The BPO Margin Leak Playbook

The anatomy of a leak

When a people business loses margin, it usually looks for a culprit: a bad quarter, a client that churned, a wage settlement that went the wrong way, a project that overran. These are the events that make it into the board pack, and they are real. But they are not where most of the money goes. The larger loss is quieter and more continuous, and it does not appear as a line in any report, because it is not an event at all. It is a leak: a slow, distributed loss of margin through hundreds of small decisions, each defensible on its own, none of them recorded, and therefore none of them ever learned from.

The macro picture makes the stakes plain. The two giants of customer-experience outsourcing are the most visible barometer, because they have to disclose their numbers. Concentrix reported a non-GAAP operating margin of 12.8% for its 2025 financial year, down from 13.7% a year earlier, a fall of nearly a full percentage point.¹ Teleperformance, larger again, held its recurring EBITA margin at 15.0% but guided to only marginal improvement, in effect flat.² Neither company is badly run; both are among the most operationally sophisticated in the sector. If margin can compress at that level of scale and discipline, the leaks are not a symptom of incompetence. They are structural.

The demand side compounds the pressure. Deloitte’s Global Outsourcing Survey finds cost reduction is still the primary motivation for more than 70% of buyers, but that they now judge providers on fully-loaded cost, management overhead, technology, training and the hidden cost of attrition, rather than a simple price per seat.³That sophistication removes the provider’s room to hold margin through opacity. What is left is operational: the difference between a contract sold well and delivered well, and one that leaks at every join.

Margin is not lost in one big event. It leaks, quietly and continuously, through decisions no system was ever built to record.

Here is the thesis of this playbook, and it is worth stating before the leaks themselves. In a people business the product is a chain of decisions: which deals to chase, who to hire and when, how to staff a shift, what to charge, whether to renew. The operating stack records the consequences of those decisions in meticulous detail (the deal in the CRM, the hire in the ATS, the roster in the workforce tool, the invoice in the ledger) and almost nothing about the decisions themselves: what was known at the time, which constraint was expected to bind, who chose, and what they were betting would happen. Margin leaks through decisions no system recorded, which is why it cannot be found in the systems that recorded everything else.

This is why a people business can be busy, growing, and quietly unprofitable all at once, and why the three states are so easily mistaken for one another. The accounting system faithfully records revenue and cost, but it records them long after the decisions that set them, and it aggregates them to a level where the individual leaks cancel into a single number that looks merely disappointing rather than diagnosable. By the time a margin slip is visible in a quarterly result, the decisions that caused it are months old, undocumented, and impossible to reconstruct, so the response is necessarily blunt: a cost programme, a hiring freeze, a push on utilisation. These treat the symptom at the level of the whole business because the cause was never captured at the level of the individual decision. A leak you can only see in aggregate is a leak you can only fight in aggregate, which is expensive, slow, and usually aimed at the wrong place.

What follows is nine leaks, in roughly the order the money flows through a people business: from the deal, through delivery and the workforce, to price, to the renewal, and finally to the structural squeeze that AI is putting on the whole model. For each, three things: the mechanism (how the margin actually escapes), what it costs (cited where a real figure exists, reasoned or labelled illustrative where one does not), and the play (the specific, concrete move that stops it). The plays share a single shape, which we will name at the end, but each stands on its own. Read them as a diagnostic. Most people businesses are leaking through several of these at once, and cannot see any of them, because the evidence was thrown away at the moment the decision was made.

Leak one: selling what you cannot deliver

The most expensive margin leak in a people business happens before a single hour of work is done, at the moment a deal is qualified. A mis-scoped deal, one sold at a price or on a timeline the business cannot actually deliver, does not lose margin once. It loses it at every subsequent stage: it is staffed under pressure, delivered at a loss, over-serviced to keep the client calm, escalated, re-planned, and eventually renewed defensively or lost. The damage is set at qualification and paid out over the life of the account.

The mechanism

The leak opens because the decision to bid is made on the wrong evidence. A sales team judges a deal on what the CRM holds: stage, value, probability, close date. But in a people business the decisive question about a deal lives entirely outside the CRM: can we hire and deliver this, at the margin we are quoting, on the date we are promising? The CRM cannot answer it, because it cannot see the labour market, the delivery backlog, or the true cost to serve. So the question goes unasked, the deal is qualified on winnability alone, and a commitment is made that delivery had no part in agreeing to.

Win rate, the metric most sales teams steer by, actively hides the leak. It is a lagging indicator: it tells you which deals closed, not which ones you can keep or deliver profitably, which is why we treat win rate as a lagging indicator rather than a target. A team optimising for win rate will, rationally, chase the deals that are easiest to win, which are not the same as the deals that are most profitable to deliver. The factors that genuinely predict a good outcome sit outside the funnel arithmetic entirely, which is why we look at what actually moves win rates rather than the rate itself.

