Twenty-two products, read in full
The half of the hire that nobody else pointed an AI at.
In September 2026 we read the public feature and packaging pages of twenty-two products across HR, CRM and applicant tracking — the ones a growing business actually shortlists. Not to score them. To find out what the category had decided not to build.
Seven things came out of it. This page is those seven, in order of how hard they are to copy, followed by what a stack costs you while you wait and the places a specialist still beats us. No vendor is named, no price appears, and the last section is where a specialist beats us.
One · the thing none of them built
Every one of them points the AI at the candidate. The uncertainty is on the other side of the desk.
A shortlist is only as good as the brief behind it, and the brief is almost always the weakest document in the process — three bullet points a hiring manager wrote between meetings, then interpreted differently by everyone who reads it.
So the AI here is pointed at that. It works out what it still does not know about the role, and it asks — a real back-and-forth with the person hiring, not a form with more fields on it. What it learns becomes a versioned understanding of the role, and the moment a human confirms an answer it stops asking about it.
Every product in the study screens candidates well. Not one of them does this, and it is the reason a shortlist here opens per requirement instead of arriving as a score.
Pointed at the candidate
Every product in the study does some version of this, and several do it very well.
- Reads a CV and pulls the fields out of it
- Ranks applicants against the role
- Screens by chat or by voice
- Re-surfaces people you saw before
- Drafts the outreach
Here too. This half is parity, not an argument.
Pointed at the person hiring
Found in none of the twenty-two products read in September 2026.
- Works out what it still does not know about the role
- Asks the person hiring — a conversation, not a form
- Keeps what it learns as a versioned understanding
- Stops asking once a human confirms an answer
- Refuses the questions it must not ask about anyone
A brief nobody can argue with is worth more than a shortlist nobody can check.
Two · where the AI actually is
Not one assistant bolted on the side. Seven places it does the unglamorous work.
Every product in the study has an AI story, and most of them are the same story: a panel that drafts something, and a score. The interesting question is not whether there is AI. It is whether it is pointed at the parts of the week nobody enjoys, and whether you can check what it did.
Everything below is answerable. A finding that cannot be traced to its source is dropped rather than shown, which is the opposite of how a confidence score behaves.
It reads the document and shows its working
Every extracted field carries the sentence it came from, so it is confirmed in a second rather than re-read. Correct one and the correction sticks — your phrasing, your clients' job titles, in your workspace.
A verdict that opens per requirement
What the role asked for, what the evidence shows, and the gaps stated plainly — instead of one percentage nobody can argue with.
The interview, returned as evidence
Quoted and timestamped, with each panellist's rating sitting beside what the interview itself produced and the rows furthest apart marked before the debrief.
Calls that come back as commitments
Objections, commitments and competitor mentions with the second each was said at — and coaching measured against your own team's medians rather than a vendor's idea of good.
Questions it refuses to answer
A search that asks about gender, age, caste, religion or disability is refused rather than answered, and you can read the query it built before it runs.
Ask your own data a question
Answers drawn from your workspace's own material rather than the open internet, in the same product as the records they are drawn from.
Three · in-suite at no tier, anywhere
Seven capabilities that every CRM in the study sends you elsewhere for.
Not held behind an upgrade. Not an add-on. Absent from the product at every tier, because the vendor's answer is a second product — theirs or somebody else's — with its own console, its own customer record and its own renewal date.
These are areas here, on the record the deal is already on.
Contracts signed where they were written
Signature is not a second vendor. What gets signed is the version the customer saw — re-rendered and re-hashed at the moment of signing, so an edit after the link went out cannot be signed against.
Commissions, with a statement the rep can open
Plans, accruals and participant statements. The study found no mainstream CRM that ships this at any tier.
Billing that satisfies an auditor
Numbered invoices that cannot be edited, credit notes, subscriptions, dunning and ageing receivables — with e-invoicing where a jurisdiction requires it, and an export to the ledger your accountant already uses.
Customer success as a module, not a report
Health scores you can open, renewals that become deals, surveys and onboarding. Elsewhere this is a separate software category.
Support and the answers behind it
One queue with SLAs and escalation, and a knowledge base in the same product as the records the answers are drawn from.
Calls as evidence, with a dialer
Recorded, quoted, timestamped and coached — in the CRM rather than in a conversation-intelligence tool bolted onto it.
