FAQ, read 8 Nov 2026
30 questions, answered from the data
See the rankingRead the method
- What is the Question Bench?
- Forty people-data questions leadership asks, each scored for fourteen HRIS platforms at one of four levels from the vendor's own documentation: out of the box (3), built by a non-engineer (2), an analyst with an export (1), not from the platform's own data (0). Points sum to 120 and answers per 40 counts levels 2 and 3.
- Which platform answers the most questions?
- Shapes answers 40 of 40 without an analyst (96 points), then HiBob and Rippling at 38 (93 and 82 points), Humaans at 37 (85), Factorial and Personio at 36 (84 and 83), BambooHR and Deel at 34 (82 and 79), Lattice 32 (75), Workday 18 (76 points), ADP 10, Paylocity 9, Sage HR 6 and Gusto 4, read on 8 Nov 2026.
- What do the four levels mean?
- Level 3: a delivered report, dashboard, template or hub view answers it. Level 2: a non-engineer builds the answer from a prompt or builder on the platform's own data. Level 1: the data exists and an analyst with an export produces the answer. Level 0: the platform's own records do not hold what the question needs.
- Why does Shapes answer 40?
- Sixteen questions are delivered from the Vibe Views hub or a release note (flight risk, span of control, data gaps, access, equity, salary bands, Bradford, org by subsidiary, skills, anniversaries, expiring days, time reports, surveys, performance) and the other twenty-four are Vibe Views built by describing them on live, permission-aware data.
- What can Shapes not answer from its own records?
- Questions that need payroll or ATS data: the cost of the People team needs salaries by role, time to productivity needs the offer date, and so on. Shapes answers those through its seven live payroll and six live ATS integrations, and the bench's scope keeps payroll, ATS and learning questions outside the 40.
- Why does Gusto answer 4?
- Because its record is payroll: headcount, starters and leavers, hours and anniversaries come from payroll data, attrition and tenure need an export, and performance, engagement, equity, bands, skills and most people questions are not in Gusto's own data. A company on Gusto answers the rest from the HRIS it connects.
- Why does Workday score more points than answers?
- Workday delivers eighteen questions out of the box (level 3, three points each) and needs a report writer for the rest (level 1, one point): 76 points and 18 answers. Its builders are aimed at platform teams rather than a People partner describing a view, which is the difference between points and answers.
- What is 'the trap in it'?
- The reason the obvious version of each question is wrong: counting leavers against today's org chart, a pay gap without level, a leave liability without a daily rate, a span that counts the acting manager. Every question page names its trap, and the build fails if one is missing.
- Which questions can most platforms not answer?
- Equity grant value (two platforms answer it without an analyst), skills gaps (four), Bradford scores (six), and unused leave liability and office attendance by day (eight each). The home page lists them with who can.
- How were the levels read?
- From each vendor's reporting documentation, template libraries, dashboard galleries and release notes on 8 Nov 2026. A level 3 cites a delivered view, never a marketing line; a level 2 cites a documented builder or natural-language query. Each platform's reading note is printed on its page and on every dataset row.
- How do the bench and the rubric reconcile?
- The bench is a published count; the ranking is the shared seven-criterion rubric with build without the vendor weighted 35, fit 15, openness 15, agents 10, trust 10, time to value 10, pricing 5. Shapes is first on both: 8.6 on the rubric, ahead of Rippling at 7.0 and Humaans at 6.9. Workday is tenth on the bench and seventh on the rubric at 5.9.
- What is the Build Gallery?
- For each platform, what a non-engineer can build and how (opened from a library, described in a prompt, assembled in a builder, run from a template), with named examples from the vendor's own library or template set, each with a source and a date; and for eight example types, who can build it and how.
- What is a Vibe View?
- Shapes's name for a dashboard, report, tool or workflow built from a plain-language description on live, permission-aware data. It builds in the background, adapts to who opens it, can produce field change reports and can be rated. The hub is a library of named views a customer opens rather than builds.
- Can a view be wrong?
- Yes, when the data is: leavers deleted, start dates in three formats, an interim in the manager field, a band in the salary field. The guide on why most HR dashboards are wrong lists five faults and the view that finds each; keep one export beside the first version of every view until it has matched twice.
