Shaped by Demand: the power of fluid teams - Daniel Terhorst-North (Talk Outline)

Daniel Terhorst-North (~30 years in IT; coined BDD) on organization design — for 10–15 years working out “what shape an org should be.” Golden nugget: take care of each other; compassion is what matters. Definition: “build the best structure we can to do the best work we can — for now” (context changes). The challenge evolved from getting 8 people pointing the same way (cross-functional teams, ~2000s) to 80 or 800 (only solvable after you’ve solved the 8).

1. The Goal of Org Design

• Big consultancies push cookie-cutter approaches; asked “what’s the outcome?” they go “crickets… (billable hours).” His pithy definition: the best structure to do the best work, for now, knowing it’ll change.

2. What Is “the Best Work”?

• Not just delivery (ship/enhance features) — the whole picture:
  • Discovery — solution-space (solve the problem better) and problem-space (understand user needs better).
  • The work about the work / Kaizen — his broader definition: continual improvement of the system of work (technical improvements, automation, and learning/development — Evette teaching him a thing is first-class work that changes the system so two people now know it).
  • BAU (business-as-usual / run-the-bank) — reports, controls, compliance/security/regulatory/audit.
  • Failure demand — work that should never have existed (outages, process/reconciliation failures) plus all the follow-up (finding/fixing corrupted data, the meetings about the meetings).
• Mandatory spend = BAU + failure demand (can’t skip and stay in business); discretionary spend = delivery + discovery + Kaizen.

3. Debunking “Stable, Long-Lived Feature Teams”

• The received wisdom sold by Scrum/SAFe/LeSS/Scrum@Scale training. If it weren’t true, none of the scaling apparatus would matter.
• “Stable” ← Bruce Tuckman’s meta-study (forming/storming/norming/performing) — but 50% of its groups were therapy groups (individuals feeling vulnerable, sometimes under duress) and 25% HR training (eyes-forward, slides); only 25% were actual working groups. Later studies (a good US military one) show the stages happen concurrently and very fast — a shared-culture consultancy team (his decade at ThoughtWorks) performs in hours because of shared values/expectations/capability. So teams don’t need to be stable to perform.
• “Long-lived” (a corollary of stable) assumes homogeneous work over time — false: at Bank of America, ~150 London / 200+ NY people faced 30–40 scary legacy systems and work that changed dramatically quarter to quarter (mainframe updates one quarter, SQL tuning the next) — Scrum couldn’t cope.
• “Feature team” ignores that some work is Kaizen/automation/tech-debt — needing different team shapes.
• Conclusion: instead of shaping demand to fit teams, shape teams to fit the demand (demand is the relatively fixed thing).

4. Demand-Led Planning

1. Identify all the demand — we’re good at identifying demand but bad at seeing what hides in plain sight. Case study (Gene Kim / HP LaserJet firmware): measuring failure demand revealed ~95% (2 hours/week of discretionary time — “methodology is irrelevant then”); 8 months of dull automation/tests drove it to 40% → 3 days/week to do work. Mandatory demand is a rich source of Kaizen (production failures → improve path-to-live/testing/resilience; reporting toil → automate).
2. Allocate discretionary spend — decide the mix of feature/Kaizen/discovery each quarter (e.g., 80/20 to Kaizen this quarter). The only rule: none goes to zero (features = “walking-around money”/reputation; Kaizen = improvement; discovery = not flying blind).
3. Pitch the demand — sponsors (product managers for features; platform leads / report-owners for Kaizen) pitch to everyone in the room (30–200 people; fly them together — virtual is “a bit rubbish”; the energy is palpable).
4. Introduce constraints:
   • Governing (rules of the road): 3–10 people per team (fewer → always in meetings; more → in each other’s way); teams fully autonomous (specific meaning: never blocked on another team — else your “priority 2” item sits at 2 forever); teams need skills, knowledge, experience, relationships, permission, and resources (things, not people); and you must do all this work.
   • Bumblebees/butterflies — people deliberately not in a team who rove between teams (≤3 teams per bumblebee, ≤3 bumblebees per team — heuristics).
   • Enabling (encourage behavior): min/max movers per quarter (early John Lewis: “only two people may stay” to force knowledge transfer; later “no more than two may leave” for stable metrics), min/max tenure (e.g., 2–3 quarters).
5. Self-organize against the constraints — expect ~30 seconds of nothing, then someone brave (a contractor) walks to the unloved reporting work and others follow (applause). Iterate 2–3 times to a workable configuration (“mostly right” is fine).
6. Each new team declares a BHAG (aspirational OKR — automating 2/3 of a 10-step process is a great quarter) and a team name (naming is hard). Then break bread (pub).