Delivery that shares the customer
A timesheet line knows the invoice line it became, and an accepted milestone with nothing billed against it is surfaced rather than discovered at quarter end.
Four · the hiring side, in full
The parts of recruitment that usually need a second system.
Agency work is where applicant trackers thin out. The study found the commercial half — what a client owes, what a vendor submitted, what the bench is costing — split across separate products or missing entirely.
The bench, priced daily
Who is idle, what it has cost so far, and the open requirement that could end it — ordered by money rather than by name.
Client and vendor portals
Their own login, scoped to what you assigned them, with submissions and invoices arriving through it rather than through your inbox.
Campus and bulk hiring
A drive register with audited bulk enrolment — cohort to pipeline to hire, without re-importing anybody.
Consent, retention and erasure as screens
Consent captured as an artefact, retention set per client, and erasure that runs — operable screens rather than a policy page.
Five · absent everywhere
Six modules that were in none of the twenty-two.
Comparison tables get written around the features everybody has, because those are the rows that line up. These six have no column to sit in — no product in the study ships them, so no grid asks about them.
Each is a module here with its own screens, permissions and trail. Not a document template, not a custom field somebody configured, and not on a roadmap.
POSH case management
An internal committee workflow with confidential case files and the statutory annual position — where the rest of the category publishes a policy template and leaves you an inbox.
Contractor and gig lifecycle
Contracts, payments and an invoice submission route for the people who are not on payroll and never will be. Nobody else treats them as first class.
A referral tree that pays every hop
Second and third-degree referrals split the bounty along the chain by weights you set. Every other referral feature in the study is flat.
The bench as a priced screen
Who is idle, what the idleness has cost so far, and the open requirement that could end it — ordered by money rather than by name.
Fairness as a screen, not a policy
Outcome distributions across self-declared protected classes, aggregate only, on a route a compliance officer reads and a recruiter does not.
Compensation revision cycles
Bands, budgets and revision letters as a cycle that runs — with the letter signed in the platform rather than routed to a signature vendor.
Six · what the category charges for
The same nine capabilities, held behind the same upgrade.
Every system in the study ships records, leave, attendance, payroll and a self-service portal at entry. That is table stakes and we are not going to pretend otherwise — it is the floor, and everybody clears it.
What is interesting is what sits above the floor. The capabilities that change how a team actually works — reviews, learning, surveys, a helpdesk, analytics, automation, an API — are held back from the tier most teams start on, in all six of the HR systems we read. Not some. Six of six.
Where these sit across six systems · Sep 2026
Sold as an add-on
A separate line on the invoice, or a separate licence
- Expense management
- People analytics
- API and webhooks
Top tier only
The plan above the one most teams start on
- Performance, OKRs and 360s
- Learning
- HR helpdesk
- Workflow automation
- Face or geo-fenced attendance
Middle tier
One upgrade in
- Engagement surveys
Entry tier
What everyone gets on day one
- Records, leave, attendance, payroll, self-service
Where they sit here
People & payroll — one area
- Performance, OKRs and 360s
- Learning
- Engagement surveys
- HR helpdesk
- People analytics
- Expense management
- Face or geo-fenced attendance
- API and webhooks
- Workflow automation
Seven · the rest of the stack
And the rest is not a tier at all. It is another company’s product.
The column on the right is the argument, and it is repetitive on purpose. Every line of it is an area of the same product, sharing one customer, one employee and one audit trail with every other line.
What that buys you is not a smaller bill — it is that the invoice knows about the ticket, the renewal knows about both, and nobody had to build an integration for either of those to be true.
| What you need | What it usually takes | Here |
|---|---|---|
| Pipeline, quotes and forecasting | A sales CRM | Sales CRM |
| Contracts signed by the customer | An e-signature product | Sales CRM |
| Sales commissions and statements | A commissions product | Sales CRM |
| Invoices, subscriptions and dunning | An accounting product | Finance |
| Health scores and renewals | A customer-success platform | Customer success |
| Tickets, SLAs and escalation | A helpdesk, per agent | Support |
| Answers drawn from your own material | A tier inside that helpdesk | Knowledge base |
| Calls recorded, quoted and coached | A conversation-intelligence tool | Call intelligence |
| SOWs, timesheets and live margin | A professional-services tool | Services delivery |
| Applicant tracking and interviews | An applicant tracking system | Hiring |
| Payroll, statutory filings and exits | An HR and payroll system | People & payroll |
| Visit plans, check-ins and expenses | A field-force app | Field visits |
| Nine kinds of product, nine logins, nine data models, nine renewal dates. | One. | |
Where the seams show
The integration is not the hard part. The disagreement is.
