A master services agreement with a staffing vendor, schedule C, clause 4.2: “Supplier attrition shall not exceed 15% per annum. For each percentage point above this threshold, Supplier shall pay liquidated damages of €5,000.” Procurement signs off because the risk has been named, priced and shifted to the other side. Legal signs off because the number is specific. The VP of Engineering signs because nobody wants people to leave.
Twelve months later the quarterly business review shows attrition of 0%. Over the same period, four of the twelve people on the team changed. Two had resigned, but after a conversation with the vendor’s account manager the engineering lead agreed they “hadn’t really been a fit”, so the departures were logged as client-requested replacements. One was moved to another account “at the specialist’s own request”, and since it coincided with the end of a module it was booked as a scope change. The fourth left in week two, inside the guarantee window the contract excluded from the count. All of it was within the letter of the agreement. The project lost roughly a quarter of momentum.
This article is for procurement leads and CTOs who draft or review contracts with team-augmentation vendors, whether the vendor sits in Poland, Portugal or Pennsylvania. The argument may be unwelcome to anyone who likes a single KPI: a percentage attrition cap backed by liquidated damages is one of the weakest protections you can put in an outsourcing contract. What protects you better is a set of separated event definitions, a handful of continuity metrics reported monthly, specific obligations attached to each event, and damages only for the events the vendor genuinely controls.
Why does a single attrition percentage measure the wrong thing?
Because a project does not lose value when someone leaves. It loses value in proportion to the continuity that leaves with them. A specialist who spends three weeks handing over to a successor, whose module is documented and whose knowledge is shared with two colleagues, costs little to lose. The only person who understands the billing-system integration, leaving on two weeks’ notice with no overlap, can stall a release. In an attrition percentage, both events weigh exactly the same.
The second problem is the numerator. Attrition is a ratio, and anyone who influences the numerator influences the result. If the contract does not say precisely which departures count, a vendor who pays for a high number will naturally look for categories where departures do not count. No bad faith is required. It is enough that the account manager sincerely believes the specialist “wasn’t going to work out” and that the client’s engineering lead is too busy to argue.
The third problem is timing. An annual cap is assessed once a year, but a project loses momentum in weeks. When damages are calculated after twelve months, the information arrives too late to fix anything, and the discussion about which departure counts happens in the middle of a money dispute rather than a joint effort to stabilise the team.
The fourth problem is the denominator. On a four-person team one departure is 25%. A 15% cap on that team means, in practice, nobody may leave for a year: a commitment no vendor can honestly make, because no vendor controls people’s life decisions. A contract that contains an impossible obligation from day one does not protect the client. It just hands the client an argument for a future dispute.
Does that mean you should not measure attrition? Quite the opposite. You should measure it more carefully than one line in a schedule allows. The cost of losing a specialist is real, as the breakdown in our piece on employer branding and the cost of losing a specialist shows. The point is to make the measurement describe what the project loses, and to make the contract assign obligations to whoever can prevent the loss.
How is attrition different from turnover, replacement and separation?
Vendor reports and contracts use attrition and turnover as if they were synonyms. In HR practice they often are not: attrition frequently means headcount that is lost and not backfilled, turnover means departures that are replaced. In outsourcing the line is blurry and vendors switch between the words freely. So the first practical conclusion is simple: do not assume the vendor reads “attrition” the way you do. Define it in the contract.
A useful starting point comes from the US Bureau of Labor Statistics. Its Job Openings and Labor Turnover Survey (JOLTS) splits separations into three groups: quits, meaning employees who left voluntarily; layoffs and discharges, meaning involuntary separations initiated by the employer; and other separations, such as retirements, transfers to other locations, deaths and separations due to disability. BLS computes the rate by dividing separations by employment and multiplying by 100. That taxonomy does not map one-to-one onto outsourcing, because a client–vendor–specialist relationship has three parties rather than two, but the logic carries over: before you calculate a percentage, establish who initiated the departure.
In an outsourced project, a specialist can leave the team because the specialist decided to (resigning from the vendor or asking to leave the project), because the vendor decided to (reassigning the person or ending the engagement), or because the client decided to (requesting a replacement, ending the role, shrinking the team, hiring the person directly). Then there are events nobody initiates: long-term sickness, parental leave, an accident. Each group tells you something different and each calls for a different contractual response.
Hence a simple drafting rule: never use the word “attrition” in a contract without a definition, and build the definition around who initiated the event and whether the vendor could influence it. Anything the client initiated must not count against the vendor. Anything the vendor initiated should count most heavily, because it is the one category where the decision sat entirely with the vendor.