5. Making It Work

• Before: evangelize up and down — get the “suits” (money/risk/compliance) comfortable with “shaking 100 people in a bag,” and get the 100 people comfortable too (expect literal tears and tantrums — a program manager rage-quit calling it a waste, then two years later: “you were right, I just needed to process it”). The scariest thing is uncertainty → road shows, lunch-and-learns, AMAs; coach sponsors to pitch (they get sharper by the 2nd/3rd session); recruit facilitators (ideally experienced); big room, natural light, fruit + good coffee.
• During: manage energy, not time (structure for meals/breaks, otherwise flex); contain loudmouths; listen to your facilitators (they see what you don’t); keep the conversation flowing (bring stationery; draw in the quiet domain expert, gently rein in the loud one — teams form/storm in real time, minutes not weeks).
• After: follow up with an academically defensible Likert survey (built with an academic — “most surveys aren’t defensible”; cf. Nicole Forsgren). Statements like “I feel connected to the work / excited about next quarter / can see how my work impacts the business” landed on 4–5 (agree/strongly agree) — real connection vs. work handed down a backlog.

6. Metrics, Identity & Management

• Teams measure themselves with throughput (things done since last time), lead time (a range, not a number), and WIP — “don’t even define ‘things’; each team counts its own.” Reviews are forward-looking — not “what have you done” but ”how much is left, and what’s the best way from here to the goal?” (OKRs).
• Identity is in the program team (~50–100, sweet spot ~60–80); the current squad is transient. (Early Spotify tribes had matrixed belonging — “streaming squad” vertically, “testing guild” horizontally — before squads calcified into long-lived teams.)
• Management in a lean org manages a system of work that has people in it, not people (not “policeman + mom”). The sponsor who cares enough to pitch de-facto leads that work, then rejoins the pool — decoupling seniority from management. Employee-lifecycle stuff (appraisals, recruiting) happens out of band, ideally owned by guilds/practices (testers appraise testers — no conflict of interest).

7. Q&A

• Ownership / on-call with fluid teams? The program team owns its world, so the work doesn’t vanish; people generally care more, and on-call becomes a rich source of Kaizen (“if we automated X, this whole class of pages goes away → pitch it next quarter”).
• Is full autonomy even possible? It’s not binary — autonomy through alignment: governing constraints (which side of the road, stop on red) let anyone drive anywhere. Alignment without autonomy = autocracy; autonomy without alignment = anarchy. Give the minimum guidance/constraints + the resources they need, and people surprise you (mission command, post-WWII US military).

People, Companies & References Cited

• Daniel Terhorst-North — speaker.
• Bruce Tuckman (forming/storming/norming/performing — critiqued), Gene Kim (HP LaserJet failure-demand case), Nicole Forsgren (surveys), John Willis (“fractional evangelist”), Cat Hicks, Kent Beck, Nigel Thurlow, Dave Snowden (referenced).
• Orgs: ThoughtWorks, Bank of America, John Lewis, Spotify.
• Concepts: demand-led planning, five kinds of work, failure demand, Kaizen (system of work), governing/enabling constraints, bumblebees, BHAGs, throughput/lead-time/WIP, autonomy-through-alignment, program-team identity.

Video: https://www.youtube.com/watch?v=IPKb9XymCEE — Transcript via yt-transcript.sh; outline generated from the transcript.