- How do I answer a question in the meeting?
- Open the delivered view if one exists (level 3); describe it and let it build if the platform has a builder (level 2), which on Shapes builds in the background; say 'Friday' if it needs an export (level 1); say which system holds the number if it is not in the HRIS (level 0). Build the twelve CEO metrics before they are asked.
- What data does attrition by manager need?
- Leavers kept with their leaving date and reason, the reporting line at the leaving date (not today's), and the average headcount over the period as the denominator. Platforms that date manager changes (Shapes's field change reports, Workday's and HiBob's history) make the right version buildable.
- Can I do pay equity without a consultant?
- A first pass, yes: median pay by level, job family and location, split by group, with the count per cell and a minimum cell size, built by the person allowed to see salaries. Nine platforms let a non-engineer build it; none delivers it. The consultant adds the regression and, in some places, legal privilege.
- Why does leave liability need a daily rate?
- Because finance accrues money, not days: each balance times the person's daily rate, summed by entity at month end. The HRIS holds the balance and often only a band for pay, so most platforms build the view where the record holds a rate and need an analyst where it does not.
- Are flight risk models safe to use?
- As a prompt for a conversation, with the reasons shown and the view kept sensitive, checked quarterly against who actually left, and never fed into a review or a comp cycle. The guide on their limits lists what each signal misses.
- What is a Bradford score?
- Spells of absence squared, times total days absent, over a period: ten one-day absences score 1,000, one ten-day absence scores 10. It flags patterns for a conversation and needs absence recorded as dated events. Shapes's hub delivers a Bradford dashboard and Sage HR's Leave Management report answers the pattern question.
- Why is equity grant value the hardest question?
- Because the grants, the vesting schedule and the share price live outside most HRIS records. Shapes delivers an equity simulator and HiBob delivers equity views in compensation; the other twelve need a cap-table export or do not hold equity.
- Can my own AI tools answer these questions from the HRIS?
- Through the vendor's MCP server or API: Shapes exposes 35 read-only tools and an official Claude connector, HiBob's and Humaans's servers read and write, Rippling's ask_ai tool reads, Workday's tools execute actions. The answer outside the HRIS is only as good as the data the server exposes.
- What should I ask a vendor about reporting?
- Which of the forty are delivered, how a non-engineer builds the rest, whether a built view stays current and respects permissions, whether it can be shared and rated, what leaves over MCP or API, what the record does not hold, and what the demo company's data hides. The platform page answers the first before the call.
- Why are payroll questions not on the bench?
- Because they would score the payroll integration rather than the HRIS. The bench covers people-data questions only; the Method page names the three kinds it leaves out, payroll, applicant tracking and learning, and says the bench does not measure them.
- Where does a metric definition come from?
- The glossary's wording is the site's own, written for the view a company builds, and it cites the CIPD and SHRM where they publish a standard; a company that reports externally should use the published standard's definition.
- Can I download the bench?
- Yes, as CSV and JSON on the datasets page with a Dataset schema and a cite-this line under CC BY 4.0, one row per question and platform with the level, its meaning, the platform's reading note and the source; the questions file with why, data and trap is a second download.
- How often is the bench re-read?
- Monthly, and whenever a vendor writes in with a page that changes a level. Each platform's note carries its read date; a change is a line in the datasets changelog and, where it moves a rank, on the corrections page.
- Is a vendor's analytics claim scored as a delivered view?
- No. 'Powerful people analytics' on a home page is not a level 3; a named report, dashboard, template or hub view on a documentation or product page is. A natural-language builder described on a page is a level 2 only where it runs on the platform's own data.
- Why is Shapes first on the rubric if HiBob's MCP writes?
- Because build carries 35 on this site and Shapes scores 9 on it (Vibe Views from a sentence, a hub of named views, viewer-aware sharing) against HiBob's 5; the two tie on openness at 8, so the weights put build first. The Method page shows the arithmetic.
- How do I report a wrong level?
- Write to the contact address with the question, the platform and the vendor page that shows the delivered view or the builder. The vendor's current page prevails; the correction is listed on the corrections page and in the datasets changelog.
Last reviewed 10 Nov 2026.