TL;DR
- What it is: a system that turns scattered HR requests into tracked cases, so nothing is answered twice and nothing is lost in a shared inbox.
- When you don't need it: under about forty employees with one HR person who still recognises every name. A shared inbox and a good handbook genuinely cover it.
- The split nobody explains: one half of this market answers questions, the other half tracks cases. Buying the wrong half is the expensive mistake.
- What changed in 2026: AI agents trained on your own policy documents now deflect the repeat questions before a ticket exists, and some of them cost less than a single help desk seat.
- The decision rule: count how many of last month's requests had an answer already written down. If most did, you have a deflection problem, not a ticketing problem.
- The outcome: HR spends its week on the conversations that need a person, not on explaining the leave policy for the ninth time.
The Inbox That Became a System Nobody Designed
An HR team of three supports four hundred people. Requests arrive by email, by Slack direct message, in corridors, and increasingly in replies to unrelated threads. There is a shared inbox. There is also a spreadsheet somebody started in March to track the ones that matter.
Nobody can answer two questions. How many requests came in last month? And how long did they take?
That is the point at which HR help desk software stops being a nice idea. Not headcount, not budget cycle, and not a directive from IT. It is the moment somebody asks a question about the function's own workload and the honest answer is that nobody knows.
Best tools for Device Management
The second question matters more than the first, and it is the one this guide is organised around: of those requests, how many had an answer that already existed in writing? Because that single ratio decides which half of this market you should be shopping in, and most teams never calculate it before they start taking demos.
When You Don't Need HR Help Desk Software
When the team is small enough to hold in one head
Under about forty employees with one HR person, a shared inbox works. The person reading it recognises every name, remembers what was asked last week, and can hold the backlog without a tool. Adding software here creates process for its own sake, and the process will be abandoned within a quarter.
The honest signal that you are still in this territory: nobody has ever asked you for a report.
There is a second signal worth watching, because it moves before volume does. If one person has started keeping a private list to avoid dropping things, the inbox has already stopped working as a system and is working only as that person's memory. That is a fragile arrangement with a single point of failure, and it is usually the point at which a tool starts earning its cost even though headcount has not changed.
When the handbook is the actual problem
If staff cannot find answers because the handbook is two years out of date, buying software will not fix that. It will faithfully route the same questions to the same person, who will give the same answer that contradicts the document. Fix the content first. It is unglamorous, it is free, and it removes a surprising share of the volume.
When friction starts showing
Three signals, in the order they usually appear. Somebody answers the same question for the fourth time in a month and says so out loud. A request goes missing and is only noticed when the person chases it. And a manager asks how long something normally takes, and the answer is a guess.
The edge case that forces it
An audit, or a near miss. The moment somebody outside HR needs evidence of how a case was handled, when it was raised and who saw it, an inbox stops being defensible. That is not a volume problem, it is a record-keeping one, and it arrives without warning.
The variant that catches growing companies is a change in what the business sells rather than how many people it employs. Taking on a client with security or employment schedules in the contract, or entering a regulated market, creates an evidence requirement that did not exist the previous quarter. That requirement does not care what your request volume is, and it frequently settles this decision on its own.
When somebody leaves and takes the context with them
Worth naming separately because it is the least discussed trigger. In a small function, how requests get handled is often held entirely in one person's head: who to route what to, which manager needs warning, what was agreed last time. When that person gives notice, the gap is not the headcount, it is the undocumented process. A case system does not fix that by itself, but it is the only thing on this page that leaves a record somebody else can read.
Five Questions HR Leads Ask at 11pm
"How many of these questions are actually new?" The question that decides everything else, and almost nobody counts. Take last month's requests and mark each one as either answered-in-writing-already or genuinely new. The ratio tells you whether to buy deflection or ticketing.
"Who is going to configure this?" HR help desk tools assume somebody will build request categories, routing rules and templates. If that person does not exist, buy the thing that works with the least configuration, not the thing with the best feature list.
"Can we keep sensitive cases out of it?" Grievances, investigations and anything involving a named third party should not sit in the same queue as a question about parking. Ask every vendor how case-level access restriction works, specifically, and test it.