Every product on that list has an integration menu, and most will sync to most of the others. That was never the problem.
The problem is that two systems holding the same customer eventually disagree about who that customer is, and neither is wrong from where it is standing. Below are the three ways that shows up. None of them is exotic and every operations lead reading this has seen at least two.
The merge that only half happened
Duplicates are merged in the system that owns the customer. Every other system keeps both, because a merge is not an event any sync knows how to replay.
A renewal is forecast twice, against two halves of one account, and the quarter is over before anyone notices the number was wrong.
The customer who is three different people
Each system holds the identity its own job needs. None of them is incorrect; none of them is the customer.
Furious in the helpdesk, healthy in the CRM, thirty days overdue in a system neither team has a login to — and the account manager walks into the QBR knowing one of the three.
The field nobody owns
Two systems both write it, the sync runs both ways, and the last writer wins on a schedule nobody chose.
A corrected billing address reverts overnight, twice, and the third invoice goes to the old office again.
One record, not two that sync
The same company is one row whether it arrives as a lead, a payer or a ticket requester — enforced underneath rather than reconciled nightly.
A trail that outlives the deletion
The audit trail refuses its own deletion, which is a different promise from nine trails in nine products with nine retention settings.
Bring what you already have
Coming off a stack means arriving with its history. Import is a screen, not a services engagement.
What you actually get
Not a smaller bill. Five things a stack structurally cannot do.
The saving people expect from consolidation is money, and that is the least interesting part of it. These are the things that become possible only when the customer is one row rather than nine reconciled ones.
- A health score that can see the invoice —
Usage, tickets, survey responses and what is owed, in one number you can open — because the money and the complaints are on the same record as the relationship.
- A forecast that knows what was said —
The calls behind a deal sit on the deal, so the number is weighted by the objection nobody answered rather than by the rep's account of it.
- A candidate who becomes an employee —
Hired on Tuesday, paid in that month's run, with nothing re-entered — because the person was one record before the offer was signed.
- A timesheet that knows its invoice —
Billable hours carry the invoice line they became, so work that was delivered and never billed is surfaced rather than found at year end.
- One trail, not nine —
A single audit history across every area, which refuses its own deletion — as against nine trails with nine retention settings and no shared clock.
What is not here
The rows we would lose.
A comparison page that only lists wins is a brochure, and the first demo corrects it. These are the places where a specialist product is genuinely the better answer, and where an honest evaluation should mark us down.
- A marketplace of third-party apps —
Some of the products on that list have hundreds of integrations built by other companies. That is a real advantage of age and scale, and one record under twelve areas is a different answer to the same problem rather than a bigger version of theirs.
- Statutory depth outside one regime —
Payroll and statutory filing go deepest in one tax regime today. Hiring, portals, the CRM and the interview layer are locale-neutral; payroll is not, and a page that implied otherwise would be found out in week one.
- Years of installed base —
Several products on that list have been filing payroll for longer than this company has existed. That is a real thing to weigh, and it is not something a feature table can answer.
›Is this a like-for-like replacement for all nine products?
For most teams, in the areas listed, yes. The test that matters is whether the depth in each area is enough for how you work, which is what the area pages are for — and the page above is deliberately about product count rather than depth.
›Do we have to take all twelve areas?
No. Areas are entitled one at a time with no dependency between them, so taking two that have nothing to do with each other is a normal configuration rather than a special case.
›Why does this page not name the products it is comparing against?
Because named grids go stale quietly. Packaging moves, tiers get renamed, and a table that was true in September is a factual claim about somebody else by March. The count above is about categories of product, which does not move.
›What happens to the data in the systems we are leaving?
It comes with you. Import is a screen in the product rather than a professional-services engagement, and the history arrives attached to the records it belongs to.
Book a demo
Bring your actual stack to the call.
Thirty minutes, with your renewal dates in front of you. The useful outcome is an honest list of what moves, what stays, and what you would give up — not a demo that avoids the question.
- Here are the nine products we pay for — show me which areas replace which
- Show me the one you would not replace, and say why
- Open an invoice and a ticket for the same customer, on one record
- Import a file from the system we are leaving, live
- Show me a capability we would lose by moving