To make the mechanism concrete, and as an illustration rather than a specific customer, picture a staffing firm bidding to place a cohort of specialist engineers in a tight labour market. The CRM shows a warm client, a clear budget and a high close probability, so the deal is qualified and won. What the CRM cannot show is that the local supply of that specialism is exhausted, that the promised start date assumes a fill rate the firm has never actually achieved for that role, and that hitting the date will require paying above the bill rate to source contractors. Every one of those facts was knowable at qualification; none of them lived in the system the qualification decision was made in. The deal was not mis-priced by accident. It was priced without the evidence that would have shown it could not be delivered at that price, and the decision to bid recorded none of what it did not know.

What it costs

The cost is hard to see precisely because it is distributed downstream, but its scale is visible in the hit rate of formal bidding. Even well-run proposal functions win only around 43 to 45% of the bids they choose to enter.¹⁵ That means more than half the effort in a disciplined bid process is spent losing, and, more importantly, that the winning half is not selected for deliverability at all: it is selected for winnability. A deal won at a loss is worse than a deal lost, because it consumes delivery capacity that a profitable deal could have used. The true cost of a mis-scoped deal is therefore the sum of the loss on delivery and the opportunity cost of the capacity it burned, a figure no single system computes.

The play

Qualify against delivery reality, not against winnability. Before a deal is committed, it should be read as a governed decision, not a pipeline entry. Name the constraint that actually binds it: the scarce skill you would have to hire, the delivery date the backlog will not clear, the rate the true cost to serve will not support. Because a people business fails at the constraint that fails, not the average of its constraints, the worst constraint decides whether the deal is deliverable at all. Attach the evidence with its quality labelled: a delivery estimate that was measured from similar past work is not the same as one someone asserted in the deal review, and the evidence hierarchy stops the second from being weighed as the first. Record the choice to bid, or not to, as an object, so that when the account is underwater eighteen months later the reasoning can be reconstructed and, crucially, scored. Do this and the pattern becomes learnable: the shapes of deal that consistently deliver below forecast get flagged before the next one is signed. We have written at length on qualifying against delivery reality and on the cost of selling what you cannot deliver; the short version is that the cheapest margin you will ever protect is the margin you protect at qualification.

Leak two: over-servicing beyond the contract

The second leak is the most sympathetic, which is what makes it dangerous. A delivery team, wanting to keep a client happy, does more than the contract requires: an extra report, an unbudgeted meeting, a senior person on a call that a junior could have handled, a change absorbed rather than charged. Each act is small, generous, and invisible. In aggregate they are a standing transfer of margin from the provider to the client, made not by a pricing decision but by a thousand un-priced kindnesses.

The mechanism

Over-servicing leaks because effort and revenue are recorded in different systems that never reconcile. The workforce and time systems know what was delivered; the contract and finance systems know what was sold; and nobody holds the two together in real time. So the gap between them, the effort spent beyond the scope that was priced, accumulates silently. It is often rewarded, too: the account is green, the client is delighted, the team is praised. The margin has gone, but everything visible says the account is healthy, which is precisely the failure mode of a dashboard that reads consequences and not decisions.

What it costs

The cost shows up first in utilisation, the share of paid time that is billable. Across professional services, average billable utilisation fell to 66.4% in 2025, down from 68.9% the year before, the lowest in the history of the benchmark; the highest-performing firms sustain around 81.2%.⁷Over-servicing is one of the reasons for that gap: hours are worked, and paid for, that no invoice ever reflects. To make the mechanism concrete (as an illustration, not a measurement), a delivery team that gives away five unbilled hours a week per ten-person pod, at a fully-loaded cost that a rate card was meant to recover, is running that pod several points below the margin the contract assumed, every week, with no event to mark it. The precise figure is specific to each firm; the shape is universal.

There is a second, subtler form of the same leak: the seniority mismatch. Work that a junior could do gets done by a senior, because the senior is closer to the client, cares more, or simply happens to pick up the call. A partner drafting a status report, a principal consultant running a routine review, a team lead handling a query a new agent was hired to handle: each is delivery being done at two or three times its costed rate, invisibly, because the timesheet, if it exists at all, records the hours but not the mismatch between who did the work and who should have. The account still looks healthy, and the client is genuinely better served, which is precisely why nobody stops it. But a business that routinely delivers junior work at senior cost is subsidising its clients out of its own margin, and it has no mechanism that would ever tell it so, because the reconciliation of effort to appropriate rate is a decision no system was asked to make.

The play

Make scope a governed decision, not a goodwill reflex.The point is not to stop delighting clients; it is to decide, deliberately and visibly, when extra effort is an investment worth making and when it is margin quietly walking out of the door. Treat a material request beyond scope as a decision object: name the constraint (the capacity it consumes, which another account needs), attach the evidence (what this client’s account is actually worth, measured, not assumed), and record the choice, to absorb it, to charge it, or to bank it toward the renewal. The value of recording it is that the pattern becomes visible: an account that has quietly absorbed forty unbudgeted hours a quarter for a year is no longer a mystery at renewal time, it is a documented decision the business chose to make. Governed this way, generosity becomes a strategy rather than a leak, spent where it earns loyalty and withheld where it merely subsidises. What was an invisible drain becomes a line the business can see, price, and, when it wants to, stop.