Which events should the contract definitions separate?
The table below is a proposed glossary you can lift into a contract schedule. Each event gets its own metric and its own obligation. The important point is that client-requested replacements and guarantee-window departures do not disappear from the report. They are reported separately, because they measure something other than team stability.
| Event | Initiated by | Counts toward vendor attrition | What it measures | Vendor obligation |
|---|---|---|---|---|
| Specialist resignation | Specialist | Yes | Team stability | Notice, replacement, overlap |
| Reassignment to another client or project | Vendor | Yes, flagged separately | Vendor priorities | Client consent beforehand, replacement at vendor’s cost |
| Vendor ends the engagement (e.g. misconduct) | Vendor | Yes | Selection and supervision quality | Immediate notice, replacement |
| Departure within the guarantee window | Specialist or vendor | No, separate metric | Matching accuracy | Free replacement within guarantee terms |
| Client-requested replacement | Client | No, separate metric | Matching accuracy | Replacement within the agreed period |
| Role ended or team reduced | Client | No | Changes on the client side | Knowledge transfer within the notice period |
| Specialist converts to client payroll | Client and specialist | No | Conversion | Per the conversion-fee clause |
| Unplanned absence (sickness, parental leave) | Nobody | No, separate continuity category | Team resilience | Temporary cover, if agreed |
Two rows deserve comment. The first is vendor reassignment. From the client’s perspective it is the worst kind of departure: the specialist is not leaving the market or the vendor, just moving to whoever the vendor currently ranks higher. It often happens under the banner of “the specialist asked for a change”. That may well be true, but the contract should require client consent for the move and treat it as vendor-initiated, because the vendor decides whether to grant the request.
The second is the guarantee window. Most vendors offer an initial period in which a poorly matched specialist is replaced at no cost. At ARDURA that guarantee runs for two weeks, and if a specialist leaves during the engagement, a replacement is guaranteed within 5 business days. Excluding the guarantee window from attrition is reasonable; excluding it from the report is not. If one in three people leaves in the first two weeks, matching is broken, however good the attrition number looks.
How do you calculate the attrition rate step by step?
The formula is simple. The hard part is deciding what goes into the numerator and the denominator. A version suitable for a contract:
Attrition rate (rolling 12 months) = number of events counted as attrition in the last 12 months / average number of filled seats over those 12 months × 100
The numerator is the sum of the first three rows of the table: specialist resignations, vendor reassignments and vendor-ended engagements, excluding the guarantee window. The denominator is the average of the filled-seat count at the end of each month. Count seats, not people: one seat filled by three successive people during the year is still one seat with two attrition events.
An illustrative example. A team averaged 12 filled seats over the year. Two specialists resigned and the vendor moved one to another client. One left within the guarantee window and the client asked for one replacement. The numerator is 3, the denominator 12, attrition 25%. Separately reported: early replacements 1, client-requested replacements 1. The same team under a one-bucket definition would show 5 events and 42%; under definitions as loose as the ones in the opening story, 0%. The gap between 0% and 42% is not reality. It is drafting.
Why a rolling 12-month window rather than a calendar year? Because a quarterly review should always look at the same horizon, not at a number that is zero by construction in January and collects the whole year in December. A rolling window also smooths single events and shows the trend. Watch for a common mistake: multiplying a monthly figure by 12 does not give annual attrition. One departure in a month when the team had 10 people becomes “120% annualised”, which has little to do with anything.
It is also worth calculating a vendor reassignment rate on its own, using only the second category over the same denominator. Resignations are partly outside the vendor’s control; reassignments are entirely within it. If the vendor blames attrition on a hot labour market and half the numerator turns out to be reassignments, the conversation changes.
Why does the percentage mislead on small teams, and what should you track instead?
Because with a small denominator every event moves the result by double digits. On a five-person team one departure in a year is 20%, two are 40%. That number cannot distinguish bad luck from a real problem. Two departures from a team of five in a year might be coincidence (a relocation, a career change) or a symptom of a vendor systematically paying specialists below market. The percentage will not tell you which.
A reasonable contract rule: for teams below ten filled seats, report absolute event counts in each category, and calculate the percentage only at the level of the whole vendor relationship if it spans several teams. The ten-seat threshold is not drawn from any standard. It is a pragmatic point above which a single event no longer shifts the annual figure by more than ten percentage points.