"Will people actually use a portal?" Usually not, if the alternative is messaging a person they know. The tools that succeed meet staff where they already are, which in most companies means Slack or Teams rather than a web form.
"What happens to the inbox?" The honest answer is that it survives for months. Plan for a transition period where both channels are live, and decide who is responsible for pulling email into the new system, or it quietly becomes two systems.
The pattern across those five is worth naming. Four of them are answered from information you already hold and have never written down, and only one needs a vendor. That ratio is normal here, and it is why the research phase feels unproductive: the questions that decide the outcome are internal, while the ones vendors are eager to answer are not the ones holding you up.
The Two Halves of This Market
This is the distinction that product pages obscure, and getting it wrong is the most expensive mistake available.
Answering: deflection tools
An AI agent or knowledge system pointed at your handbook, policies and intranet pages, which answers employee questions directly from that content. No ticket is created because no ticket is needed.
Right when: most of last month's requests had an answer already written down.
Fails when: the problem is that cases get lost, because deflection tools do not track anything they successfully answer.
Tracking: case management
A help desk proper. Requests become tickets, tickets have owners, states, due dates and an audit trail, and you can report on volume and resolution time.
Right when: requests disappear, nobody owns them, and you cannot answer a question about turnaround.
Fails when: you buy it to stop repeat questions. It will record them beautifully, and a person will still answer every one.
The arrangement most teams end up with
Deflection in front, ticketing behind. The AI handles the repeat questions, anything it cannot answer becomes a tracked case with the conversation attached, and the ticket volume you pay to manage is the volume that genuinely needed a person.
| Half of the market | What it does | Right when | Fails when |
|---|---|---|---|
| Deflection | Answers from your own documents | Most requests are already answered in writing | Cases need tracking and reporting |
| Case management | Tracks, assigns and reports | Requests get lost and nobody owns them | You wanted the repeat questions to stop |
| Both, layered | Deflect first, track the remainder | You have volume and an audit obligation | Nobody owns the handoff between them |
Calculating the Deflection Ratio, Concretely
This gets recommended on every page like this one and explained on almost none, which is unhelpful because it is the number that decides which half of the market you are shopping in. It takes an afternoon and it needs no software.
Take one month of requests, wherever they arrived. Shared inbox, Slack direct messages, anything a colleague forwarded. Thirty to sixty items is plenty. Do not sample a quiet month and do not exclude the ones that annoyed you.
Mark each one with a single letter. W if the answer was already written down somewhere a member of staff could have found it. N if it was genuinely new, specific to that person, or needed a judgement call. If you hesitate, mark it N, because an answer buried in a policy appendix nobody links to is not realistically findable.
Count a third category, R, for repeats. The same question from a different person. These are W by definition, but counting them separately tells you which five documents are carrying your whole volume, and that is the list your content review should start from.
Then read the ratio. Above roughly sixty per cent W, you have a deflection problem and a case management system will document it rather than fix it. Below about thirty per cent W, deflection will answer very little and the real work is tracking. In between, you will probably end up running both, and the useful question becomes which half to buy first.
| W as a share of volume | What it means | Buy first |
|---|---|---|
| Above 60 per cent | Staff cannot find answers that exist | Deflection, and fix the handbook |
| 30 to 60 per cent | Mixed, both problems are real | Deflection first, it is cheaper and faster |
| Below 30 per cent | Requests genuinely need a person | Case management |
| Any ratio, requests going missing | Volume is not the problem, ownership is | Case management regardless |
One more count while you are in the data. How many requests arrived somewhere a tool would never see, such as a corridor conversation or a direct message to a colleague outside HR. That number is your reporting blind spot, and knowing it before you buy stops you trusting a dashboard that reports forty cases in a month generating ninety.
The output is four numbers on one page: W, N, R and the invisible ones. That page is worth more in a vendor conversation than any feature comparison, because it describes your organisation rather than somebody's product.
How to Choose: Five Questions Before You Talk to Any Vendor
Calculate the deflection ratio. Last month's requests, split into already-answered-in-writing versus genuinely new. One afternoon of work, and it rules out half the market.