Leak three: attrition and the churn tax

Attrition is usually filed under human resources, as a wellbeing or culture problem. In a people business it is a financial one, and a large one, because every departure resets a ramp the contract already priced as productive. The leak is not the recruiting fee. It is the tax the whole operation pays, continuously, to run below the productivity the pricing assumed.

The mechanism

When an experienced agent or consultant leaves, the loss is not a single cost but a compound one: the open seat produces nothing, the replacement is recruited and trained at a cost, and then the replacement ramps for weeks or months at below full productivity while being paid in full. A contract priced on the assumption of a tenured, productive team is, in a high-attrition environment, quietly delivered by a team that is permanently part-new, part-ramping, and part-covering for gaps. The pricing assumed one thing; the delivery is another; the difference is margin.

The compounding is what makes it a tax rather than a one-off charge. Each departure also removes knowledge that was never written down: the workaround for a fiddly client system, the reason a particular account escalates, the relationship with the client-side manager who actually approves things. That knowledge walks out of the door with the person, and the replacement rebuilds it slowly and partially. Meanwhile the people who stay carry the gap, covering shifts and mentoring the new joiners, which raises their own load and, often, their own likelihood of leaving. Attrition, in other words, is partly self-feeding: the conditions that produce it are worsened by it. A business that reads its attrition number once a quarter and files it under culture is watching a financial process unfold in an HR report, and mistaking the two.

What it costs

The turnover itself is measured, and it is high and rising. Metrigy’s tracking of contact-centre agent turnover found it climbing from 21.8% in 2022 to 28.1% in 2023 to 31.2% in 2024.⁹Deloitte Digital’s contact-centre research puts reported average annual attrition higher still, at around 52%.¹⁰A per-agent replacement cost is where honesty is required. The figure most often quoted, “USD 10,000 to 20,000 to replace a contact-centre agent,” is attributed across the industry to McKinsey and to SQM Group, but it is very hard to trace to a specific primary publication that actually states it, so this playbook does not treat it as a fact. The traceable signals are enough: with a third of agents turning over each year, and reported attrition around half, the ramp cost is not an occasional charge but a permanent operating condition. The precise price per departure is uncertain; the scale of the drain is not.

Attrition is not an HR metric. It is a pricing input the P&L has already agreed to pay, whether or not anyone decided to.

The play

Treat attrition as a leading indicator that decides pricing and staffing, not a lagging one you report on. The move is to stop reading attrition after the fact and start reading it forward. The signals that precede a departure (occupancy held too high, schedule instability, a team whose tenure profile is thinning) are visible before the resignation, and attrition treated as a leading indicator lets the business act while it still can. Govern the decisions that drive it: a staffing plan that names the constraint (a tenure-thin team cannot absorb the occupancy the forecast wants), carries the evidence for its assumptions, and is scored against what actually happened to attrition. And price the churn tax in explicitly: a contract delivered by a team you know will turn over at 30% is not the same contract as one delivered by a stable team, and the pricing decision should say so, as a recorded object rather than an optimistic assumption. Attrition stops being a number the HR report laments and becomes an input the pricing and staffing decisions actually use.

Leak four: bench time and the utilisation drag

The fourth leak is the plainest arithmetic in the whole playbook, and the one most firms have simply made peace with. Every person on the bench, hired, paid, and not assigned to billable work, is margin leaving the building at a fixed rate per day. Some bench is unavoidable, the cost of having capacity ready. But most firms carry far more than they think, and cannot see it, because utilisation is reported as an average that hides where the drag actually is.

The mechanism

The bench leaks because the two halves of the decision, how much capacity to hold and how much work is coming, live in different places and different time horizons. Hiring is planned against an optimistic pipeline; the pipeline slips or reshapes; and the capacity, already hired, sits idle while the forecast that justified it quietly evaporates. Worse, the average hides it: a firm at 70% blended utilisation can be running one team at 90% and burning out while another sits at 50% and drains margin, and the single reported number conceals both problems at once.

There is also a skills dimension the headline number hides completely. A person on the bench is not generic capacity; they are a specific skill at a specific level, and the work waiting to be won needs a different specific skill at a different level. A firm can be carrying idle capacity and turning away deliverable work at the same time, because the people it has are not the people the pipeline needs. This is the bench as a matching problem rather than a volume problem, and it is invisible to any metric that treats headcount as fungible. It is also the join where the sell and the hire have to reason together: the pipeline is, in effect, a forecast of the skills the business will need, and the bench is the standing bet it has already placed on that forecast. When the two are decided in separate systems by separate owners, the bet and the forecast drift apart, and the gap between them is paid in idle salary on one side and premium sourcing on the other.

What it costs

The measured picture is stark. Average billable utilisation across professional services fell to 66.4% in 2025, from 68.9%, the lowest in the survey’s history, while high-performing organisations sustain around 81.2%.⁷ That gap, roughly fifteen points, is the utilisation drag made visible. On a workforce of any size it is enormous: fifteen points of a fully-loaded salary base, earning nothing, every day. The reason the best firms hold 81.2% is not that they work their people harder; it is that they decide capacity against demand deliberately, and the rest do not.

