Day 48 · Phase G

A team isn't a group of people who can sell. It's a road anyone can walk

Splitting the four jobs · hiring without trusting your gut · commission buys only what you count · four mandatory CRM fields·4 moves · 4 diagrams

"Everything gets better once we hire a great salesperson" is the most expensive illusion in the business. A team is what survives the person leaving. If the process only lives in one head, you didn't hire an employee — you bought a one-off piece of luck.

Sales teams rarely die of bad people. They die because four jobs of completely different character get piled onto one person, who is then steered by a commission that can only count closed revenue. One move per part. One, split the work before the headcount — handoffs matter more than job titles. Two, don't trust the rapport in an interview; trust the sample of work he does in front of you. Three, commission never buys effort, only whatever you can count. Four, a CRM is where the next step lives when it isn't in someone's head.

MOVE 01

A deal is really four different jobs — split the work first, the people second Split the work before you split the headcount.

SDR / AE / CSMspecialisationhandoff cost

The principle in one line

Every deal holds at least four jobs. Sourcing — screening a crowd for people who might need this, on volume and discipline. Qualifying — deciding whether it's worth your hours, on the nerve to say no. Closing — diagnosis, value, objections, on experience and presence. Keeping — onboarding, renewal, expansion, on patience. The skills and tempos barely overlap: sourcing is repetitive labour, closing is a sprint, keeping is slow work. Force all four onto one person and the outcome is always the same — he quietly drops the most uncomfortable stretch, usually sourcing. So the first decision isn't how many people to hire, it's who does each of the four today, where the handoffs sit, and what travels across them.

Four jobs, three handoffs

Checklist · three steps from job titles to handoffs

1. Do a time audit on the last three deals you won: hours spent in each of the four jobs. Sourcing will come out near empty — not because it wasn't needed, but because it was being dodged.

2. Make the most painful cut first: almost every first split goes between sourcing and closing, the two whose tempos clash hardest. Watch for a month: did the closer's meeting count rise? If not, the bottleneck was never sourcing.

3. Put the handoff on one page: three required items plus a line saying "send it back if these are missing". Rejectability is the point — a handoff nobody can return just gets silently redone downstream.

What a handoff sounds like: "This one's yours. His words were 'reconciliation eats two days at every month end.' Budget never came up. The finance director decides. Confirm those three before you book the meeting."

Why it works · mechanism and evidence

Why it works

Mechanism: specialisation pays twice — fluency from repetition, and not having to switch mental gear between unlike tasks. It costs once, at every seam: handoff friction and lost information. So finer is not better. The net depends on your deal volume, and at low volume the friction outweighs the fluency.

Hard evidence (experimental psychology, clean mechanism) · switching costs something: Rubinstein, Meyer & Evans (2001, Journal of Experimental Psychology: Human Perception and Performance) had subjects alternate between task types; the switch itself reliably cost time, and cost more as tasks grew complex or rules unfamiliar. That's why "cold calls in the morning, a major negotiation in the afternoon" does both badly.
Hard evidence (fMRI, specific brain region) · multitasking is a hardware limit: Dux, Ivanoff, Asplund & Marois (2006, Neuron) used time-resolved fMRI to localise a central bottleneck in the posterior lateral prefrontal cortex: when two tasks both need it, they queue rather than run in parallel. It isn't a willpower problem — the channel is only that wide.
Field practice (not research, don't treat it as a finding) · separating sourcing from closing: Ross's Predictable Revenue documents what Salesforce did internally. No control group. It reproduces well at high volume with a standardised product; copied into a small team it often backfires.

Honest boundary: those experiments run in milliseconds, and converting them into "switching costs a salesperson X% of the day" is extrapolation, not a research finding — they support "switching has a cost", not any number. Also, splitting narrows what each person sees: whoever does sourcing stops hearing real closing conversations and their judgement drifts. Sitting them in on a live meeting periodically isn't a perk, it's calibration.

MOVE 02

Don't hire the person who looks like a salesperson. Hire the one who does well in the sample Stop hiring the one who interviews well. Hire the one who does the work in front of you.

structured interviewswork samplesthe extraversion myth

The principle in one line

The most expensive misjudgement usually happens the moment you think "we really clicked". A free-flowing interview measures whether he is likeable to you — which is precisely the skill every salesperson has practised most, so you have chosen the most easily faked signal available and bet a hire on it. Only two things replace it: the same questions for everyone, scored on the same scale, and watching him do a piece of the real work. And one more filter to dismantle on purpose: stop screening for extraversion.

Which signals actually predict later performance

Template · a 45-minute work sample (worth more than ten questions)

Sent in advance: a one-page description of your real product, plus public information on a real customer (site, job ads, news).

