TL;DR
- Most equipment SLAs measure the wrong event. Days to dispatch is the vendor's convenience; days to the person having a working machine is what you are buying.
- Retrieval usually has no commitment at all, which is remarkable given it is the half of the service that actually differs between vendors.
- A single global target is either unachievable somewhere or trivial everywhere. Set targets per market tier, using your own markets rather than theirs.
- An SLA with no remedy is a statement of intent. An SLA with a remedy nobody ever claims is only slightly better.
- Measure the exceptions, not the average. A service that is excellent ninety per cent of the time and unaccountable the other ten is the one that generates all your work.
- Ask for reporting you do not have to request. Anything you have to ask for will stop being asked for by month four.
The SLA that was met every month while the service got worse
A 400-person company had a three-day delivery SLA with its equipment vendor and monthly reports showing compliance in the high nineties. The arrangement was, on paper, working well.
What the reports measured was dispatch. Three days from order to the device leaving the vendor's facility. What the company cared about, and what everybody internally assumed the number meant, was the new joiner having a working machine.
Those two numbers had been drifting apart for about eight months. Dispatch stayed at three days. Delivery into the company's two slower markets had gone from six days to eleven. Customs handling, which was outside the measured window, accounted for most of it. The SLA was met every single month while the experience got steadily worse, and nobody could point to a breach because there had not been one.
Best tools for HR Operations
When it was finally raised, the vendor's response was entirely reasonable: the SLA says dispatch, it has always said dispatch, and they had met it. They were right. The failure was in what had been agreed, not in what had been delivered.
When an SLA is not the right instrument
The commitment is small and you can leave. A month-to-month arrangement with no minimum gives you a better remedy than any service credit: stopping. Negotiating detailed SLAs into a reversible arrangement is effort spent on the wrong lever.
You have no baseline. An SLA set without knowing your current performance is a guess, and a guess written into a contract is worse than no number because it becomes the standard. Measure first.
The vendor is small and the relationship is the mechanism. With a small supplier, a conversation works better than a clause and both sides know it. Write the measurement in; be realistic about whether a penalty regime is the thing that will change behaviour.
And when you will never enforce it. Companies routinely negotiate service credits they then never claim, because claiming is work and the amounts are small. If that is you, be honest and spend the negotiation on measurement and reporting instead, which you will actually use.
Measure the right event
This is the single highest-value change available and it costs nothing to ask for.
Order to hand, not order to dispatch. The clock starts when you place the request and stops when the person has a device they can work on. Everything in between is the service, including customs, including the last mile, including a failed first delivery attempt.
Vendors resist this and the resistance is not unreasonable. Customs genuinely is partly outside their control, and a courier failing to find an address is not their doing. The honest resolution is not to abandon the measure but to agree exclusions explicitly: a documented customs hold stops the clock, a failed delivery where the recipient was unreachable stops the clock, and both are reported rather than silently excluded.
What you must not accept is a measure that ends at dispatch with no commitment beyond it, because that is a promise about the easy part of a journey whose difficulty is concentrated elsewhere.
And define working machine. A device that arrives requiring ninety minutes of setup before the person can do anything is not the same as one that arrives ready. If the vendor configures and enrols, the SLA should end at usable rather than at delivered. Where your own team does the final configuration, split the measure explicitly: the vendor's portion ends at delivery and yours begins there, with both halves reported. That is the only way to see which side of the handover is responsible when a new joiner waits, and in practice the answer is frequently the internal half, which is uncomfortable and worth knowing.
Set targets per market, from your own list
A single global number is the second most common SLA mistake. It is either generous enough to be met everywhere, in which case it commits nobody to anything, or tight enough to be realistic in your main markets, in which case it is breached constantly in the hard ones and the breach stops meaning anything.
Tier your markets yourself rather than accepting the vendor's tiers, because the vendor's tiers reflect their operation and yours should reflect your headcount.
| Tier | Typical markets | What to ask for | Why |
|---|---|---|---|
| Tier one | Where most of your people are | A tight target, measured to hand | High volume, vendor has depth, no excuse |
| Tier two | Secondary markets with some volume | A realistic target plus a reporting commitment | Performance varies, you need visibility |
| Tier three | Single-person and difficult markets | A committed quote per request, not a standing target | A standing target here is fiction |
| Exceptions | Anywhere new | A response time for a quote, not a delivery time | You are asking them to do something unusual |
The tier three approach is worth dwelling on because it is the one most people miss. In a market where a vendor realistically cannot promise a standing lead time, asking for one produces either a padded number that tells you nothing or a commitment they will breach. What you actually want is a commitment to respond with a specific, binding timeline within a short window of your request. That is achievable, it is useful, and it gives you something to plan the person's first week around.