The play

Decide capacity against the pipeline as one governed judgement, not two disconnected plans. The bench is a symptom of a decision made in the wrong place: hiring committed against a forecast that delivery and finance never got to challenge. The play is to make the capacity decision a joined one, where the pipeline’s evidence quality is labelled (a weighted forecast is not a signed contract), the binding constraint is named (the skill you are hiring ahead of, and whether the demand for it is measured or hoped), and the outcome is scored so the next hiring wave is calibrated against how the last one actually landed. This is the heart of workforce planning done as decision intelligence rather than as a spreadsheet: not a headcount number defended in a meeting, but a governed decision that remembers whether the capacity it authorised was ever used. Fixing the bench does not mean hiring less; it means hiring against evidence you can trust and a constraint you have named, so the capacity you carry is capacity you chose.

Leak five: forecast miss and the premium-hiring scramble

The fifth leak is the mirror image of the fourth. Where bench time is the cost of holding too much capacity, the forecast miss is the cost of holding too little, and then buying the shortfall at a premium under pressure. A demand forecast that is wrong in the tight direction forces a scramble: overtime, agency staff, contractors at a spot rate, all paid to cover a gap that better foresight would have staffed at standard cost. The premium is pure leaked margin.

The mechanism

The leak opens because the forecast is treated as a number rather than a decision with evidence behind it. When it misses, the response is reactive and expensive by construction: the cheapest capacity (planned, permanent, trained ahead) is unavailable precisely because the miss was not seen in time, so the business is left with the most expensive capacity (unplanned, temporary, un-ramped). The gap between what the work was priced to cost and what it actually cost to cover is the leak, and it recurs every time the forecast is wrong in that direction.

The asymmetry is the part most planning misses. A forecast that is too high leaves you with the bench of leak four, expensive but visible, sitting on a report where someone will eventually question it. A forecast that is too low leaves you with the premium scramble, which is often invisible, because it is absorbed as overtime, buried in an agency invoice, or paid in the burnout that shows up as attrition two months later. The two errors are not symmetric in how they are seen, so an organisation that fears the visible bench will systematically under-forecast to avoid it, and pay the larger, hidden premium instead. The horizon matters just as much as the number: standard-cost capacity has a lead time, the weeks or months it takes to hire and ramp a skill, and a forecast that is accurate but arrives inside that lead time is, for practical purposes, no forecast at all, because the cheap option has already expired by the time the signal is read.

What it costs

The direct cost is the premium on unplanned labour, which is real but firm-specific and not something this playbook will put a fabricated number on. The indirect cost is more insidious and better evidenced: the scramble is usually absorbed by pushing the existing team harder, and occupancy is the pressure gauge. Occupancy sustained above the mid-eighties looks like efficiency and behaves like burnout; across a large sample of more than 160,000 calculations, the average maximum occupancy target sits around 83.3% for exactly that reason.¹¹ A forecast miss covered by cranking occupancy past that line does not just cost the agency premium; it seeds the attrition of leak three, which resets the ramp, which worsens the next forecast. The leaks compound, which is the recurring lesson of reading these numbers in isolation, as we describe in occupancy and shrinkage, the numbers that mislead.

The play

Govern the forecast as an evidence-weighted decision, and staff to the constraint, not to the hope. A forecast is not a fact; it is a claim, and it should carry the quality of the evidence behind it, so that a number built from measured seasonality is trusted differently from one built on a hopeful pipeline. Name the binding constraint the staffing plan must clear (the ramp time for the skill in question, which sets how early a decision must be made to avoid the premium), and record the staffing choice as an object so that when the forecast misses, the miss is scored and the next forecast is better calibrated. The goal is to convert a recurring, reactive scramble into a governed decision made early enough that standard-cost capacity is still an option. A forecast the business can interrogate and learn from is worth more than a more accurate one it cannot, because the first one gets better every quarter and the second one only gets lucky.

Leak six: discount and pricing leakage

The sixth leak is the fastest-acting of them all, because price is the most sensitive lever on the entire P&L. A discount is decided in a second, often to close a deal or placate a client, and paid for across the whole life of the contract. And discounting is only the visible part. The larger leak is the quiet erosion between the price on the contract and the price the business actually realises, through concessions, waived fees, unbilled overruns and off-list terms that nobody added up.

The mechanism

The classic name for this is the pocket price waterfall: the series of deductions, some contractual, many discretionary, that separate the invoice price from the money the business actually keeps. Each deduction is a small decision, made by a different person, at a different moment, none of them recorded as a pricing choice. There is no single culprit and no single event, only a waterfall of un-tracked concessions, which is why pricing leakage is so hard to see and so easy to keep doing.