Task one (15 min, live): "I'm head of operations at that company. You have ten minutes. Go." You play an ordinary difficult buyer: busy, mildly sceptical, volunteering nothing. Watch for: how many questions he asks; whether they're about his pitch or my situation; whether he follows up when I go vague or just keeps talking.

Task two (15 min): "What went badly there, and what would you change on a second run?" Watch for: whether he can review himself concretely. A person who can is one a process can amplify; a person who blames the customer isn't.

Task three (15 min): hand him a real objection ("we're already using someone else") and have him write three lines back. Watch for: does he acknowledge before he answers, or collide head-on.

Scoring sheet (1–5 each, identical for everyone): question quality / listens before speaking / self-review / handles vagueness / clarity. Score independently before anyone speaks — discussing first lets the first voice contaminate the room.

Why it works · mechanism and evidence

Why it works

Mechanism: the accuracy of any selection depends on how much of the signal is genuinely job-related. A free chat hands the criterion over to momentary warmth and similarity, which pours noise into the judgement. Fixed questions on a fixed scale put candidates on one baseline. A work sample is more direct still — behaviour predicting behaviour, with one less layer of translation.

Hard evidence (large meta-analysis, and an important revision) · the ranking: Schmidt & Hunter (1998, Psychological Bulletin) produced the validity table that got cited for two decades; Sackett, Zhang, Berry & Lievens (2022, Journal of Applied Psychology) recorrected it and found the old figures systematically inflated by overcorrection, revising most downwards (structured interview around .42, work sample around .33, unstructured around .19). Honestly: stop quoting the old absolute values — but structured beats unstructured in both versions.
Hard evidence (very clean design) · an unstructured interview actively adds noise: Dana, Dawes & Peterson (2013, Judgment and Decision Making) had interviewers interview real people, some of whom were instructed to answer at random. Interviewers still formed confident impressions, most never noticing the answers were random, and predictions made with such interviews were worse than using academic records alone. It isn't "slightly less information" — it's confidence in noise.
Medium strength (single large sample, no large-scale replication seen) · extraversion isn't the filter: Grant (2013, Psychological Science) followed 340 call-centre sellers over three months. Extraversion and revenue weren't linearly related; the ambiverts in the middle earned most. One role, one sample, so not a law — but enough to strike "must be outgoing" from your requirements.

Boundaries and one red line: structured doesn't mean stiff — probe and follow up freely; what's fixed is the question list and the scoring dimensions, not the conversation. A work sample only works if the task really resembles the job, so no brain-teasers. And the hard line: a sample task must never become free labour — beyond an hour or two, or if you intend to actually use the output, pay for it.

MOVE 03

Commission can't buy effort. It buys what you can count — and quietly defunds the rest Commission never buys effort. It buys whatever you can count, and quietly defunds everything you can't.

multitask agencyaccelerators and bunchingdiscount authority

The principle in one line

Commission was never about how much. It's about what you pay on. People move their hours toward whatever is both measurable and convertible into money, and the dimensions you can't measure don't merely shrink — they get systematically crowded out. Pay on new logos and nobody handles handoffs or renewals. Pay on booked revenue and discounts get handed out freely, because a discount barely touches the base he's paid on. Pay on calls made and you'll receive a pile of call logs and zero customers. So ask one question first: what is the laziest way to game this number? Whatever that is will happen — not because people are bad, but because you're paying for it.

The shape an accelerator always grows

Checklist · four questions to answer before designing any commission

1. What's the laziest way to game this metric? On revenue → give discounts. On new logos → ignore renewals. On activity → pad the logs. Once you know, either change the metric or block the route directly. The most useful block: charge discounts against his commission base, so giving margin away costs him his money.

2. Who owns the things you can't count? Handoff quality, telling the truth, helping a colleague — none of it ever fits in a commission. Only three fixes work: lower the intensity of the single metric (don't run pure commission), tie part of the payout to the team or to renewals, and delay vesting. Adding one more commission line to cover the gap mostly just creates a new surface to game.

3. Accelerators or not? They do drive a sprint, and the price is the shape above. If you use them, lengthen the period and set the threshold within reach; anyone far below it will write off the quarter and hide deals until the next one.

4. How do you set the rate? There's no universal number, only a derivation: take gross margin per deal, decide what share of that margin selling can consume (your unit economics decide this, not your competitors'), multiply by expected deals, and that's the ceiling on one person's total package — then split it into base and commission. Common industry practice is a roughly even split (a checkable convention, not a research finding); the simpler and more individual the deal, the more can sit in commission.