What to track instead of a percentage on a small team? Three things. First, average tenure on the project: how many months the current line-up has, on average, worked on it. If eighteen months into the engagement the average tenure is four months, the team is effectively always new, whatever the percentage says. Second, vacancy days: the total working days during which a seat was empty or held by someone still onboarding. Third, key-role stability: whether the people the client named as critical at the start (an architect, a team lead, the integration specialist) are still on the project.
That last measure matters most to a CTO. Not every seat is equally important, and a contract does not have to pretend otherwise. Naming two or three key roles in a schedule and giving them stricter terms (longer notice, mandatory overlap, client consent for any change) protects a project better than one cap for everyone.
Which continuity metrics complement attrition?
Attrition answers the question of how often someone leaves. A project also needs to know how much it loses each time. A few simple metrics cover that, and a vendor can report them without special tooling if it collects the data from the start.
Time to replacement. Working days from the day the vendor learned of a departure (or the day the client requested a replacement) to the successor’s first working day. It is the metric most easily compared with the contractual promise. If the vendor commits to replacement within 5 business days, the report should show how often that promise was kept.
Overlap. Working days during which the departing specialist and the successor worked in parallel. Overlap is the cheapest knowledge-transfer tool there is, and the first thing cut under cost pressure. Zero overlap on a departure with notice is a sign the vendor started looking for a successor too late.
Vacancy days. Total working days a seat stood empty. Under time-and-materials billing the client formally does not pay for an empty seat, but it pays in delay and in overload for the rest of the team. This metric lets you put a number on it.
Early replacements. The share of people who left or were replaced within the guarantee window. It measures matching quality and complements attrition well, because a vendor can show low post-guarantee attrition precisely because poorly matched people drop out early and are never counted.
Average tenure on the project. As above: the average number of months the current team members have worked on the project. It is the best single proxy for accumulated knowledge of the domain and the codebase.
Keep the proportions sensible. Five continuity measures plus attrition are enough to see a problem and few enough that someone will read the report. Ten more will produce a report that is complete and unread. Continuity also depends on the client: a solid process for onboarding external specialists shortens the time a successor needs to reach full productivity and lowers the cost of every departure.
A note on cost, for when you need to show the board the scale. Each of these metrics converts into money. At an illustrative day rate of €600 for a senior engineer (a round number for the arithmetic, not a quote), ten working days of overlap cost €6,000. That is typically the sum the parties argue over when deciding who pays for knowledge transfer. The cost of onboarding a successor with no overlap is harder to calculate but usually higher, because it includes the time of everyone else on the team and the slippage that follows.
What should a vendor’s monthly attrition report look like?
It should fit on one page and arrive unprompted on an agreed day each month. Attach the format to the contract as a template: it forces both sides to agree definitions before the first departure rather than after it.
| Report line | This month | Rolling 12 months | Contract target | Vendor comment |
|---|---|---|---|---|
| Average filled seats | – | |||
| Specialist resignations | – | Reason category, if known | ||
| Vendor reassignments | 0 without client consent | Date of client consent | ||
| Vendor-ended engagements | – | |||
| Attrition rate (teams ≥ 10 seats) | – | To be agreed | ||
| Early replacements (guarantee window) | – | |||
| Client-requested replacements | – | |||
| Average time to replacement (working days) | e.g. ≤ 5 | Breaches | ||
| Average overlap (working days) | e.g. ≥ 5 for key roles | |||
| Vacancy days | – | |||
| Average project tenure (months) | – | |||
| Changes in key roles | 0 without client consent | |||
| Known departure risks in the next 60 days | – | – | Without personal data where required |
The last line is the most valuable and the one least often filled in. A vendor often knows before the client that a specialist is looking for a new project, that a contract is ending or that a competing offer is on the table. A contract cannot force a vendor to disclose what it may not lawfully disclose, but it can require a general risk flag for a specific role. A client who learns of a risk sixty days ahead can plan overlap, spread knowledge and shift priorities. A client who learns of it with the resignation can only react.
One practical caution: the report should not contain more personal data about specialists than project management genuinely requires. In the EU that is a GDPR data-minimisation question; everywhere it is a matter of decency towards the people the report describes. Record departure reasons by category (compensation, location, growth, personal) rather than detail.
Once a quarter, review the report in the vendor governance meeting. Not to argue about the percentage, but to answer three questions: could any event have been spotted earlier, were the contractual obligations (notice, replacement, overlap) met, and is the distribution of knowledge in the team safe if one more person leaves?
Which attrition clauses belong in the contract?
The clauses below form a coherent protection package. Not every contract needs all of them. For two specialists over six months, the first three are enough; for a team of fifteen over several years, it is worth having the full set.