The same body of work names a second pattern worth watching: the pocket price band, the spread of realised prices across customers for essentially the same work. In most businesses that band is wide, and wider than anyone believes, because the concessions are granted account by account with no view of the whole. Two clients buying the same service can be paying materially different realised rates, not through any deliberate segmentation but through the accumulated history of who negotiated hardest and who happened to be handled by whom. For a people business the band is aggravated by time: multi-year contracts, particularly rate-capped ones, cannot reprice as fast as the cost of the labour delivering them rises, so a rate that was healthy at signature erodes silently across the term. The leak is not only the discount granted today; it is the discount granted three years ago that the contract will not let you take back, quietly widening the gap between what the work costs and what it earns.

What it costs

The cost is disproportionate to its cause, and this is the crucial point. In the foundational study of the subject, Marn and Rosiello showed that for the average company a 1% improvement in realised price raises operating profit by 11.1%, a far larger effect than a 1% gain in variable cost (7.8%), volume (3.3%) or fixed cost (2.3%).⁸ Price flows straight to the bottom line with no offsetting cost, which cuts both ways: the same sensitivity that makes a small price improvement so valuable makes a small, un-noticed leak enormously expensive. A people business that lets one point of realised price erode across its portfolio through discretionary discounting is not losing one point of revenue; it is losing something closer to eleven points of operating profit, decided a fraction at a time by people who never saw the cumulative effect.

The discount is decided in a second and paid for across the life of the contract, and price is the most sensitive lever there is.⁸

The play

Make every discount and concession a governed decision with a visible waterfall. The leak closes the moment the pocket price waterfall is made visible and each deduction becomes a recorded choice rather than an invisible reflex. Name the constraint a discount is meant to clear (is this concession the thing that actually wins the deal, or a habit?), attach the evidence (what this client is worth, and what the last three discounts of this shape actually bought), and record the choice so the pattern can be seen across the portfolio. Confidence in a pricing decision should be composed from that evidence, not asserted by whoever is most eager to close, which is what business confidence means in practice. Governed this way, discounting becomes a deliberate instrument used where it earns its keep, rather than a slow leak nobody chose. Given how sharply profit responds to price, this is the single highest-return place in the playbook to impose discipline.

Leak seven: unbilled work and scope creep

The seventh leak is close cousin to over-servicing, but it is worth separating, because its cause is different. Over-servicing is generosity; scope creep is drift. Work expands past the boundary of what was sold, one reasonable request at a time, and because no single request is large enough to renegotiate over, the scope grows while the price stays fixed. The client is not exploiting anyone; the boundary simply moved, and nobody was watching it.

The mechanism

Scope creep leaks because the contract is a document filed at signature and the work is a living thing that changes every week, and the two are never reconciled after the ink dries. The statement of work said one thing; the actual work has quietly become another; and the gap between them is unbilled labour that the original price never contemplated. It is a governance vacuum: there is a moment, each time the scope shifts, where a decision could be made, to absorb it, to charge it, to renegotiate, and that moment passes unrecorded, again and again, until the accumulated drift is large and no longer attributable to any single change.

The deeper problem is ownership: nobody is standing at the boundary. The delivery team is measured on client satisfaction and delivery, not on defending scope, and often has no easy way to raise a change without appearing difficult. The commercial team that owns the contract is not in the daily flow of the work and does not see the small requests as they land. So the boundary is everyone’s responsibility and therefore no one’s, and it moves by default rather than by decision. This is the structural difference between a fixed-price engagement and a time-and-materials one: the second bills the drift automatically, so it cannot hide, while the first absorbs it silently unless someone actively stops it. A firm that runs fixed-price work without a live mechanism for governing scope has, in effect, written its clients an open option to expand the work at no extra cost, and options written for free are exercised.

What it costs

An honest note on the numbers here. A figure is often cited that companies lose “1% to 5% of realised EBITDA” to revenue leakage, attributed to EY, but like the agent-replacement figure it is repeated secondhand across many sources and is hard to trace to a primary publication, so this playbook does not rest a claim on it. What can be said with evidence is that unbilled work depresses realised utilisation (the measured 66.4% average already carries some of this drag⁷) and that, given how sharply profit responds to price, revenue that is delivered but never billed is close to the most expensive revenue to lose, because its cost has already been fully incurred. To make it concrete without dressing an assumption as a measurement: a project that has silently drifted ten per cent beyond its scoped hours, at a fixed price, has surrendered ten per cent of its margin base, and the drift will not appear in any report until the project’s profitability is reconciled at the end, long after the decisions that caused it could have been changed. That figure is illustrative, offered to show the shape, not to report a specific outcome.

The play

Put a decision at every scope boundary, so drift becomes a choice instead of an accident. The play is not to police clients; it is to make the boundary visible and to force a small, recorded decision each time it moves. Treat a material scope change as an object: name the constraint it consumes (the hours it adds against the fixed price it does not), attach the evidence (what has already drifted on this account, measured against the statement of work), and record the choice, so that at reconciliation the profitability is not a surprise but the sum of decisions the business can see and, next time, make differently. The value is cumulative: an account whose every scope shift is a recorded decision cannot drift silently, because there is no longer anywhere silent for it to drift to. Scope creep is a governance failure before it is a commercial one, and governing the boundary is what turns invisible drift into a line the business controls.