Why it works · mechanism and evidence

Why it works

Mechanism: when one person must do several things and only some of them can be measured accurately, strong incentives on the measurable part pull effort away from the rest. That isn't an attitude problem — it's a rational response to the price list you handed him.

Hard theory (Nobel-backed) · multitask agency: Holmström & Milgrom (1991, Journal of Law, Economics, & Organization) proved exactly this, with a counterintuitive corollary: when the important dimensions are hard to measure, optimal incentive intensity is lower, sometimes flat pay — because strong incentives there mainly distort. Holmström received the Nobel Prize in Economics in 2016.
Hard evidence (deal-level data from a real firm) · accelerators get paid for in timing and discounts: Larkin (2014, Journal of Labor Economics) analysed deal-level data at an enterprise software company: reps timed deals (pulling them forward or pushing them out) and discounted more to cross accelerator thresholds, at an estimated cost to the firm on the order of a few percent of revenue. Earlier, Oyer (1998, Quarterly Journal of Economics) used fiscal-year-end sales spikes to make the same point: that seasonality is manufactured by the incentive structure, not by demand.
Hard evidence (natural experiment, big effect, narrow boundary) · commission does work, under strict conditions: Lazear (2000, American Economic Review) documented Safelite auto-glass installers moving from hourly pay to piece rates: output per worker rose about 44% — and crucially about half of that came from sorting (slower workers left, faster ones joined), only half from existing staff working faster. The boundary is clear: output that is simple, measurable and purely individual. The closer your selling is to that shape the better commission works; long cycles, team effort and hard-to-judge quality make it less so.

Boundary: commission addresses willingness, never capability — a low conversion rate is usually a skill and process problem, and adding commission there just disappoints your best people into leaving first. Same for individuals: reward yourself only on countable metrics (posts published, people met) and the uncountable things that actually compound get crowded out by your own scheme.

MOVE 04

A CRM isn't surveillance for the boss. It's where the next step stops living in one head A CRM isn't surveillance. It's where the next step lives when it isn't in one person's head.

externalising processfour required fieldspipeline hygiene

The principle in one line

A CRM becomes a filing punishment for exactly one reason: the fields were designed for management's report, not for the seller's next action. The test is plain — if someone is away for two weeks, can a colleague pick up his deals from the CRM alone? If not, what you're storing is a record, not a process. So make only four fields mandatory: what the next action is, who does it, the date it's done by, and which specific obstacle the deal is stuck on. The first three turn a vague "keep following up" into a plan with a trigger; the fourth forces someone to name the blocker instead of parking the deal in the comfort of "in progress" forever.

One deal, two ways of writing it down

Template · three hard rules that make people willing to fill it in

1. Four fields required, everything else optional. Each extra mandatory field costs you a slice of data quality; compute the reporting fields yourself from those four instead of levying them on the front line.

2. "Next step" must be an action plus a date, never a status. "Keep following up" and "waiting on the customer" don't count. "Turn the reconciliation piece into a 20-minute demo and send it to Zhang by 6 November" does.

3. Never make CRM activity counts a performance metric. The moment calls logged becomes something people are judged on, you've imported the crowding-out problem from the last move straight into your system, and the data turns beautiful and useless. A CRM should be the seller's own workbench: it remembers what he has to do, rather than watching what he did.

What to say in the weekly meeting: "The next-step dates on these three have passed. Those are the only three we're discussing — are they stuck, or should they be closed out?"

Why it works · mechanism and evidence

Why it works

Mechanism: intentions become actions not through resolve but by being hung on a specific time and situation, so that nothing has to be decided again in the moment. Meanwhile, writing down a tacit process that exists only inside one head is what lets a team stop depending on that person — but externalising has no magic of its own. What works is people genuinely changing behaviour because of it.

Hard evidence (meta-analysis of 94 studies) · "when and what" lifts follow-through: Gollwitzer & Sheeran (2006, Advances in Experimental Social Psychology) pooled roughly 94 studies and over 8,000 participants, finding that implementation intentions in "if X, then I do Y" form had a medium-to-large effect on goal attainment (d around 0.65). The mechanism is pre-attaching the trigger to a concrete situation. "Action plus date" in a CRM is the workplace version of exactly that; "keep following up" is not.
Hard evidence (cross-industry, and the negative result must be told too) · turning tacit process into a checklist: Haynes et al. (2009, New England Journal of Medicine) introduced a surgical safety checklist across eight hospitals: post-operative mortality fell from about 1.5% to 0.8% and complications from about 11% to 7%. But Urbach et al. (2014, same journal) found no significant improvement after a mandated province-wide rollout in Ontario. Together they teach precisely this section's point: a checklist has no power of its own; what works is a team actually using it to change behaviour, and mandated form-filling changes nothing — which is also how CRM projects most commonly die.