Name the administrator and ask what they will do in month one. A platform whose value depends on configuration needs somebody to configure it. If the answer is vague, choose the option that assumes less.
Get the compliance requirement in writing from whoever owns it. Not from HR. If somebody needs case records retained, restricted and retrievable, that requirement eliminates options and should do so before a demo, not after a pilot.
Establish what you already pay for. If the company runs an IT service desk, it very likely has an HR module or an employee-service tier sitting unused. Check before buying a second system.
Decide where staff will ask. If the answer is Slack or Teams, a tool whose primary surface is a web portal will be used by the HR team and nobody else.
Seven Options Worth Knowing
Pricing below is what each vendor published on its own site, read on 5 October 2026. Where a vendor publishes no rate card, this says so rather than estimating.
Matram
Disclosure: Matram is our own product. It is assessed here on the same criteria as everything else on this page, its limitations are stated in full, and no vendor on this page has paid for placement.
Best for: deflection, when the answers already exist in your documents.
Why teams choose it: it reads your handbook, policy pages and intranet and answers from them, citing the source so an employee can check the answer against the actual document. Pricing is flat at $29, $69 or $199 a month with unlimited seats, so the cost does not move as HR or the company grows. It answers in 95 or more languages from one set of source content, which removes the usual cost of maintaining a handbook per market.
Where it struggles: there is no free tier, only a 30 day trial that does not ask for a card. More importantly for this category, it answers questions but does not action tickets, and it carries no HR-specific workflows for leave, payroll or onboarding. If your problem is case tracking, this is the wrong half of the market.
Leena AI
Best for: enterprise HR service delivery where the workflow behind the question also needs automating.
Why teams choose it: purpose-built for employee service rather than adapted from customer support, with HRIS integrations that let it complete a request rather than only explain it. For an organisation with an established HR operations function, that depth is the argument.
Where it struggles: it publishes no pricing, so budgeting starts with a sales conversation. Deployment is a project with a security review attached, not an afternoon, and it is more platform than a mid-sized team needs.
Moveworks
Best for: large organisations that want one assistant across HR, IT and facilities.
Why teams choose it: employees do not know or care which department owns their question, and Moveworks is built around that, which avoids the awkward end state of three separate chatbots in the same Slack workspace.
Where it struggles: no published pricing at all, and its pricing page returns an error. Enterprise deployment timeline. Below a few hundred staff the evaluation alone costs more time than a simpler tool costs to run for a year.
Freshservice
Best for: HR sitting inside, or alongside, an existing IT service desk.
Why teams choose it: genuine case management with queues, SLAs and reporting, plus asset records in the same platform, and a published rate card you can budget against without a call.
Where it struggles: the AI layer is a separate charge. Freddy AI Copilot is an additional $29 per agent per month on top of the licence, so a team wanting deflection pays twice.
Pricing: $19 per agent per month on Starter, $49 on Growth and $99 on Pro, all billed annually, with Enterprise quote-based.
Zendesk
Best for: larger HR functions that need real ticketing and real reporting.
Why teams choose it: mature case management, queues, SLAs and resolution-time reporting that deflection-only tools cannot provide.
Where it struggles: per agent pricing compounds, and the AI assistance is separate again. It is also built for customer support and configured into an HR shape, which shows in the vocabulary.
Pricing: Suite Team is $55 per agent per month billed yearly, or $69 monthly, with Copilot AI a further $50 per agent per month. Pricing is shown by region, so confirm the currency you will be billed in.
Crisp
Best for: small teams who need somewhere to start with no approved budget.
Why teams choose it: a genuinely usable free plan rather than a countdown, and pricing per workspace rather than per agent, so adding HR colleagues does not raise the bill.
Where it struggles: AI answers are metered through credits on top of the plan, so a busy month costs more than the sticker suggests. No HR-specific features at all, and answers do not cite their source.
Pricing: Free, then $45 a month for Mini, $95 for Essentials and $295 for Plus, per workspace, with $5, $25 and $75 of included AI credits respectively.