Leak eight: weak renewals and contract drift

The eighth leak is where several of the earlier ones come home to be paid. A renewal is the moment a people business gets to reset the terms of an account, to reprice for the cost that has crept in, the scope that has drifted, the attrition that has eroded productivity. It is the single best opportunity to recover leaked margin. And it is routinely wasted, because the renewal is treated as a date that arrives rather than a decision that is made.

The mechanism

The leak opens because, at the renewal, the business cannot see what it needs to decide well. On the surface the account is green: every service level is met, the client is satisfied, the dashboard is calm. But the value actually delivered has often never been measured, the true cost to serve has crept up through the leaks above, and the relationship may run through a single sponsor. So the renewal is signed at the old rate, or with a token uplift, because there is no assembled evidence to justify more, and the accumulated erosion is locked in for another term. A renewal made on a green dashboard is a renewal made blind to everything that actually moved the margin.

Two hidden risks make the blindness worse. The first is the single sponsor: an account that runs through one relationship is one departure away from a cold renewal, and the provider that has never measured or communicated the value it delivered has no defence when the friendly sponsor is replaced by a procurement-led review that starts from price. The second is the absence of a baseline. Because the value delivered was never captured as it accrued, the renewal conversation has nothing to point at: no record of the outcomes achieved, the problems averted, the scope absorbed as goodwill. The provider is reduced to arguing for a rate increase on trust, at exactly the moment the client is most inclined to test it. A renewal is won or lost in the eighteen months before it, in whether the account kept a running, evidenced account of the value it created, and most do not, which is why so many renewals default to the path of least resistance: the same rate, one more year, a little more margin gone.

What it costs

The cost is the compounding of every prior leak, extended for another contract term, but the deeper evidence is about decision quality itself. The research on organisational decision making is unambiguous that this is where value is won or lost. McKinsey finds that only about 20% of organisations believe they are good at decision making and that 61% say most of the time they spend on it is wasted.⁴ The same body of work finds that speed and quality go together rather than trading off: respondents who called their decisions fast were nearly twice as likely to call them high quality, and organisations that combined both were about twice as likely to report top-quartile financial returns.⁵Bain’s decade-long study across a thousand companies is blunter still, finding a correlation above 95% between how well a company makes and executes decisions and its financial performance, with the best decision makers delivering around six percentage points more total shareholder return.⁶ A renewal is one of the highest-stakes decisions a people business makes, and most are made in exactly the un-evidenced, un-recorded way this research shows is most expensive.

The play

Make the renewal a governed decision that starts long before the date, not a signature that confirms the status quo. The play is to assemble the renewal as a decision object across the life of the account, not to improvise it at the deadline. Carry the evidence of delivered value and true cost with its quality labelled, so the negotiation rests on measured facts rather than on the comfortable green of a service dashboard. Name the constraint the renewal must clear (the rate below which the account no longer earns its capacity) and record the choice, to reprice, to renegotiate scope, or to walk, so the outcome can be scored against what the account actually did next. What happens after a deal is won is not an epilogue; it is where the margin is defended or surrendered, which is why we treat what happens after a deal is won as a first-class subject. A renewal governed this way stops being a date that arrives and becomes the moment the business recovers what the year quietly leaked.

Leak nine: the AI and labour-arbitrage squeeze

The ninth leak is different in kind from the eight before it. Those are operational: money escaping through decisions inside the business. This one is structural: the ground under the business model itself shifting, as artificial intelligence rewrites the economics that made labour arbitrage rational in the first place. It is the slowest-moving leak and the most dangerous, because it does not just erode a margin, it can erode the reason the margin existed.

The mechanism

The offshore labour-arbitrage model, moving standardised, well-defined, heavily-monitored work to a lower-cost location, was rational precisely because the work was standardised, well-defined and monitorable. Those are exactly the properties that make a task automatable. As the Harvard Business Review put it, AI is rewriting the economics of outsourcing by automating the very characteristics that made arbitrage worthwhile.¹⁴A provider whose proposition is “the same work, cheaper, somewhere else” is selling the thing automation does best, and the client increasingly knows it, which is why the pricing conversation of the opening section has an edge to it: buyers now assume AI-driven efficiency and expect the saving passed through.

The likely outcome is a two-speed industry, and which speed a provider ends up on is itself a decision being made now, mostly by default. Providers that treat AI as a way to do the same commodity work for less will find themselves in a race whose finish line is a machine, competing on the exact dimension the technology is built to win. Providers that use the productivity it releases to move up, from executing tasks to governing outcomes, from selling hours to selling judgement a client can trust and audit, will find the ground under them firmer, because that is the part of the work automation does not replace and buyers cannot easily insource. The distinction is not about how much AI a provider adopts; it is about what the provider is for. A firm that has never been able to show a client the reasoning behind its work has nothing to move up to, which is why the ninth leak is, at bottom, the same problem as the other eight: a business that does not govern its decisions cannot sell the one thing that will still be worth paying for.