Retrieval needs a commitment, and usually has none
Delivery gets an SLA because it is visible and because vendors are comfortable committing to it. Retrieval frequently gets nothing, which is strange given it is the half of the service where vendors genuinely differ.
The difficulty is real: retrieval depends on a person who no longer works for you cooperating, and no vendor can commit to somebody else's behaviour. That is a reason to structure the commitment carefully, not a reason to have none.
Commit the vendor to their own actions rather than to the outcome. Days from your notification to the first contact attempt with the leaver. Days to a collection being booked once the person responds. Number of contact attempts before escalation, and what escalation means. All of those are within the vendor's control and all of them are where retrieval actually fails.
Define the stopping point. After how many failed attempts does the vendor tell you it is not working, and in what form. The common failure is a retrieval that sits open indefinitely with nobody telling you it has stalled, which means you discover it during an audit rather than in week three.
Ask for a completion rate commitment with a stated denominator. Of collections where the leaver responded, what proportion complete within a stated window. Excluding non-responders is fair to the vendor and keeps the measure meaningful.
And agree what happens to the device if it is never recovered. Who writes it off, at what point, and what evidence you receive. This is not strictly an SLA matter and it belongs in the same conversation because nobody has it until it is needed. The question underneath is whether an unrecovered device stays open in the vendor's system indefinitely or reaches a defined end state, since an open record nobody closes is how a fleet count drifts from reality one leaver at a time.
Getting a baseline before you agree a number
An SLA negotiated without knowing your current performance is a guess, and a guess written into a contract becomes the standard against which everything is judged for the length of the term.
Measure what you do today, for a month. Order to hand on every device you issue, by market. It is tedious and it is a single spreadsheet, and it converts the entire negotiation from an argument about what sounds reasonable into a comparison against a known position.
Separate the easy cases from the hard ones while you do it. You will almost certainly find two distributions rather than one: domestic deliveries clustering tightly, and international ones spread widely. That shape is the argument for tiering and it is far more persuasive with your own numbers attached.
Ask the vendor what they actually achieve, as distinct from what they will commit to. Those are different numbers and most vendors will share the first in a conversation even where they will not contract to it. A vendor whose typical performance is well inside their proposed commitment has built in headroom, which is sensible of them and useful for you to know.
And find out what your people currently experience, which is not the same as what your records show. A handful of conversations with recent joiners will tell you whether the four days in your spreadsheet felt like four days or like a fortnight of uncertainty with no communication.
The baseline also protects you in the other direction. Companies occasionally negotiate hard for a commitment they were already comfortably beating, spending negotiating capital on a clause that changes nothing. Knowing the starting position prevents that.
Exceptions are the number that matters
Averages hide everything important in this category. A vendor delivering in four days on average, where ninety per cent arrive in two and ten per cent take three weeks, looks identical on a monthly report to one delivering reliably in four. The experience is completely different and so is the work it generates for you.
Ask for percentiles rather than means. The ninetieth or ninety-fifth percentile tells you about the tail, and the tail is what your team spends its time on. A vendor unwilling to report percentiles is telling you about the shape of their distribution.
Count the exceptions separately. How many shipments required a human conversation rather than running through the normal path. This single number predicts your internal workload better than any delivery statistic, because it measures how often somebody on your side had to stop what they were doing. A vendor with excellent delivery times and a high exception count is consuming more of your team than its headline numbers suggest, and that cost appears nowhere in the contract.
And ask for the breaches listed, not summarised. A report saying ninety-four per cent compliance is less useful than a list of the six that failed, with what happened to each. The list is actionable; the percentage is a grade. It also surfaces patterns a percentage cannot: six breaches spread evenly across markets is noise, while six breaches in the same country is a coverage problem that will recur next month and the month after until somebody names it.
Set the target on the tail rather than the average. A commitment that ninety-five per cent of tier one deliveries reach the person within the target is a meaningfully different promise from an average, and it is the one that matches what people experience.
