Book review: The Machine, by Justin Roff-Marsh

1. Use the book for routing, handoffs, scheduling and measurement. That part is right, and most of it is obvious.
2. Reject its theory of people. Incentives trump processes, and talent is heavy-tailed. Fix the rewards, don't remove them.
3. The caveat. Incentives get gamed, the performance skew has a measurement critique, and one point I agreed with has evidence against it. Section 09.
4. The alternative. Three elements: idiot-proof structure, a performance culture, and incentives on a short cycle. Section 12.

The route: the book's prescriptions and my verdicts, the frame, incentives, talent, territories, the counterevidence, the ledger, and my own framework.

The Machine: A Radical Approach to the Design of the Sales Function was published in 2015 by Justin Roff-Marsh, founder of the sales consultancy Ballistix. It applies Goldratt's Theory of Constraints to sales. The model: marketing creates opportunities, a coordinator fills the salesperson's calendar as if it were machine capacity, the salesperson only sells, engineers design the solution, operations owns repeat business, management controls the flow. Individual commissions are removed.

I scored each of the book's 24 points. The table compresses them to ten.

scroll the table sideways →

The third column is the argument of this review. Every point about routing, scheduling and measurement gets an agree or an obvious. Every point about people gets an incomplete or a disagree.

The book's strongest material is division of labour applied to the salesperson. Salesforce's State of Sales puts selling at 28% of a rep's week, down from 34% in 2018. The figure is self-reported and matches what I have seen on my own teams.

The fix is the one the book proposes: take administration off the reps, split prospecting from closing, define stages and handoffs. Firms with a formal sales process reported 18% higher revenue growth in a Vantage Point Performance survey of 62 B2B companies. All of that is true, and obvious.

Three qualifications from my own teams. Small teams lack the scale: below a handful of reps, an SDR, an AE and an account manager are three salaries doing one job. Information has to flow both ways: a good AE knows which accounts and which messages will land. And the book does not separate inbound from outbound: a central calendar is right for inbound; in outbound, sequencing, timing and account strategy belong to the rep.

Rittel and Webber coined the term in 1973. A tame problem can be stated fully, has a stopping rule, and its solutions are true or false. A wicked problem cannot be formulated without committing to a solution, never ends, and its solutions are good or bad depending on who judges. Manufacturing is tame, and the Theory of Constraints was built for it. Sales has humans at both ends.

The book's answer is to declare the problem tame. Reps become capacity, conversations become throughput, deviations become scheduling errors. My reading is that the author manages a wicked problem as if it were tame. That is a reading of the text, not a diagnosis of the author. Ashby's law of requisite variety states the limit: a controller needs at least as many responses as the system has states, and four principles is a small set.

Point 18 says salespeople should not receive individual commissions, because the system produces the revenue. Point 19 says commission creates dysfunctional ownership of leads and accounts. Point 19 is true. Point 18 does not follow from it.

Incentives trump processes. I have seen this over and over in my career. I have run sales teams of 3 to 80 reps in marketplaces and SaaS, in Asia and Europe, and the pattern has held in each of them. One operator's sample.

Two results are enough. Charlie Munger's Xerox case: salespeople pushed an inferior machine because the commission plan paid more for it, and Munger's wider point is that people who understand incentives still underestimate them. Lazear's Safelite study: when the company moved from hourly pay to piece rates, output per worker rose 44%, about half of it from more productive workers joining and staying.

The book's point 19 has a number too. Larkin found that accelerating commissions at an enterprise software vendor cost 6% to 8% of revenue, because reps timed deals and discounted to hit the accelerator. That is a case for a linear plan, not for no plan.

Selling is against human nature. Rejection, repetition, prospecting, follow-up and asking for commitment are activities people avoid unless something realigns them. Compensation does that. The problem is badly designed incentives, not incentives themselves.

Point 24 is the book's goal: replace dependence on rainmakers with a predictable production system. My working heuristic is that 20% to 30% of salespeople generate 70% to 80% of results. No study tests those percentages, so they stay an estimate. The shape behind them is documented.