What it costs

The trajectory is measurable, even if the timing is contested. Gartner forecasts that by 2029 agentic AI will autonomously resolve 80% of common customer-service issues without human intervention, cutting operational costs by around 30%.¹² If even directionally right, that reshapes the unit economics of contact-centre outsourcing entirely, and a provider still priced on the old head-count model is exposed on both sides at once: undercut on price, and undercut on the very labour it sells. But the same analysts counsel against believing the timeline uncritically. Gartner also predicts that more than 40% of agentic-AI projects will be cancelled by the end of 2027, undone by escalating cost, unclear value and inadequate controls.¹³ The lesson is not that AI is overhyped; it is that AI deployed without judgement fails expensively, which is its own margin leak, an investment that does not return.

The play

Move up the value chain from doing the work to governing it, and adopt AI only where its evidence is exposed. The structural answer is to sell what automation cannot: judgement, governance, and outcomes rather than hours. A provider that can show a client not just that work was done but that the decisions behind it were sound, evidenced and defensible is selling something an arbitrage competitor cannot, and something AI on its own does not provide. On adoption, the discipline is the same one that runs through this whole playbook: an AI that produces a confident recommendation you cannot trace is instinct with a bigger vocabulary, and deploying it in a consequential decision is how the 40% of projects fail. Insist that any AI in a pricing, staffing or renewal decision exposes the evidence beneath it and records the human choice on top of it, so the recommendation can be interrogated rather than merely trusted. The shift from doing work to governing decisions is explored in our companion report, The State of Decision Intelligence for People Businesses 2026; the point for this playbook is that the ninth leak is the one that turns the other eight from an operational nuisance into an existential question, and the answer to it is the same discipline that plugs the rest.

The common cause, and the summary table

Read the nine leaks together and a single cause comes into focus. None of them is really a pricing problem, or a hiring problem, or an AI problem, though each wears one of those costumes. Every one of them is a decision that was made without the evidence it needed, recorded nowhere, and therefore never learned from. The mis-scoped deal, the un-priced kindness, the reactive hire, the reflexive discount, the silent scope drift, the renewal signed on a green dashboard: each is a moment where a judgement was made and immediately discarded, so that the same judgement, made badly, could repeat next week without anyone noticing it had happened before.

You cannot govern what you never wrote down, and you cannot stop a leak you cannot see.

This is why the plays all share one shape. In each case the move is to take a decision that was made in a meeting or a reflex and turn it into a governed object: name the constraint that actually binds it, attach the evidence with its quality honestly labelled, record the human choice, and score the outcome against what was expected. The mechanics differ (a pricing waterfall, a staffing plan, a renewal file) but the discipline is identical, and it is the discipline of decision intelligence. A decision captured this way stops being a leak and becomes an asset: searchable, auditable, and, above all, learnable, so that the next decision of the same shape is better informed than the last.

The table is a diagnostic, not a scorecard. Most people businesses are leaking through several of these at once, and the reason the leaks are so persistent is not that they are hard to understand, it is that they are invisible in the systems the business already runs. The CRM, the workforce tool and the ledger each hold a slice of the truth; none of them holds the decision that spans them, which is exactly the decision each leak turns on. That missing layer is the whole opportunity. It is why the fix is not a tenth system but a place for the thing the other systems were never built to keep, and why a leak the business could never see becomes, once decisions are governed, a line it can finally control. The record that layer leaves behind, evidence, options, choice and outcome held together, is what we call a Decision Object, made in a Decision Room, and it is the reason margin, once it leaks through decisions no system recorded, can be recovered by making the decisions governable.

The final reason to govern decisions rather than merely plug leaks is that governance compounds where firefighting does not. A cost programme recovers margin once and then decays; the leaks reopen because nothing changed about how the decisions that cause them are made. A governed decision, by contrast, leaves a record, and a thousand records of the same shape become something a competitor cannot buy: the firm’s own history of what worked. If deals with a certain profile consistently deliver below forecast, the next one of that shape is flagged before it is signed. If discounts of a certain size never change the win, they stop being granted. If a staffing pattern reliably precedes an attrition spike, it is caught while it can still be changed. The leaks do not just close; they teach. That is the difference between fixing a margin problem and dissolving it, and it is why the destination is not tighter cost control but an organisation that gets measurably better at the decisions its margin is made of, quarter after quarter. This is the loop that turns a pile of decisions into enterprise intelligence, and it is explored across our companion note on why BPO margins keep shrinking.

Methodology & a note on honesty

A playbook about margin that leaks through un-recorded decisions has no business making un-recorded claims of its own. A note, therefore, on how this one was assembled and where it deliberately stops short.

Every quantitative figure here is drawn from a named, dated, public source, analyst firms, peer-reviewed and long-established research, and public-company filings, and each is listed in the References below. The macro margin picture rests on the disclosures of Concentrix¹ and Teleperformance² and on Deloitte’s outsourcing survey³; the decision-quality evidence on McKinsey⁴⁵ and Bain⁶; the operational figures on SPI Research’s utilisation benchmark⁷, Metrigy⁹ and Deloitte Digital¹⁰on attrition, and Call Centre Helper¹¹ on occupancy; the pricing-sensitivity point on the foundational Marn and Rosiello study⁸; and the structural AI argument on Gartner’s two forecasts¹²¹³and the Harvard Business Review’s analysis of outsourcing economics¹⁴, with proposal win rates from Loopio¹⁵.