Event definitions. The glossary from the table above, stating clearly which categories count toward attrition and which are reported separately. Without this clause every other one can be worked around.
Notice obligation. The vendor notifies the client of any departure promptly, and no later than a set number of business days after the vendor itself learns of it. It is also worth requiring the vendor’s own contracts with specialists to carry a notice period no shorter than the one in the client contract. If the client has 30 days’ notice towards the vendor (the standard at ARDURA) but a specialist can leave the vendor on a week’s notice, the client’s protection is illusory.
No reassignment without consent. The vendor may not move a specialist to another project or client without the client’s written consent, and the client will not unreasonably withhold consent if the vendor provides a successor with overlap. The second half matters: without it the clause becomes a ban on the specialist’s career development, which in turn raises the risk of resignation.
Replacement within a deadline. The vendor presents successor candidates within a set number of business days, and the client may interview and approve them. The deadline should match the role. Five business days is realistic for mainstream technology; for niche skills, set a longer deadline but require interim cover.
Overlap at the vendor’s cost. For departures that count as attrition, the vendor provides an agreed number of days of parallel work between the leaver and the successor and does not invoice the client for one of them. This is the clause that most directly ties the cost of attrition to the party who can prevent it, which is why it works better than a penalty on the percentage. A vendor who knows that every departure costs two weeks of double staffing has a very concrete reason to retain people and to flag risks early.
Knowledge transfer as a vendor obligation. On every departure the vendor is responsible for handing over documentation, access and work in progress against an agreed list. We have set out a detailed plan for this in the article on offboarding a contract specialist, so here only the principle: the contract should say who is accountable and what gets handed over, rather than relying on goodwill.
Report and review. The monthly report template as a schedule, its delivery date and a quarterly review attended by a decision-maker from the vendor’s side.
Remedy for repeated events. Instead of damages for exceeding a percentage: if the number of attrition events in a rolling 12-month window exceeds an agreed level, the parties agree a remediation plan within a set time, and if the plan is not delivered the client may terminate the affected part of the contract without the standard notice period. This protection works because it gives the client a way out, not just a claim.
Bear in mind that these clauses interact with the commercial model. When the client pays for named people’s time, attrition is visible and easy to attribute. Under outcome-based pricing or a managed-team subscription, the vendor can argue that team composition is its own business as long as it delivers. Even then, keep the notice obligation and key-role protection, because it is the client who is left holding the system knowledge when the contract ends. For a broader list of clauses that frame this kind of engagement, see the body leasing contract checklist.
When do liquidated damages for attrition help, and when do they hurt?
Start with what the law will tolerate. Common-law courts police clauses that look punitive. In Cavendish Square Holding BV v Talal El Makdessi [2015] UKSC 67, the UK Supreme Court restated the test: a clause is penal if it is a secondary obligation that “imposes a detriment on the contract-breaker out of all proportion to any legitimate interest of the innocent party in the enforcement of the primary obligation”. Civil-law systems take a different route to a similar place. Polish law, relevant whenever the parties to a contract with a Polish vendor choose it, allows contractual penalties for non-monetary obligations (Article 483 of the Civil Code) and lets a court reduce a penalty that is grossly excessive (Article 484 § 2). Whatever the governing law, a penalty that bears no relation to real loss invites a fight.
That raises the first question: which obligation has the vendor breached when a specialist resigns of their own accord? If the contract obliges the vendor to “ensure attrition below 15%”, it is in practice an obligation to achieve a result the vendor only partly controls. The damages attached to it work less as discipline than as a risk premium, which the vendor will price into the rate. The client pays for it up front, whether or not anybody leaves.
The second question is what behaviour the clause produces. Damages on a percentage create an incentive to keep departures out of the numerator. The cheapest way to do that is not to retain the specialist but to reclassify the event, exactly as in the opening story. The same incentive discourages early risk flags, because an early warning is an admission of a problem that carries a penalty.
When do damages help? When they attach to an event the vendor fully controls and that is easy to establish. Three good examples: moving a specialist to another client without consent, failing to give notice of a departure within the agreed time, and failing to present a successor within the agreed period. In each case the vendor made a decision or failed to take an action within its reach, and establishing the breach does not require an argument about categories. Tie the amount to a realistic estimate of loss (vacancy days at the contracted rate, for instance), which also makes it easier to defend.
Present this to procurement not as a concession to the vendor but as stronger protection. A damages clause that cannot be sidestepped by reclassification protects better than a larger one that depends on definitions. And the overlap-at-vendor’s-cost clause described above works like damages without the dispute: the cost of attrition lands on the party who can prevent it, at the moment it happens.
