Process as a Pillar, Not a Support Function

Two quotes left the same eleven-person firm in the same week, for nearly the same job. One came in at £14,000 with installation included. The other came in at £11,500 with installation listed separately, to be confirmed. Neither person did anything wrong. Each had worked out a sensible way of pricing that kind of job, and the two ways were different. The client had asked both. That is how the firm found out.

Why business processes are important, and why they get ranked last

Business processes matter because they are the medium a product reaches the customer through, not a support function beside it. Product and marketing decide what gets promised. Process decides whether that promise survives repetition, volume, and staff who were not in the room when it was made.

Almost nobody denies that process matters. What happens instead is ranking. Product first, because without it there is nothing to sell. Marketing second, because without it nobody knows. Then sales, then hiring, then finance, and somewhere below that, in the part of the list where things go to wait, sit systems, process and documentation. Important. Not urgent. Something to do properly once the urgent things are handled.

The ranking is rational, which is why it survives

Product and marketing have short, visible feedback loops. A price change moves conversion inside a week. A new feature produces a reaction in the first ten calls. The signal arrives fast enough to be attached to the decision that caused it.

Process has the opposite shape. The investment lands now and the result arrives months later, mixed in with everything else. Nelson Repenning and John Sterman documented what that delay does. While the improvement is still invisible, managers blame the people doing the work rather than the system they are working inside, and the control measures that follow suppress the very capacity that would have produced the improvement (Administrative Science Quarterly, 2002). The delay is not a minor inconvenience. It actively generates the wrong conclusion.

Operations also has no natural owner, as Michael Hammer observed in 2004, because nobody holds the title Vice President of Operational Innovation. Which is why companies formalise when something outside forces them to, rather than when the process is ready: across 78 early-stage companies, the rate of adoption of management systems tracked headcount, investor presence and time to revenue (Davila and Foster, The Accounting Review, 2007).

So the ranking is not laziness. It is what a reasonable person concludes from the evidence in front of them. The problem is that deprioritising process does not postpone the cost. It converts it into two forms that are slow, hard to see on a monthly dashboard, and expensive to reverse.

Are processes more important than product? The question is built wrong

A product is not a thing that sits beside the processes that deliver it. A product is a set of processes that held together long enough to produce something a customer paid for. Ranking process below product treats the container as optional and the contents as the whole job.

The reduction underneath this is simple. A business is three things: people or machines, data, and processes. People process data by following processes to deliver value to a customer. That model is developed at length in the companion piece on business process improvement and who it is actually for, and this article takes it as a premise rather than a subject.

Quality stops being a property of the company

With no agreed path, competent people improvise competently. That is the difficult part. Nobody is underperforming. Everyone is solving the problem in front of them with the judgement they have, and the judgements differ.

The size of that difference is larger than managers expect. In a noise audit at an insurance company, professionals independently priced identical cases: the median difference between two underwriters on the same policy was 55 per cent, and between two claims adjusters on identical claims, 43 per cent. Asked beforehand what variation they would expect among experts doing the same job, 828 senior executives gave a median answer of 10 per cent (Kahneman, Sibony and Sunstein, Noise, 2021). That is an account of a client engagement rather than a published experimental study, so it carries a different evidential weight. The gap between 55 and 10 is still worth sitting with.

Variation is not a softer version of a low average. It is a separate quantity with its own consequences. Walter Shewhart built statistical process control in 1931 around predictability of variation rather than defect counts, and J. F. C. Kingman's 1961 result puts variability into the waiting-time formula as a straight multiplier: hold average demand and service time constant, raise the variability, and the queue grows anyway.

The business version has been measured. Studying eleven processes common to retail banking, Frances Frei and colleagues found banks performing well on one process frequently performed badly on another, and that low process variation contributed significantly to financial performance, possibly more than the average level of process performance did (Management Science, 1999).

Customers experience the variance, not the average, which is why internal quality metrics can hold steady while customer perception slides. Tracking 287 patients through two painful procedures, Donald Redelmeier and Daniel Kahneman found retrospective judgement followed the worst moment and the final moments, with duration barely registering (Pain, 1996). A client who has had six good months and one bad handover is carrying the handover.

Two qualifications, both honest. Consistency gets a company to acceptable, not to admired: the researchers who established reliability as the dimension customers weight most heavily in meeting expectations also found that responsiveness, assurance and empathy are where firms exceed them (Sloan Management Review, 1991). And reducing variability has a customer cost, since constraining the options can drive customers away (Frei, Harvard Business Review, 2006). Standardisation moves cost from the customer's ledger to the company's. Usually a good trade. Still a trade.