A playbook about traceable decisions has no business making untraceable claims of its own.

Two widely-quoted figures were deliberately notused as load-bearing facts, because they could not be traced to a primary source. The first is the frequently-cited “USD 10,000 to 20,000 to replace a contact-centre agent,” attributed across the industry to McKinsey and to SQM Group but very hard to pin to an original publication that states it; the playbook names the caution in leak three and rests the point instead on measured turnover⁹ and attrition¹⁰. The second is the “1% to 5% of realised EBITDA lost to revenue leakage” figure attributed to EY, which is repeated secondhand in many places but similarly hard to trace to a primary EY publication; leak seven names it as such rather than relying on it. Naming these is the point, not a footnote to it.

Where the playbook illustrates a mechanism with a worked example, the five-unbilled-hours pod in leak two, the ten-per-cent scope drift in leak seven, those are explicitly labelled as illustrative, offered to show the shape of a leak, not to report a measured outcome or a specific customer. This playbook invents no customers, logos, quotes or results. It holds itself to the same evidence hierarchy it recommends: measured facts are cited, modelled or illustrative content is named as such, and an assumption is never dressed up as a measurement.

The nine leaks are a framework, a synthesis of the patterns visible across BPOs, consultancies, managed service providers and staffing firms, not a ranked benchmark; treat the ordering as a flow of money through a business, not a league table of cost. The playbook will be revised as the numbers move, particularly the AI-adoption and margin figures, which are changing quickly. If you find a claim here you believe is wrong, or a source that has been superseded, we want to know. A reference work earns its standing by being corrected, not by being defended.

References

Every figure in this report is sourced below. Third-party links open in a new tab. Where a figure is modelled rather than measured, or where a widely-quoted statistic could not be traced to its primary source, the report says so in the text.

1. Fourth Quarter and Fiscal Year 2025 Results · Concentrix Corporation, 2026. Non-GAAP operating margin 12.8% (FY2025), down from 13.7% (FY2024).
2. 2024 Annual Results · Teleperformance, 2025. Recurring EBITA margin 15.0% (FY2024); 2025 guidance implied broadly flat margins.
3. 2024 Global Outsourcing Survey · Deloitte, 2024. Cost reduction remains the primary outsourcing driver for more than 70% of organisations; buyers increasingly evaluate fully-loaded cost.
4. Decision making in the age of urgency · McKinsey & Company, 2019. Only 20% of organisations say they excel at decision making; 61% say most decision-making time is used ineffectively; inefficiency costs a typical Fortune 500 around 530,000 manager-days (~USD 250m) a year.
5. Three keys to faster, better decisions · McKinsey & Company, 2019. Respondents who called their decisions fast were 1.98x more likely to also call them high quality; winning organisations were about twice as likely to report top-quartile financial returns from their decisions.
6. Measuring decision effectiveness · Bain & Company, 2020. A 95%+ correlation between decision effectiveness and financial performance; top-quintile decision firms averaged around 6 percentage points higher total shareholder return.
7. Professional Services Maturity Benchmark 2025 · SPI Research, 2025. Average billable utilisation fell to 66.4% (2025) from 68.9% (2024), the lowest in the survey history; high-performing firms reach around 81.2%.
8. Managing Price, Gaining Profit · Harvard Business Review (Marn & Rosiello), 1992. For the average company a 1% improvement in realised price raises operating profit by 11.1%, a larger effect than a 1% gain in variable cost (7.8%), volume (3.3%) or fixed cost (2.3%); introduces the pocket price waterfall and pocket price band.
9. What Metrigy’s Latest AI Data Reveals About Contact Center Staffing · Metrigy, 2024. Contact-centre agent turnover rose from 21.8% (2022) to 28.1% (2023) to 31.2% (2024).
10. Global Contact Center Survey · Deloitte Digital, 2023. Reported average annual agent attrition around 52%; AI-centric contact centres reported being 85% more profitable.
11. What Is the Ideal Occupancy Rate for a Contact Centre? · Call Centre Helper, 2023. An analysis of more than 160,000 calculations put the average maximum occupancy target at 83.3%.
12. Gartner Predicts Agentic AI Will Autonomously Resolve 80% of Common Customer Service Issues by 2029 · Gartner, 2025. Forecast issued March 2025, with a projected 30% reduction in operational costs.
13. Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 · Gartner, 2025. Cancellations driven by escalating costs, unclear business value and inadequate risk controls.
14. AI Is Rewriting the Economics of Outsourcing · Harvard Business Review, 2026. Generative AI automates precisely the standardised, monitorable tasks that made offshore labour arbitrage rational.
15. RFP Response Trends and Benchmarks · Loopio, 2025. Average RFP win rates cluster around 43 to 45%; general B2B win rates are lower and vary widely by deal size.