SiteGPT
Best for: one focused bot over an internal handbook and nothing beyond that.
Why teams choose it: it does one job, trains on your pages and answers from them, with no inbox or ticketing layer to configure or pay for.
Where it struggles: page limits on the entry plan, and the headline price is the annual one. No HR workflows, no case tracking.
Pricing: Starter is $468 a year, which the site presents as $39 a month, or $59 billed monthly, and covers one chatbot and up to 1,000 pages. Growth is $948 a year, shown as $79 a month.
What Each One Published
| Option | Published price | Unit | Half of the market |
|---|---|---|---|
| Crisp | Free, then $45 to $295 | per workspace, per month | Deflection, metered AI |
| Matram | $29, $69 or $199 | per month, unlimited seats | Deflection |
| SiteGPT | $468 or $948 | per year | Deflection |
| Freshservice | $19, $49 or $99 | per agent, per month | Case management, AI extra |
| Zendesk | $55, plus $50 for Copilot | per agent, per month | Case management, AI extra |
| Leena AI | Not published | n/a | Both, enterprise |
| Moveworks | Not published | n/a | Both, enterprise |
Read the last column before the second. Comparing a deflection tool's price against a case management licence is comparing two different purchases, and it is the most common error in this category. The per agent options also behave differently as you grow: at three HR seats the gap is small, and at twelve it is the whole decision.
The Decision Table
| Situation | Scale | Setup | Primary Pain | Recommended Starting Point |
|---|---|---|---|---|
| One HR person, no reports requested | Under 40 | Any | Nothing is broken yet | Shared inbox and a current handbook |
| Most requests already answered in writing | 40 to 500 | Any | The same question, repeatedly | A deflection tool, Matram or SiteGPT |
| No budget approved, want to test the idea | Under 100 | Any | Proving it before spending | Crisp free tier |
| Requests get lost, no owner, no reporting | 100 plus | Any | Cases, not questions | Freshservice or Zendesk |
| HR already inside an IT service desk | Any | Existing ITSM | Paying twice for the same thing | The HR module you already own |
| Audit needs case records on demand | Any | Any | Evidence, repeatedly | Case management, budget for it now |
| Enterprise, several departments, one assistant wanted | 1,000 plus | Any | Three chatbots in one Slack | Leena AI or Moveworks |
| Laptops and equipment, not HR questions | Any | Remote | Devices, not cases | Not this category at all |
Most teams land in rows two and four, and the deflection ratio tells you which. The fifth row is the cheapest outcome available and the one least often checked.
Where Teams Get This Wrong
Buying ticketing to stop repeat questions. The single most expensive error here. A help desk will record the same question forty times with perfect fidelity, and a person will still answer each one. If the volume is repeat questions, you need deflection, and the ticketing system you bought will simply document the problem.
Not checking what IT already owns. If the company runs a service desk, an employee-service or HR module very often sits inside the existing licence. Buying a second system because it was procured by a different team is a real recurring cost paid for nothing.
Putting sensitive cases in the general queue. Grievances and investigations need case-level restriction, and discovering that after go-live is a trust problem rather than a configuration one. Ask about it specifically and test it during the pilot.
Choosing a portal nobody will use. Staff ask where it is easy to ask. If the tool's main surface is a web form and the cultural default is Slack, the portal will be used by HR and nobody else, and the inbox will continue as before.
Deploying on stale content. Every deflection tool is only as accurate as the documents behind it. A bot pointed at a two-year-old handbook will confidently repeat two-year-old policy, and staff will act on it. Audit the content before deployment, not after the first complaint.
Underestimating the transition. The shared inbox does not close itself. Decide who moves email into the new system and for how long both run, or you will operate two queues indefinitely.
Comparing prices across the two halves. A flat deflection price and a per agent case management licence are not alternatives to each other. Work out which half you are buying first, then compare inside it.
What HR Help Desk Software Will Not Tell You
It knows about requests that reached it. That sounds complete until you need the other number.