The knowledge that leaves is the knowledge nobody wrote down

Undocumented process lives in people, and people leave. In July 2026 alone there were 5.1 million separations from US payrolls, a rate of 3.2 per cent, and across 2025 there were 62.8 million (Bureau of Labor Statistics). Those count events rather than individuals, but the shape holds: this is a rate, not an event. US median tenure with a current employer was 3.9 years in January 2024, the lowest since 2002.

Europe looks different, and both things are true at once. Across six Western European countries, mean job tenure was flat between 1993 and 2021, with roughly half of workers at the same employer ten years or more (Goulart and Oesch, European Journal of Industrial Relations, 2024). Churn is not spread evenly. It concentrates in a minority of short-tenure roles, which is where a small company's frontline work usually sits.

What a departure costs has been measured properly, and the real figure is higher than the one that circulates. Using German social security records covering around 34,000 unexpected worker deaths at small firms, Simon Jäger and Jörg Heining used a model to estimate marginal replacement costs of roughly 2.3 to 3.0 annual salaries of an incumbent worker, well above standard estimates drawn from employer surveys (NBER, 2022). That is an estimate for their sample, not a universal replacement-cost multiplier.

Hiring a replacement of equal experience does not restore what left. Cardiac surgeons working across several hospitals improved with recent volume at that specific hospital and not with volume elsewhere: performance was partly a property of the organisation, and it did not travel (Huckman and Pisano, Management Science, 2006).

It also reaches the customer. Linking factory staffing records to field failures across roughly 50 million mobile devices, Ken Moon and colleagues found that devices built in the high-turnover weeks after paydays failed 10.2 per cent more often than those built in the calmest weeks (Management Science, 2022). Turnover this week becomes a broken product in somebody's hand next year. And a meta-analysis of 110 samples covering over 309,000 organisations and units found the damage moderated by size: the same turnover rate hurts a smaller unit more (Park and Shaw, Journal of Applied Psychology, 2013).

The two mechanisms feed each other, and this is the part that compounds. Quality drift makes a workaround necessary. The workaround is never written down, because it was a fix rather than a decision. It becomes the tacit knowledge of whoever invented it. Then it leaves. Anita Tucker and Amy Edmondson observed 26 nurses across nine hospitals for 239 hours and recorded 194 failures. In only 7 per cent did anyone act on the underlying cause. The rest were worked around, and the system was left as it was (California Management Review, 2003).

Does writing it down actually help?

Partly, and less than the intuition suggests. Zeynep Ton and Robert Huckman analysed 48 months of turnover data across a large US retail chain and found that in stores with high process conformance, rising turnover had no negative effect on performance at all, while in low-conformance stores the damage was pronounced (Organization Science, 2008). The authors scope that to work which depends on repeating known tasks. Against it, Moon and colleagues studied 52,214 workers across 44 assembly lines in an environment about as standardised as a workplace gets, and turnover degraded output anyway: what had gone was workgroup coordination knowledge, which they describe as difficult to codify (Management Science, 2023). A 2023 review of 91 empirical studies confirms documentation has barely been tested as a protective mechanism at all (Galan, The Learning Organization).

The defensible reading is that written process protects the codifiable part and leaves the coordination part exposed. That is a great deal of the work and worth having. It is not the whole of it, and anyone claiming otherwise is selling something. It is also why documents fail on their own.

Building clean costs less than optimising later

The common sequence is to build the process with gaps, ship, and optimise when there is time. That sequence is more expensive than it looks, and the reason is structural rather than motivational. Each gap gets filled with a workaround. Each workaround becomes a dependency for the next one. Optimising later is not tidying. It is untangling dependencies that were created deliberately, usually without a record of why.

Software engineering is the one field that has measured this, so the evidence is borrowed. Ward Cunningham coined the debt metaphor in 1992, and his original is more precise than the version in circulation: a little debt speeds development as long as it is repaid promptly, and the danger lies only in not repaying. Alan MacCormack and colleagues quantified how far a single change ripples through a system and found Mozilla carrying a propagation cost of 17.35 per cent before a deliberate redesign and 2.78 per cent after it, on a codebase of comparable size (Management Science, 2006). The earlier design had more than six times the potential change propagation. That is a measure of dependencies, not evidence of a sixfold difference in money spent. Separately, developers reporting twice weekly for seven weeks said existing debt consumed 23 per cent of their time, and that in 24 per cent of cases the debt already there forced them to add more (Besker and colleagues, 2019). Self-reported, so the compounding mechanism is the finding rather than the figure.

Why do SOPs fail? Usually because the standard is not the one leadership holds

Process is not a virtue on its own. A process disconnected from strategy is bureaucracy with better formatting. Documentation nobody reads is filing. Two failure modes are worth naming exactly.

