Weekly Editorial, W36
Salesforce put $166m into an HR platform and named Slack as the place employees will meet it, while three other vendors shipped their own answer to the same question in the same week. Not one of the four published a price, an adoption figure or an accuracy rate.
Theme 01
The unsolved problem in HR software is that employees do not open it. Four vendors moved on that problem in four days, and they did not agree on where the fix belongs.
On 1 September HiBob announced a $166m investment led by Salesforce. Later reporting put the valuation at $3.2bn, against $2.7bn at its 2023 round, on annualised revenue past $400m.
The sum is not the interesting part. The surface is. The announcement's product substance is Slack, and the phrase that matters is what the integration carries: "people, data, roles, skills, teams, permissions, and structure."
That is not a chat integration. That is the organisational graph, which is the asset every agent strategy in this market is competing to own. Routing it through Slack solves the engagement problem without HiBob having to win the desktop itself.
An HR platform's weakest point has always been that employees visit it a handful of times a year and forget the password in between. Every move this week is an answer to that, and the four answers are incompatible.
Rippling gave the opposite answer. It shipped Helpdesk on 1 September, an IT support agent resolving app access, password resets and device replacements. That is its fourth distinct agent surface in four weeks, a cadence no other vendor covered here matched in 2026.
Note where that lands. Helpdesk is IT service management, not HR and not payroll. Rippling is arguing that owning the employee record entitles it to run the systems downstream of that record, and it is spending the argument against a different budget line.
Workday's answer was to put agents inside the most conservative buyer that exists. The City of Akron, 2,300 employees, went live running a Self-Service Agent, a Payroll Agent and a Financial Audit Agent.
Public payroll is unionised, statutorily rigid, disclosable under freedom-of-information law and politically punished for errors. An agent executing there is a materially stronger artefact than a vendor demonstration, and it is the first of its kind recorded here.
The fourth answer removes the HR function altogether. Leo HR launched a UK platform aimed explicitly at employers with no HR team, covering grievances, falling performance and absence.
Almost everything agentic shipped into this market in 2026 automates administration. This automates judgement, in the least reversible category of HR work, and sells it to the buyer least equipped to tell good guidance from bad.
No price. No adoption figure. No accuracy or deflection rate. Not one of the four attached a number a buyer could evaluate, and a fifth vendor made the same silence audible in its accounts.
SD Worx disclosed €30.2m of digital investment in its first half, described a scaled multi-agent customer service deployment, and attached no adoption, deflection, accuracy or cost-saved figure to any of it.
A results release is a harder place to stay silent than a press release. A company that can state what a programme cost to the hundred thousand, and cannot state what it produced, has disclosed which of the two it measures.
Rippling's omission is the sharper one. Vendors have justified missing payroll-agent accuracy figures by arguing correctness is genuinely hard to measure. A helpdesk has no such defence: tickets resolved without escalation is a number every ticketing system already emits.
For a buyer, that is the question to carry into renewal. Not whether a vendor has agents, but which of them will put a number next to one. On this week's evidence, none will until a competitor forces it.
Theme 02
Almost every platform here is priced per employee per month. Four separate measurements landed this week, and they agree.
Gusto reported small businesses adding 29,500 jobs in August, the seventh straight month of growth and, on its own account, the weakest of the streak.
ADP put private hiring at 38,000, and the composition mattered more than the total. Employers of 1 to 19 added 20,000, those above 500 added 34,000, and the 20 to 49 band lost 17,000.
That band sits at the centre of the mid-market most vendors price and sell against. A flat aggregate was concealing a shrinking middle, which is a sharper finding than the headline it hid inside.
The first three are vendors measuring employment inside their own customer books, and the same objection answers each of them: a book is not a market. The fourth is the official national projection of the base itself.
Against the Bureau of Labor Statistics projections for 2025 to 2035: the US labour market grows 3.5% over ten years while GDP grows 22%. Workforce growth is roughly a third slower than the decade before it.
If the employed population is flat for a decade while output rises, every per-employee-per-month line in this market is attached to a base that does not grow. Vendors priced per employee do not participate in the 22%.
The nearer consequence is not repricing. It is that a buyer with flat headcount and rising output is a buyer whose per-seat bill should also be flat, which is exactly when contractual indexation stops being invisible.