It cannot tell you about the questions asked in corridors, in direct messages to a colleague, or in a reply to an unrelated thread. Those are the requests that never enter the system, and in most companies they are a substantial share of the real volume. A tool that reports forty cases a month in an organisation generating ninety is not wrong, it is partial, and the gap is invisible precisely because it is outside the tool.
It also cannot tell you whether the answer given was correct. Deflection tools report what they answered and how often, not whether the policy they quoted was current. That is a content problem wearing a reporting problem's clothes, and the only fix is somebody owning the handbook.
The diagnostic is one question. Ask what went wrong last quarter. If people could not find an answer that existed, this category solves it. If a case was mishandled and nobody can reconstruct what happened, this category solves that too, with the other half of the market. If the real issue is that nobody knows how much work HR is actually absorbing, start by counting for one month before buying anything, because software will only count what you route through it.
What to Establish Before You Commit
The deflection ratio. Last month's requests, split into already-answered versus new. The single most decisive number and the one least often calculated.
The named administrator, and what they will do in month one. Not a role, a person, with a specific first task.
The compliance requirement, in writing, from its owner. Retention, restriction, retrievability. One sentence from whoever carries the obligation.
What IT already owns. Specifically whether the existing service desk includes an HR or employee-service tier.
Where staff will actually ask. Observed, not assumed. Look at where last month's requests arrived.
Your handbook's last review date. If it is older than a year, that is the first project regardless of which tool you choose.
How sensitive cases will be separated. Decided before go-live, tested during the pilot.
None of those seven needs a vendor conversation. They are answerable inside your own organisation in about a day, and together they eliminate most of this page before anybody books a demo.
What Getting This Wrong Costs
The first cost is the licence paid against the wrong half of the market. A case management system bought to stop repeat questions renews annually while the repeat questions continue, and the tool is blameless because it was never built to do that.
The second is the abandoned configuration, and it is the most common outcome in small HR functions. Somebody sets the system up in a burst of effort, nobody maintains the request categories, and eighteen months later it holds the cases raised in the first month and none since. The subscription is live and the coverage is partial, which is worse than either not having it or doing it properly, because it produces a confident answer to a compliance question that is not true.
The third costs trust rather than money, and it is specific to HR. A grievance that sits in a general queue, visible to the wrong colleague, is not a configuration error to the person who raised it. That damage is slow to repair and entirely preventable by asking about case-level restriction before signing.
The fourth is the one that only appears at an audit. A system that captures forty of ninety monthly requests produces records that look complete and are not, and the gap is discovered by somebody who is not on your side. Count your real volume before you trust any tool's reporting.
When You Are Ready to Decide
Calculate the deflection ratio first. One afternoon with last month's requests, split into already-answered-in-writing and genuinely new. If most were already answered, you are shopping for deflection and half of this page does not apply to you. If most were new, or were lost, you need case management and the other half does not apply.
Then check what you already own, because an HR module inside an existing service desk is the cheapest answer available and the one nobody looks for.
Then pilot on real requests rather than test ones. Point a deflection tool at your actual handbook and ask it the five questions HR answered most often last month. Read the answers against the policy documents, not against your memory of them. That exercise finds stale content as reliably as it evaluates software, and the stale content is usually the bigger problem.
Two more things to test while the pilot is live, both easier to check now than to discover later. Ask it something deliberately outside your documentation and watch what it does: a clean handoff with the conversation attached is the behaviour you want, and a confident guess is disqualifying, because an employee told the wrong notice period will act on it. Then ask a colleague without permission to search for a sensitive case and confirm they see nothing. Both tests take ten minutes and both are the kind of thing that is uncomfortable to find out about in month four.
And give the pilot to whoever will own it rather than whoever is most curious. The useful signal is not whether the tool is impressive, it is whether the person who will live with it can get an answer corrected without asking anybody for help.
And whatever you choose, put the handbook review on somebody's calendar. Every tool on this page answers from documents a person has to keep current, and no amount of licence spend substitutes for that.
A closing note on sequencing, because it saves money. Deflection is cheaper, faster to deploy and reversible, so it is the better first purchase in almost every mixed case. Case management is the larger commitment and the harder thing to unwind once request categories and routing rules are built around it. If the deflection ratio puts you in the middle band, buy the cheap half first, run it for a quarter, and let the remaining volume tell you whether the second half is actually needed.
