding (fixed or not) and reports new problems only when they are serious. A page whose text changed by more than 30%, a new page and a weekly random sample of pages get a full review again.
• Recorded: every review is kept with its date, version and findings, and each page shows its review status under "How this page was made".
• People: the review is automated; no page is reviewed by a person yet.
• What it is not: the reviewer is not a human expert and does not test products. No page is signed by, or claims the review of, a named expert.
• Fixed at the source: problems that recur across pages are fixed in the system that builds them, not page by page. The first review of 30 finished pages found prices shown as "Quote only" for vendors that publish them, and plans priced at company sizes they are not sold to; both are fixed and listed on the corrections page.

• Not a user rating. It does not measure satisfaction or reviews.
• Not a hands-on test. It is an editorial assessment of public information. We do not claim to have tested products we have not tested.
• Not a recommendation for every company. A high-scoring product can be the wrong choice for your size, industry or budget. Read the "best for / not for" notes on each page.
• Not a verdict on a vendor. A low score or "Not scored yet" usually means a product publishes less of what we score, or that our research is still in progress.

Some products are also tested hands-on. A tested product's page shows the test under "Tested by LeadChange Research", with its date, the plan used and a screenshot behind each finding. Pages of products we have not tested say so.

• How a test is run: LeadChange Research opens a free trial in the product's own app and follows a written protocol: add people, set up jobs or records, build a schedule, track and approve time, run reports and exports, and look at the billing page. Each step is logged with its time, and clicks are counted where they matter.
• Demo data only: every employee, customer and job is fictional and labeled as such, and invitations go only to our own research email addresses.
• What we never do in a test: pay for a plan or add a card, contact the vendor's sales or support team, connect a real integration, send anything to a real customer, or use real people's data.
• What we publish: the steps, the measurements with their caveats, the findings, what we did not test and the prices the product's billing screen showed. Screenshots are cropped to the part that matters; nothing the product shows is edited. A price seen in the app appears beside the published price of the same plan, each with its date.
• Tests and the score: a test does not change the LeadChange Score, which reads verified public facts only. What a test observes is stored as a fact whose source is our own dated screenshot, next to the facts from the vendor's pages.
• Dated and re-run: every test carries its date. When the product changes in a way that affects a finding, we test it again and date the new test.

• Sell a score, a rank, an award or a "best for" label, or change one for a fee.
• Count a feature we could not verify, in either direction.
• Use undisclosed sources, anonymous tips or private demos as scored facts.
• Invent ratings, reviews, test results, customer counts or experts.
• Describe a vendor or product in mocking or sweeping negative terms. Weaknesses are stated as facts and fit, for example "no published pricing" or "less suited to companies with hourly staff".
• Give a vendor an advance look at, or a veto over, its score.

When sources conflict, we rely on them in this order:

1. Primary vendor sources: official pricing pages, product documentation, help centers, trust centers and published certifications.
2. Official government sources: for example, agency guidance, public registries and regulatory publications.
3. Reputable secondary sources: used only when no primary source is available, and always recorded as secondary.

Facts supported only by secondary sources count toward lower confidence.

We record the date each fact was last checked against its source. A weekly scheduled check re-reads the sources of the facts checked longest ago, prices first; facts are also re-checked when we update a product's research and when we act on a correction request. Every page shows the date its content last changed and the date its facts were last checked.

Each fact is stored once and reused everywhere it appears on the site, with its value, the document it was verified against, the date of that check, a confidence level and its history of previous values. Because each fact has one value, a product's price or certification cannot differ between pages. When a fact changes, every page that uses it updates and records the change in its change log.

Where we show ratings or reviews from third-party sites, we quote them as published and name the site they come from. They are never mixed into the LeadChange Score and never included in our structured data.

• Version 1.5, October 31, 2026: a product without a number on a segment page reads "Not scored for this segment" (see "Segment pages" above). Price tables say when a payroll price is the cheapest plan that files the taxes and name any priced plan its vendor does not sell at the table's headcount; an annual-billing rate is labeled beside its figure. A check that a product falls short of a capability's standard reads "Verified as not meeting the standard". Weights and the scoring rules are unchanged.
• Version 1.4, October 30, 2026: payroll products are compared like for like: rankings, scores and comparisons use each product's cheapest plan that files the payroll taxes, and plans that leave tax filing to you are listed apart under "You file the taxes yourself" (Patriot Payroll and SurePayroll rank at their full-service plans; their own pages still show every plan, labeled). A plan's user limit and seat minimum are read from its own billing line, which changed one score: Timeero went from 7.8 to 7.5 in time and attendance software, because its Basic plan is sold for up to 10 users. The independent review has two levels, Verified and Top, and counts a serious problem only once a second check confirms it (see "Independent review" above). Weights are unchanged.
• Version 1.3, October 26, 2026: an independent review must pass a page before it is marked as finished, with a veto (see "Independent review" above). Price tables say "See plans" for a vendor that publishes prices on another basis and keep "Quote only" for vendors that publish none; a plan sold only to some company sizes is "Not offered" outside them; a flat fee up to a headcount is never added to the per-employee rate. Value now uses only plans sold to a company of 25 employees, which changed one score: Paychex Flex went from 8.1 to 7.2 in payroll software, because its one published price is for businesses with fewer than 20 employees. This page now says plainly that there is no scheduled re-check of every fact yet. Weights are unchanged.
• Version 1.2, October 26, 2026: scores no longer show a date of their own: each page shows one "Facts checked" date. A hub whose top product leads the next by more than 1.5 points shows both products' scores by dimension. Hands-on testing is described and labeled on the pages of tested products. Weights are unchanged.
• Version 1.1, October 25, 2026: service categories (PEO, ASO, COBRA, HSA and FSA, ICHRA, unemployment claims and relocation) score service facts in place of the software checks: the free trial, webhooks, SSO, SCIM, the native integration count, ISO 27001, GDPR and mobile apps give way to named integrations, electronic file feeds, a SOC 1 or SOC 2 report, public registrations (IRS CPEO, ESAC), published support hours and implementation time; their comparison tables compare service facts. Weights and every software category are unchanged.
• Version 1.0, October 7, 2026: items we could not verify are left out of the score instead of counting as zero; a product needs 60% coverage and Medium or High confidence for a published number ("Not scored yet" otherwise); capability coverage counts toward coverage until capability absence is researched; every score shows its confidence label, the date its facts were last checked and the label "Editorial assessment based on public information".

If you believe a fact or score is wrong, use the correction request form. We review every request within 5 business days, and a fact changes only with public evidence. A score changes only when a fact it uses changes. See Corrections for how we handle requests.