O'Boyle and Aguinis tested normal against Paretian fits on 198 samples covering 633,263 people and found the power law fit better. Aguinis and colleagues repeated the test in 2016 on about 625,000 people, including sales occupations, with the same result. Quota data from the field agrees.

If a standardized process made output homogeneous, attainment would cluster near 100%. Instead the company hits its number while most reps miss theirs, because a minority carries the plan.

A salesperson is not a factory worker. Verbeke's meta-analysis of the drivers of sales performance puts two under the system's control, knowledge and role clarity, and three with the person: adaptiveness, cognitive aptitude and engagement. Process moves the first two. Hiring moves the rest. There's a reason manufacturing workers are not particularly charismatic salesmen.

Points 15 to 17 say remote selling should be the default, an engineer should take the meeting when needed, and territories stop mattering. Buyer data supports the first two: McKinsey's B2B Pulse found two thirds of buyers preferring remote or self-serve at many stages in 2021, and by 2024, 20% of B2B buyers would spend $500,000 or more without meeting the supplier.

Territories are about trust, and trust has a cultural shape. Guiso, Sapienza and Zingales found that trust between European countries tracks religion, conflict history and somatic similarity, and that lower trust means less trade and less investment. In an ideal and rational world territories go. In reality prospects are humans and humans have biases. If you sell to humans, you need to entertain the human theatre. I saw this running commercial teams in Jakarta, Kuala Lumpur, Hong Kong and Singapore.

This section is hypothesis; I have no dataset for it. Applied rigidly, the model produces a robotic team. The sales head controls each small step. Each failure triggers another rule. Reality would escape the manager's mental net; the manager would make the net tighter. A system built to remove wasted effort would become its main source.

The mechanism is Goodhart's law: when a measure becomes a target, it ceases to be a good measure. Larkin's 6% to 8% is the sales version. Time-use data before and after a full implementation would test the hypothesis.

1. Commissions get gamed. Larkin's 6% to 8%. Conceded. It is a case for linear plans, not for no plan.
2. Deming agreed with the book. Points 11 and 12 of Out of the Crisis abolish quotas and merit rating, on the argument that most variation belongs to the system. In sales, the quota distributions say the worker's share is large.
3. The tail may be thinner. Beck, Beatty and Sackett argue part of the skew O'Boyle and Aguinis found is measurement. Quota attainment still shows most reps missing.
4. Retention is not as clean as the book, and I, said. Palmatier, Scheer and Steenkamp: salesperson-owned loyalty moves sales growth, and it leaves with the rep. Handing accounts to operations should be paced by it.
5. My sample. One operator, teams of 3 to 80 reps. The 20-30% / 70-80% heuristic is untested on any dataset I could publish, and "incentives trump processes" is a repeated observation, not a controlled comparison.

None of these changes the verdict; together they narrow it.

scroll the table sideways →

1. Use its basic structure. Specialize, define stages, take administration off reps, route inbound centrally.
2. Keep the rep's discretion on outbound.
3. Keep individual incentives and design them. Linear where possible, short cycle, uncapped for stars.
4. Plan for the tail. A minority will carry the number.
5. Draw territories on trust, not on distance.

The book confuses control with understanding. Use its basic structure. Reject its promise that process can contain human behavior.

This is what I do instead. It is one operator's framework, and every number in it is a rule of thumb.

A sales team needs three elements in place: structure and operations, a performance culture, and incentives.

scroll the table sideways →

Element 3 has a dose. Too much delegates control of the team to the scheme: acceptable for acquisition, not for account managers. Too little means the head of sales has to build the performance culture with personal energy. Multi-level marketing companies pay poor commissions and compensate with cult-like machines. A head of sales is not an MLM cult leader; a leader with a personality strong enough to need none of this can stop reading and go enjoy the sports cars.

Three constraints size the scheme.

1. The motivation threshold. Bonuses worth 50% to 70% of base salary change behaviour. 10% to 20% does not; 70% does.
2. The power law. In a mostly outbound motion, 20% to 30% of reps produce 70% to 80% of sales. Pay them without a cap.
3. Payback. A rep's gross margin has to cover salary plus commission within a period the company's finances can carry: 3, 6 or 12 months depending on the business.

When wishful thinking about payback produces a stingy commission, the trouble arrives at rollout.