SD Worx named price indexation among its organic growth drivers this week, with payroll software growing 2.7% organically against a staffing arm at 7.3%. Indexation arrives as a mechanic rather than a negotiation.
One caution against reading a single conclusion here. A flat base is equally consistent with charging more per unit as with expanding sideways, and at least one employment marketplace has already answered a shrinking unit by raising revenue per unit 35% while postings fell 4%.
Theme 03
A payroll engine is only as good as its instructions. Those instructions degraded three ways in three jurisdictions, and not one of them was a rate change.
On 3 September the Fair Work Agency named 658 employers for minimum-wage failures: £4m repaid to more than 27,000 workers, against £7m in penalties.
Read the per-employer figures rather than the totals. B&Q appears for £456,934 across 4,530 workers. Five Guys for an average of £14.77 a head. Tesco for £227.84, owed to one worker.
A £227.84 error affecting one employee now earns the same public listing as £456,934 across four and a half thousand. Exposure has stopped scaling with the size of the mistake.
Minimum-wage breaches are overwhelmingly not calculation errors. They are configuration and process, and almost all of them are arithmetically detectable before a pay run commits.
The recurring causes are known: unpaid pre-shift briefings and travel between sites, deductions for uniform and tools dragging pay under the floor, salary sacrifice pushing an employee below it, apprentice rates that never step up on the anniversary.
Each is a comparison between net-of-deductions pay and the applicable hourly floor, per employee, per reference period. It is cheap arithmetic. Very little payroll software runs it as a pre-commit check, which is why a £227.84 exception reaches a government list instead of a validation screen.
The second failure is what the release omits. The comparable 2021 round published the reasons: 47% wrongful deductions, 30% unpaid working time, 19% wrong apprentice rate, alongside a bulletin telling employers what to fix.
None of that appears in the September 2026 release. That breakdown was effectively a public specification, naming which interaction between a deduction type and the statutory floor to harden first. Enforcement widened its aperture and narrowed its feedback in one document.
The third failure is a vendor's. BrightPay, which pays roughly 674,000 Irish employees a month, published on 13 August that employer PRSI rises to 11.30%. It rises to 11.40%.
The correction appeared on 2 September as a new post, leaving the August page live and unmarked. Guidance is not the calculation engine, and nothing suggests the software computes PRSI wrongly. But the wrong figure read as current for twenty days.
Ireland supplied the fourth data point, and it points somewhere more interesting. After employers routed staff into schemes paying employer contributions as low as 1%, the government required workplace schemes to match the state scheme's levels.
A parity floor is a different computational object from a rate. A rate is a number applied to a pay element. Parity is a comparison between two schemes, evaluated per employee on every run, and re-evaluated at each step-up to 2035.
It also arrived as a response to employer arbitrage rather than through the legislative pipeline. Systems built to implement rules as drafted will keep meeting rules written in reaction to what employers actually did.
Theme 04
Three European vendors were running three visibly different expansion strategies. One switched this week, and what it bought says more than that it bought.
On 2 September Factorial acquired Empion, a Berlin AI talent-assessment platform whose customers include Deutsche Telekom, Procter & Gamble and Siemens Mobility. Value undisclosed.
Six weeks earlier the same company entered southern Africa through a network of regional partners. That was one of three models on view: partner-led implementation, buying the domestic payroll filer as Visma does, or building the statutory engine country by country as PayFit does.
Factorial has now used the second, in the market its June round named as the priority, three months after raising $150m at a $2.5bn valuation. Partner-led expansion looks less like a philosophy than like what a company does before it can afford to buy.
Talent assessment carries no statutory obligation in any jurisdiction. This advances a vendor's position in the German HR software market and not one step toward a German payroll engine.
Set it beside what the same money buys elsewhere. A French or Italian payroll practice is statutory capability, expensive to build and slow to rent. Empion is distribution and credibility, and the enterprise logos are the point of the announcement.
That is a real position, worth stating precisely rather than inflating. This is a company buying its way up the German HR software market while its payroll route stays rented, and in the UK that route runs through another vendor's payroll product.
A company that owns the assessment layer and rents the statutory layer is building a system of engagement on somebody else's system of record. The test that settles it is whether the next German acquisition touches payroll rather than talent.
Next