Boundary: with two or three people the benefit is mostly not forgetting things, and a shared spreadsheet is plenty. Only two signals really call for a system — handoffs starting to happen and forecasting starting to matter. Adopt too early and you'll spend months configuring a system with no data in it. Also, a checklist governs "don't drop things", not "should we work this deal at all" — that's qualifying, and no field will decide it for you.

Your Day 48 Action

Ninety minutes with the team you have — even if that's just you.

One (20 min) · time-audit the four jobs: take the last three deals you won and write down the hours spent on sourcing, qualifying, closing and keeping. Whichever is near empty is the one you've been avoiding.

Two (20 min) · write the three handoffs: one page — what each of the three must carry, plus "send it back if these are missing". Send it out today and ask, "could you actually pick a deal up from this?"

Three (25 min) · run a gaming test on your commission: write down the rules you use (or plan to), then force yourself to answer — what's the laziest way to game this? Who currently owns the things you can't count? Change one thing to block the route; most often that's charging discounts against the commission base.

Four (25 min) · four-field audit: open the CRM or the spreadsheet and fill in next action / owner / done-by date / what it's stuck on for every deal. Then count: how many have no next step at all? Those aren't pipeline, they're hope — take them out of the forecast today.

A boundary worth keeping: all four assume the thing already sells. Building a team and a system before that isn't scaling, it's copying an unvalidated process to more people and then confirming it doesn't work the expensive way.

Think It Through

1. There are only two or three of us. Is this something to revisit once we're bigger?

Don't split the roles yet — but build the process now. These two get conflated constantly, and they cost very different things.

What not to do: carving three people into SDR / AE / CSM only leaves everyone waiting on everyone else. At low volume the handoff friction outweighs the specialisation, and one person carrying a deal end to end is faster.

What to do now, and it's only two things, both free:
1. Write a next action and a date on every deal, a shared spreadsheet is enough. It's not there to stop anyone slacking, it's so you can remember three weeks later where the deal stalled.
2. Write three lines after every win and every loss: why he bought or didn't, which sentence changed the temperature, what you'd say differently. Six months of those is the training material for your first hire — and it's exactly what most teams lack, since all they can hand a new person is "go sell".

When to make the first cut: when you notice you've gone two straight weeks without doing any sourcing, only working existing deals. That means closing and sourcing are already fighting over the same person's time.

2. What percentage should the commission actually be? There has to be a number.

There's no universal number, and copying competitors is the most common mistake: their margin structure and volume aren't yours, and the same rate can simply lose you money. But there's a derivation you can run yourself:

One: work out gross margin per deal (revenue minus cost to deliver), not revenue.
Two: decide what share of that margin selling can consume. Your unit economics set this: a high-margin, low-delivery-cost business can afford it; a thin-margin one can't.
Three: margin × that share × deals one person closes a year = the ceiling on their total package. Then split it into base and commission. Common practice is a roughly even split (a checkable convention, not a research finding); the simpler the deal, the more individual the effort and the shorter the cycle, the more can sit in commission.

Three things that matter far more than the percentage: the payout period (paying quarterly grows the quarter-end spike), whether discounts hit the base (if they don't, you're subsidising every concession), and whether renewals and handoffs are tied in (if not, nobody owns how the customer fares after signing). The same money, structured differently, buys entirely different behaviour.

One more thing that gets missed: the more often you change the rules, the faster the incentive decays. An imperfect plan held steady for a year usually beats one optimised every quarter — people only change behaviour for something they believe will pay out.

3. My best seller won't use the CRM or follow the process. What do I do?

Separate two cases first; the treatments are opposite.

One: the process really is useless. He has fifteen fields to fill, twelve of them for somebody else's report — that one's on you. Cut the required fields to four, and cut them in front of him; it works better than any amount of persuasion. One question decides it: is there a single thing he fills in that he himself will use tomorrow? If not, it's pure tribute, and the resistance is entirely reasonable.

Two: he's protecting an information monopoly. The relationships live only with him, and that's where his leverage comes from — rational self-protection, not a character flaw. Don't moralise, change the structure: put handoff quality into how he's evaluated, give new leads first to the people who share, and be straight about the cost — an account nobody else can touch is one the company won't keep investing in.

And the hardest case: he really is winning on instincts that can't be written down. Don't force the process on him — extract it instead. Have him walk through a real win while you ask, line by line, "why did you say it that way?" He can't articulate it, but you can hear the pattern. Sales processes were never designed from scratch anyway; they're stripped off the best person in the room. So don't ask him to comply with a form — ask him to author the method.

One last honest note: someone who neither produces nor cooperates isn't a process problem, it's a people problem, and facing it early is cheaper.