How should you draft non-solicitation and conversion fees?
Attrition has another side. Vendors worry that the client will hire their best specialists directly, so they write non-solicitation or no-hire clauses into the contract. For the client this matters for continuity, because a specialist converting to the client’s payroll is one of the best possible “departures”: the knowledge stays on the project and even deepens.
Mind the regulatory context. On 2 June 2025 the European Commission fined Delivery Hero and Glovo a total of €329 million for taking part in a cartel in which, among other things, the companies agreed not to poach each other’s employees. The Commission described it as its first decision finding a cartel in a labour market. That case concerned an agreement between competitors, and a clause in a client–vendor contract is different in kind: it sits inside a commercial relationship in which the vendor is protecting its recruitment investment. Still, the direction of travel is clear. Competition authorities look harder at arrangements that restrict worker mobility than they did a few years ago, and broad, long or absolute bans deserve a review by counsel in your jurisdiction.
In practice, an agreed conversion fee usually works better than a ban, often tapering as the engagement lengthens. It protects the vendor’s legitimate interest (recovery of recruitment cost and lost margin) and gives the client a predictable path to keep the knowledge in-house. For attrition reporting it is also fairer: conversion does not count against the vendor, because continuity is not lost, merely transferred.
Draft two points precisely. First, whether the clause covers only active solicitation or also a specialist who approaches the client independently, for example through a job advert. Second, how long after the engagement ends it applies. A clause with no end date, or one covering every employee of the vendor including people who never worked for the client, is hard to defend and breeds disputes. For when bringing a team in-house makes economic sense, see our analysis of why body leasing stops paying off when it starts working.
What should you ask a vendor about attrition before signing?
You learn the most about a vendor’s attrition before the contract is signed, because that is when both sides are motivated to agree rather than argue. These are the questions to put to every shortlisted vendor, with answers in writing.
How do you calculate retention or attrition? Every vendor has a number in its proposal. ARDURA cites 99% specialist retention. Ask every vendor, including us, for the definition: which events are in the numerator, what the denominator is, what period the figure covers and whether it includes the guarantee window. A number without a definition is marketing; a number with one is information you can compare across bids.
What is the average tenure of your specialists on client projects? This answer often says more than the retention figure, because it is harder to present flatteringly.
What notice period do your contracts with specialists carry? If it is shorter than the notice period in the client contract, ask how the vendor guarantees continuity when a specialist leaves at short notice.
What happens, step by step, when a specialist leaves? Who tells the client and when, where successor candidates come from, who pays for overlap. A good answer is concrete and describes steps, rather than assuring you that “it doesn’t happen”.
When do you move specialists between projects? “Never” is not very credible. An answer that describes when, and with whose consent, is more so.
Can you deliver a report in our format? If the vendor cannot populate the report from the previous section, it does not measure continuity itself, and therefore does not manage it.
Compare the answers across vendors during selection and carry the important ones into the contract. If a vendor promises replacement within five business days in its proposal, there is no reason that promise should not appear in a schedule. It helps to frame the comparison in terms of IT contractor engagement models and how each one allocates the cost of a departure, and of how the vendor keeps scalable IT teams stable as seats are added or removed.
What are the key takeaways for procurement and the CTO?
Specialist attrition in outsourced projects is a real risk, but one percentage in a contract schedule protects against it poorly. It counts departures rather than lost continuity, depends on definitions that are easy to stretch, produces random results on small teams and, combined with damages, teaches the vendor to reclassify events and to flag risks late.
Six things to keep. First, put event definitions in the contract and separate resignation, vendor reassignment, client-requested replacement, guarantee-window departure, role ended by the client and conversion. Second, calculate attrition over a rolling 12 months on seats, and for teams under ten seats report absolute counts. Third, add continuity metrics: time to replacement, overlap, vacancy days, early replacements, average tenure and key-role stability. Fourth, require a monthly report in an agreed format, including known departure risks. Fifth, instead of damages on a percentage, put the cost of overlap on the vendor and reserve damages for events the vendor controls. Sixth, replace an outright no-hire ban with a time-limited conversion fee.
If you are drafting a contract with a team-augmentation vendor, or comparing the continuity terms of several bids, see how our staff augmentation services handle it: a two-week replacement guarantee, a replacement within 5 business days if a specialist leaves, and 30 days’ notice. Or get in touch and walk through the definitions and the report template with us before they go into a schedule.