The commitments that are not about time at all
Delivery speed dominates these conversations and several non-temporal commitments are worth more in practice.
Configuration accuracy. What proportion of devices arrive built to the agreed specification, enrolled correctly and with everything in the box. A machine that arrives on time missing its power adapter has met the delivery SLA and failed the person waiting for it. Ask for first-time-right as a measure.
Communication. A commitment that the recipient is told when the device dispatched and given tracking, and told if it is delayed. This costs the vendor nothing and it is most of what the employee experiences. A device arriving on day nine with two updates feels better than one arriving on day seven in silence.
Data accuracy. A commitment about the asset record: that assignments are updated within a stated period of delivery and collection. Register drift starts here, and a commitment on it is unusual and cheap to ask for.
Named contact continuity. Not that your account manager never changes, which nobody can promise, but that you are told when it happens and given a handover. The unannounced change is the one that costs you.
And responsiveness to queries, separately from delivery. Time to first meaningful response on a support request, which is different from time to resolution and considerably easier to commit to.
These rarely appear in standard templates and vendors generally accept them, partly because they are reasonable and partly because they are not the clauses where commercial risk sits. That makes them cheap to obtain and disproportionately useful to have.
Remedies, and whether they are worth negotiating
Most equipment SLAs carry service credits, and most service credits are never claimed.
The reasons are structural. The amounts are small relative to the effort of claiming, the claim process requires somebody to assemble evidence, and the person who would do it is the person already dealing with the consequence of the breach. So the credit exists, is theoretically available, and sits unused while the service degrades.
Three alternatives are usually worth more.
An escalation commitment. On a breach, a named person at the vendor contacts a named person at your company within a stated period with an explanation and a remedy. This is free for the vendor to give, costs you nothing to use, and changes behaviour more than a credit does because it creates a conversation rather than an accounting entry.
A review trigger. Breaches above a stated threshold in a quarter trigger a formal service review with both sides present. Vendors take this seriously because it is visible internally on their side.
And a termination right tied to sustained failure. The strongest remedy available and the one that most changes behaviour. Persistent breach over a defined period releases you from the term without penalty. Vendors resist it and it is the clause most worth spending negotiation on, because it is the only one whose existence you never need to act on for it to work.
If you do take credits, make them automatic rather than claimable. A credit applied by the vendor on breach without you asking is a real remedy; one you have to request is a remedy for a company with more time than yours.
Reporting you do not have to ask for
Anything requiring a request will stop being requested. Not through negligence, but because requesting it is a task on somebody's list that competes with work that has deadlines.
Monthly, unprompted, to a named recipient. Agree it in the contract including the format and the day. A report that arrives automatically gets read; one you request gets requested twice and then forgotten.
Include the exception list. Breaches itemised with what happened, as above.
Include retrieval separately. Open collections, their age, and anything stalled. This is the single most useful operational report in the whole arrangement and it is almost never provided by default.
And include the asset position. What the vendor believes you have and where, so you can compare it against your own record. Divergence between those two is the early signal of a register problem, and catching it monthly is much cheaper than catching it during a migration.
| What to ask for | Frequency | Why it matters |
|---|---|---|
| Order to hand by market, percentiles | Monthly | The actual service, not the dispatch proxy |
| Breach list with narrative | Monthly | Actionable where a percentage is not |
| Exception count | Monthly | Predicts your internal workload |
| Open retrieval ageing | Monthly | Stalled collections are otherwise invisible |
| Asset position | Monthly or quarterly | Early warning of register divergence |
Running the review meeting so it does something
An SLA produces a monthly report and a quarterly meeting, and both can be theatre. The difference is almost entirely in how the meeting is run.
Open with the exceptions, not the compliance figure. The percentage is context and the breaches are the content. Starting with ninety-four per cent frames the conversation as broadly successful before anybody has looked at the six failures.
Ask what changed rather than what happened. A vendor explaining an individual breach is answering a question about the past. A vendor explaining what they altered as a result is answering a question about the future, which is the one that affects you.
Track recurrence explicitly. Keep a short list of issues raised and whether they recurred. Three consecutive quarters mentioning the same market is a pattern that a single meeting cannot see and that nobody will remember unaided.
Bring one thing from your side. Equipment arrangements fail in both directions, and a review where the customer arrives with a list and no acknowledgement of their own approval delays or stale address data gets defensive quickly. Naming one thing you will fix changes the tone and usually gets more in return.