The first is a company documenting a standard its leadership will not enforce. This is worse than not writing it down, because the document makes the gap between stated and actual practice visible and then tolerated. Nothing teaches a team that standards are optional faster than a standard nobody is held to. John Meyer and Brian Rowan described the general case in 1977: organisations adopt formal structures for legitimacy, then decouple them from the actual work and replace inspection with what they called a logic of confidence and good faith. Olivier Boiral found employees performing the quality system as ritual inside certified firms (Organization Science, 2003), and John Gray and colleagues, using FDA plant inspections, found process compliance deteriorating steadily after certification (Production and Operations Management, 2015). Diane Vaughan's name for where this ends is normalisation of deviance: repeated unchallenged departures stop being seen as departures. In every one of these cases, the document was never the system.

The evidence on certification is genuinely mixed and worth reporting as mixed. David Levine and Michael Toffel matched 471 pairs of California firms and found ISO 9001 adopters growing faster, with the benefit concentrated in smaller firms (Management Science, 2010). But across a panel of 19,713 US facilities, Ann Terlaak and Andrew King found certified plants growing faster while operational improvement did not account for it: the value was in the signal to buyers, not the system (Journal of Economic Behavior & Organization, 2006). Installing a standard and using it are different things, and only the second produces operating advantage (Naveh and Marcus, Journal of Operations Management, 2005).

The second failure mode is optimising a process that should not exist. Between 1987 and 1990 Analog Devices ran a quality programme that worked on every operational measure: defects in shipped product fell from 500 to 50 parts per million, on-time delivery rose from 70 to 96 per cent, yield climbed from 26 to 51 per cent, and cycle time halved from 15 weeks to 8. Over the same period the share price fell from $24 to $6, operating income fell from $46.6m to $6.2m, and the company made the first layoff in its history. John Sterman and colleagues traced the mechanism: manufacturing improved far faster than the demand side, capacity went idle, cost-plus pricing stopped working, and the improvement was never coupled to anything the market wanted (Management Science, 1997). Operational effectiveness is not strategy, as Michael Porter put it in 1996, and competing on it alone converges firms and destroys margin.

So the argument is not document everything. It is that process is the mechanism by which strategy, leadership and values reach the customer, and a company treating that mechanism as secondary has decided its strategy will be delivered by whoever happens to be on shift.

What that looks like in practice is ordinary. A client had bought a CRM and had no system around it, so barely anyone used it. The workflow was designed first, then the automations built behind it: assignments, status triggers, escalations, reporting views. Then it was documented and the team trained, so it held after handover. The order of those steps is the whole point.

When is it too early to document a process?

An article arguing that process is a pillar, which refuses to say when process work is a waste of money, is a sales document. So, plainly.

A process still being discovered should not be frozen. Before a company knows what it is selling and to whom, what would get documented is a hypothesis, and documenting a hypothesis gives it authority it has not earned. Maurizio Zollo and Sidney Winter put the risk precisely: codification attempted prematurely risks hasty generalisation from limited experience, with inflexibility and negative transfer of learning behind it (Organization Science, 2002). James March's simulation work points the same way: a code everyone converges on quickly has no deviants left to learn from, and ends up holding less knowledge (Organization Science, 1991). Reproducibility is also what makes an organisation hard to change (Hannan and Freeman, 1984), and reversing a committed model is expensive in practice. Tracking 154 young Silicon Valley firms, Hannan and colleagues found that changing the founding organisational blueprint roughly tripled the net hazard of failure, while swapping the founding CEO for an outsider had no significant effect (Industrial and Corporate Change, 2006). Changing the model cost more than changing the leader.

Now the part that cuts the other way, because it does. Wesley Sine and colleagues tested that advice in the most turbulent sector available, new ventures in the emerging Internet sector, and found higher founding-team formalisation, specialisation and administrative intensity improved performance (Academy of Management Journal, 2006). And a meta-analysis of the planning literature concludes that planning is beneficial on average, with firm newness moderating the effect (Brinckmann and colleagues, Journal of Business Venturing, 2010).

Which leaves an honest position rather than a tidy one. There is no headcount threshold in the evidence. Not five people, not fifteen, not fifty. Anyone naming one has invented it. What the evidence gives instead is a test about the work rather than the company: a process worth writing down is one that has been performed enough times to know which parts are stable, on work the company intends to keep doing. Everything else is a hypothesis, and hypotheses are cheaper held loosely.

The firm with the two different quotes did not have a documentation problem. It had never decided how that job should be priced. The document would only have been the record of the decision.

Back to blog