For a side-by-side comparison of the AI options specifically, including what each one costs and where each one loses, see our [HR chatbot software comparison](/best-hr-chatbot/).
Frequently Asked Questions
What is HR help desk software?
HR help desk software turns scattered employee requests into tracked cases with an owner, a state and a record, so nothing is answered twice and nothing is lost in a shared inbox. The category actually contains two different products sold under one name. One half deflects questions by answering them from your existing handbook and policy pages, so no case is ever created. The other half manages cases properly, with queues, assignment, due dates and reporting on how long things took. Most teams need one of those far more than the other, and working out which is the first job.
Do we need HR help desk software at under 100 employees?
Often not, and the honest test is whether anybody has asked you for a report. Under about forty employees with one HR person, a shared inbox and a current handbook genuinely cover it, because the person reading the inbox recognises every name and can hold the backlog without a tool. Between forty and a hundred it depends entirely on your deflection ratio: if most requests already have an answer written down somewhere, a deflection tool pays for itself quickly, and if requests are getting lost instead, you need case management regardless of headcount.
How much does HR help desk software cost?
Published prices range from nothing to $99 per agent per month, and the spread is mostly about which half of the market you are buying. Verified on 5 October 2026: Crisp starts free and then $45 a month per workspace, Matram is $29 a month flat with unlimited seats, SiteGPT is $468 a year, Freshservice is $19 to $99 per agent per month billed annually with its Freddy AI add-on a further $29 per agent, and Zendesk is $55 per agent per month with Copilot a further $50. Leena AI and Moveworks publish no rate card at all, so budgeting for either begins with a sales conversation.
Can an HR help desk answer questions about our own policies?
Yes, and for the deflection half of the market that is the main thing it does. You point the tool at your handbook, policy documents and internal pages, and it answers from that content rather than from general knowledge. The quality of the answers is therefore the quality of your documents, which catches teams out badly: a tool deployed against a handbook last reviewed two years ago will confidently repeat two-year-old policy, and an employee told the wrong notice period will act on it. Audit the source content before you deploy, and ask whether answers cite their source so staff can check.
Should HR use the same help desk as IT?
Often yes, and it is the cheapest option most teams never check. If the company already runs a service desk, an employee-service or HR module frequently sits inside the existing licence, which means the answer costs nothing extra. The argument against is vocabulary and access: a system configured for IT incidents uses language that reads oddly for HR requests, and case-level restriction becomes essential because grievances cannot sit in a queue IT colleagues can browse. Ask about restriction specifically before deciding, and test it during the pilot rather than taking the answer on trust.
How do we keep grievances out of the general queue?
By making case-level access restriction a requirement before you shortlist, not a configuration task afterwards. Grievances, investigations and anything naming a third party need to be visible only to named people, and the mechanism for that differs considerably between tools. Ask each vendor to demonstrate it on real-looking data during the pilot, and specifically test what a colleague without permission sees when they search. Discovering the limits of this after go-live is not a technical problem but a trust one, and it is slow to repair with the person who raised the case.
Will staff actually use a help desk portal?
Usually not, if the alternative is messaging somebody they already know. People ask where it is easy to ask, which in most companies now means Slack or Teams rather than a web form. The tools that get adopted meet staff in that surface and create the case invisibly behind it, rather than asking anybody to learn a new destination. Before choosing, look at where last month's requests actually arrived rather than where you would prefer them to arrive, and treat that observation as a requirement rather than a habit to be corrected.
What is the difference between deflection and ticketing?
Deflection answers the question so no case is created, and ticketing records the case so it can be tracked, assigned and reported on. They solve opposite problems and are frequently sold as the same thing. If your volume is mostly questions whose answers already exist in writing, deflection removes the work and ticketing merely documents it. If requests are getting lost and nobody can say how long anything takes, ticketing fixes that and deflection will not touch it. Many teams eventually run both, with deflection in front and the remainder becoming tracked cases, which is a sound arrangement provided somebody owns the handoff between them.