And invite the person who does the work. Reviews attended only by account managers and procurement produce accounts of the relationship. The operations person on each side knows what actually happened, and their presence shortens the meeting considerably.
If a quarterly review has produced nothing for three cycles, that is worth noticing. Either the arrangement is genuinely working, in which case move to twice yearly and save everybody the time, or nobody is bringing anything to it, in which case the SLA is not being used.
What not to bother asking for
Negotiation time is finite and some clauses are not worth the trade.
Uptime on the portal. It matters far less than it sounds, and it is the clause buyers most often lead with. Equipment logistics are not real-time, and an hour of portal downtime costs you almost nothing beyond mild irritation. Spend the negotiating capital elsewhere.
Aggressive targets in markets you barely use. A tight commitment in a country where you ship twice a year is a clause you will never test, traded against something you would have used monthly.
Penalties large enough that the vendor prices them in. A remedy severe enough to represent real risk gets covered in the fee, which means you are paying for insurance against a failure that mostly will not happen.
And anything you have no way to verify. An SLA you cannot measure independently is one you have agreed to take on trust, which is fine as long as you recognise that is what it is. Prefer commitments you can check against your own records.
The definitions that cause arguments later
Most SLA disputes are definitional rather than performance-related, and all of them are preventable with a paragraph agreed in advance.
When does the clock start? When you submit the request, when the vendor acknowledges it, or when approval completes on your side. These can differ by days. Internal approval delays are genuinely yours and should be excluded, but the exclusion should be written rather than assumed by one party.
What counts as delivered? Handed to the named person, left at the address, signed for by somebody, or delivered to a reception desk. In home-delivery markets this matters enormously, and a device signed for by a neighbour has met one definition and not another.
What is a working day? Whose calendar, whose public holidays. In a multi-market arrangement this is not pedantic: a three-working-day commitment spanning a local holiday week means something different in each country.
What stops the clock? Customs holds, failed delivery attempts where the recipient was unreachable, your own approval delays. Each is legitimate and each should be enumerated, reported when invoked, and capped so that an indefinite hold does not become an indefinite exclusion.
And what counts as one request? A person needing a laptop, a dock and a monitor is one onboarding and possibly three shipments. Measuring per shipment flatters the vendor; measuring per person reflects the experience. Agree which.
Thirty minutes on these five at contract stage prevents the conversation where both parties are reading the same clause and reaching opposite conclusions, which is the single most common way an SLA stops functioning as a shared reference point.
Where the vendors fit
Established platforms will usually have a standard SLA and will negotiate at the margins. The standard one almost always measures dispatch, so the order-to-hand change is the first thing to raise and the most likely to be accepted, since it is reasonable and they get asked.
Smaller vendors may have no formal SLA at all, which is not disqualifying and is often honest. Write a simple one yourself: order to hand per tier, retrieval actions, monthly reporting, escalation on breach. Four commitments in plain language beat a template nobody has read, and a smaller vendor will frequently agree to more than a large one because the clauses are being written rather than defended.
Bundled arrangements, where equipment sits inside a larger platform, frequently have equipment SLAs that are thinner than the rest of the contract because it is a secondary product. Read that section specifically rather than assuming it matches the service levels on the primary product.
RemoAsset is our own product and that is the disclosure rather than a footnote. It covers procurement and retrieval across 80+ countries and publishes a Subscription tier from $1,499 a year, which includes SLA-backed priority support at that tier while the pay-as-you-go tier does not, and that difference is worth knowing before you assume a commitment applies. Three limits relevant to an SLA conversation specifically: it enrols devices but is not an MDM in its own right, so do not write endpoint management commitments into an equipment contract; it wipes devices but is not a certified ITAD, so disposal evidence commitments may need to sit with a separate provider; and coverage and lead times vary by market, which is exactly why the per-tier approach above applies to us as much as to anybody and why a single global number from any vendor in this category should be treated sceptically.
The decision table
| Situation | Scale | Setup | Primary Pain | Recommended Starting Point |
|---|---|---|---|---|
| First equipment contract | 50 to 200 | No prior SLA | Do not know what to ask | Order to hand, two tiers, monthly report |
| Existing contract, service drifting | Any | SLA met, experience worse | Measuring the wrong event | Check what the clock actually measures |
| Many markets, uneven coverage | 200 plus | One global target | Breaches mean nothing | Re-tier using your own markets |
| Retrieval is the problem | Any | No retrieval commitment | Collections stall invisibly | Commit vendor actions, add ageing report |
| Small vendor, no template | Any | Nothing formal exists | Over-engineering it | Four plain commitments, not a schedule |
| Reversible arrangement | Any | Month to month | SLA is the wrong lever | Skip it, keep the exit |
What you owe the vendor
SLAs are written as obligations flowing one way and most of them have dependencies flowing the other, which is where a good proportion of breaches actually originate.
Accurate addresses. A delivery failing because the address on file is two years old is not a vendor failure, and your address data is probably worse than you think. Confirm at the point of request rather than relying on the record.
Timely notification of leavers. Retrieval commitments start from your notification, and a vendor told three weeks after somebody left is working with a cold trail. If your HR process notifies late, no retrieval SLA will save you.
Approvals that do not sit. If your internal approval takes four days before the request reaches the vendor, the person waited four days and the vendor's clock had not started. That is legitimately excluded and it still happened to the employee, which is why your own cycle time belongs in the same report.
Forecasting where you can. A vendor given notice of a twenty-person intake can prepare for it. One discovering it as twenty separate urgent requests cannot, and the resulting delays will show as breaches that were partly yours.
And a single route in. Requests arriving through three channels from four people produce exactly the inconsistency you will later complain about.
Raising these yourself at contract stage is worth doing, both because it makes the SLA achievable and because it changes the relationship. A customer who acknowledges their own dependencies gets better engagement on the commitments they are asking for, and the vendor's own numbers improve for reasons that benefit you.
What getting this wrong actually costs
The visible cost is a breach you cannot claim against, which is usually small.
The larger cost is the one in the story at the top of this article: an SLA that is met while the service degrades. That is worse than having no SLA, because the reports provide active reassurance. Everybody stops paying attention, the drift continues, and the problem is discovered by accident months later. A number that is wrong and monitored is more dangerous than no number at all.
The cost with the longest tail is retrieval that nobody is accountable for. Collections that stall quietly accumulate into a gap between your register and reality, and that gap is invisible until something forces a count. An ageing report on open retrievals, which costs the vendor nothing to produce, would surface it monthly.
So the question is not whether your vendor meets its SLA. It is whether the thing being measured is the thing you care about, because in most equipment contracts it demonstrably is not.
A minimum viable SLA, if you only get one page
For a smaller arrangement, or where the vendor has no template and you are writing it yourself, this is the short version. Five commitments in plain language, which beats a long schedule nobody reads.
One. Order to hand, measured from request to the person having a usable device, with three market tiers and a target for each. Customs holds and unreachable recipients stop the clock, are reported when invoked, and are capped.
Two. Retrieval actions: days from notification to first contact attempt, days to book once the person responds, number of attempts before the vendor tells you it has stalled.
Three. A monthly report sent unprompted to a named person, containing order to hand by market at the ninetieth percentile, an itemised list of breaches with what happened, open retrievals by age, and the vendor's view of your asset position.
Four. On breach, a named person at the vendor contacts a named person at you within two working days with an explanation and a remedy.
Five. Sustained breach over a defined period releases you from the remaining term without penalty.
That is the whole thing, and it covers more of what matters than most multi-page schedules. If you can only get three of the five, take one, three and five, because measurement plus visibility plus an exit covers the rest by implication.
What to leave out of a short version: portal uptime, detailed credit calculations, and anything you cannot verify from your own records.
When to revisit what you agreed
SLAs are set once and then inherited, which is how a dispatch-based measure survives three renewals.
Three triggers are worth acting on. A new market, because your tiering was built on the footprint you had and a new country may not fit any existing tier. A change in the vendor's partner network, since a market that moves from in-house to partner may no longer support the commitment you agreed. And a pattern in the exception list, where the same market or the same failure mode appears repeatedly, which is evidence that the target there is wrong rather than that the vendor is failing.
The renewal conversation is the natural moment and it is usually treated as a pricing discussion. Spending part of it on the measurement is better value, because a better-defined SLA improves the service you receive for the whole term while a small discount improves only the invoice.
And if you take one thing into that conversation, make it the clock. Ask what event starts it and what event stops it, and whether the answer matches what the people inside your company believe they are being promised.
Frequently asked questions
What is the most important thing to get into an equipment SLA?
Change what the clock measures. Most contracts commit to days from order to dispatch, which is the vendor's convenience, while everybody internally assumes the number means the new joiner has a working machine. Those two drift apart over time, usually because customs and the last mile sit outside the measured window, and the SLA can be met every month while the actual experience deteriorates. Ask for order to hand, with documented exclusions for customs holds and unreachable recipients rather than silent ones.
Should we have one SLA target or different ones per country?
Per market tier, built from your own footprint rather than the vendor's. A single global target is either loose enough to commit nobody to anything or tight enough that it is breached constantly in the hard markets, at which point breaches stop carrying meaning. Three tiers usually suffice: a tight target where most of your people are, a realistic one plus reporting in secondary markets, and in genuinely difficult markets a commitment to respond with a binding quote rather than a standing lead time.
How do you write an SLA for retrieval when it depends on the employee?
Commit the vendor to their own actions rather than to the outcome. Days from your notification to the first contact attempt, days to book a collection once the person responds, how many attempts before escalation and what escalation means. Add a completion rate with a stated denominator, such as collections completed within a window where the leaver responded, which is fair to the vendor and still meaningful. Then define the stopping point, because the common failure is a retrieval sitting open indefinitely with nobody reporting that it has stalled.
Are service credits worth negotiating?
Usually less than three alternatives, because credits are small relative to the effort of claiming and the person who would claim is already dealing with the consequence of the breach. An escalation commitment, where a named person contacts you within a stated period with an explanation, costs the vendor nothing and changes behaviour more. A review trigger at a breach threshold is taken seriously internally on their side. And a termination right on sustained failure is the strongest remedy available, and works without ever being used.
What reporting should we ask for?
Monthly, unprompted, to a named person, with the format and the day agreed in the contract, because anything you have to request will stop being requested by month four. Include order to hand by market as percentiles rather than averages, an itemised breach list with what happened in each case rather than a compliance percentage, an exception count, and an ageing report on open retrievals. Add the vendor's view of your asset position so you can compare it against your own record.
Why percentiles instead of averages?
Because averages conceal the tail and the tail is where your work comes from. A vendor averaging four days, where ninety per cent arrive in two and ten per cent take three weeks, reports identically to one delivering reliably in four, and the two experiences are entirely different. The ninetieth or ninety-fifth percentile describes what the bad cases look like. A vendor unwilling to report percentiles is telling you something about the shape of their distribution that the average was hiding.
What should we not bother negotiating?
Portal uptime, which matters far less than it sounds because equipment logistics are not real-time and an hour of downtime costs almost nothing. Aggressive targets in markets you use twice a year, which you will never test and which trade against something you would use monthly. Penalties severe enough that the vendor prices them into the fee, since you then pay for insurance against a failure that mostly will not occur. And anything you have no independent way to verify.
When should we revisit an SLA we already have?
At three triggers rather than on a schedule. When you open a market, because your tiering reflects the footprint you had. When the vendor's partner network changes in a market you use, since a country moving from their own operation to a partner may no longer support the agreed commitment. And when a pattern appears in the exception list, where the same market or failure mode recurs, which usually indicates the target is wrong rather than that the vendor is failing. Renewal is the natural moment and is usually spent entirely on price.
Where this leaves you
An SLA is only as good as the event it measures, and most equipment SLAs measure the easy half of a journey whose difficulty sits elsewhere.
Move the clock to order to hand, with exclusions written down rather than assumed. Tier your markets using your own footprint. Commit the vendor to their own actions on retrieval and get an ageing report on open collections. Ask for percentiles and an itemised breach list rather than a compliance percentage. And take an escalation commitment over a service credit you will never claim.
Then ask the question the company in the opening story never asked: what event starts this clock, and what event stops it.
Two habits make the difference over a three-year term. Measure your own baseline before you agree anything, so the negotiation is a comparison rather than a guess. And read the monthly report starting from the exception list rather than the compliance figure, because the percentage is designed to reassure and the exceptions are where the service actually lives.
Everything else in this article is detail hanging off those two. A company that knows its own numbers and reads the failures will notice drift within a quarter. A company that receives a green report every month and files it will notice when somebody finally asks why onboarding feels slower than it used to, which in the story at the top of this article took eight months.
