Stage 4: Build & delivery (4a Build, 4b Engagement lead)
Stage 4: Build & delivery (4a Build, 4b Engagement lead)
Detail for stage 4 of e2e-stages. The stage runs from a signed SOW to a portal the client accepts. It has two halves that need different skills, and the aim is that one person holds both: no PM/coder split. You own stage 4 when you are at level 2 on both halves, on the same engagement.
Who is developing here (from e2e-stages):
- 4b Engagement lead: Ola (main move, starting with JoPACC), and possibly Daniela. João is already engagement lead on NGED (National Grid) and MoEI. Luccas is developing in 2/2a instead (decided 2026-09-28).
- 4a Build: Daniela, possibly Monika, Osahon to a small extent. Luccas to evaluate.
The flow
signed SOW (from 3 / 3a)
│
▼
┌─ 4b ENGAGEMENT LEAD (client side) ───────────────────────┐
│ │
│ 4b.1 Kick off & plan plan the client agrees │
│ 4b.2 Run the cadence weekly update, demo, risks │
│ 4b.3 Guard scope in scope, or change request │
│ 4b.4 Review & accept review cycles, sign-off │
│ 4b.5 Lead the team people + agents, throughout │
│ │
└──────────┬────────────────────────────────▲──────────────┘
what to │ │ what's done,
build ▼ │ what's hard
┌─ 4a BUILD (engineering; agents do the labour) ───────────┐
│ │
│ 4a.1 Architecture call │
│ 4a.2 Data migrate, harvest, schema │
│ 4a.3 Features & design search, charts, brand │
│ 4a.4 Harden & stage QA, quality bar, staging │
│ │
└────────────────────────────┬─────────────────────────────┘
▼
4b.6 Hand over into 5 (launch & run) and 6 (renewal)
4b Engagement lead
What it is: being the one accountable name the client deals with during delivery. You turn the SOW into a plan, keep the client informed and deciding, guard the scope, lead the team, and get sign-off.
| sub-step | owns it (level 2) when they… | evidence | judgement that stays human |
|---|---|---|---|
| 4b.1 Kick off & plan | run the kickoff and turn the SOW into a milestone plan the client agrees to, naming the client's decision-makers and reviewers | kickoff notes; plan agreed in writing | what to tackle first; what the client really needs versus what the SOW says |
| 4b.2 Run the cadence | send a regular (e.g. weekly) update and demo, with progress, risks and decisions needed, without being chased | the update series; demo recordings or notes | what to raise, and when |
| 4b.3 Guard scope | triage every client request as in scope or a change request, and route change requests through change control (3a) rather than absorbing them silently | change-request log; any change orders | what is in scope; when to give something away and when to charge |
| 4b.4 Review & accept | run structured client review cycles and get written sign-off against the SOW | review feedback log; sign-off email or document | when the work is actually done |
| 4b.5 Lead the team | allocate the work across people and agents, unblock, and hold the quality bar | task board; the team's view | who does what; when output is good enough to put our name on |
| 4b.6 Hand over | hand over into launch & run (5) and the client relationship (6), noting renewal and expansion signals | handover note; expansion notes in Twenty | what to promise next |
Escalation (guardrail). Tell Rufus or Anuar early when margin, scope or the relationship is at risk. Escalating is part of owning it, not a failure.
A good sign it's real: the client emails them first, not Rufus or someone else.
4a Build
What it is: building the portal, with agents doing the labour and the operator making the calls and checking the output. For commercial people (Daniela, Monika), the aim is not hand-coding. It is to specify, direct and review agent-built work well enough to stand behind it.
| sub-step | owns it (level 2) when they… | evidence | judgement that stays human |
|---|---|---|---|
| 4a.1 Architecture call | choose the stack and shape (PortalJS, CKAN, Arc hosting, integrations) and write it down in a page | architecture note | which architecture fits |
| 4a.2 Data | run the migration or harvest, define the metadata schema, map to DCAT where needed, and check data quality | migrated catalogue; schema; QA report | what "good enough" data looks like |
| 4a.3 Features & design | build custom features (search, charts, maps, integrations) and apply the client's brand, using the PortalJS skills | feature list delivered; design screenshots | design quality; avoiding competent-but-generic output |
| 4a.4 Harden & stage | harden a 2a PoC or rebuild it, test it, pass the quality bar and accessibility checks, and deploy to staging | staging URL; checks passed; review notes | when to throw a PoC away |
Quality bar. Osahon owns the quality bar (training). Level 2 in 4a means work passes that bar with a builder reviewing it, not doing it.
Levels
Counted per engagement or major milestone, not per week, since delivery is lumpier than outbound. The markers are suggestions, to calibrate.
| level | 4b Engagement lead | 4a Build |
|---|---|---|
| 0 Not yet | has not led client-facing delivery work | has not built on a client project |
| 1 With help | runs parts (e.g. the weekly update, reviews) while someone else holds the relationship | builds parts with a builder alongside doing the hard calls |
| 2 Owns it | named client lead for one engagement or major milestone, from kickoff to sign-off, with a change-request log (no silent scope creep) | delivers a milestone to staging, directing agents, passing the quality bar with review only |
| 3 Improves it | owns it, and improves the kit: acceptance workflow, update template, change-request intake | owns it, and improves the kit: a PortalJS skill, hardening checklist, regression case |
Pairing: learn the other half on the same engagement
The gaps are mirror images, so pair people on one engagement, each leading the half they are learning and backing up the other:
| pair | learning 4b (leads client side) | learning 4a (leads build) |
|---|---|---|
| e.g. one engagement | Ola | Daniela or Monika |
João is the in-house model for 4b. Ola can learn from how he runs NGED and MoEI while she leads JoPACC.
Each person takes the lead in their learning half, and the other acts as the safety net. Swap on the next engagement. This is the fastest route to one person holding both halves, and it avoids reintroducing the PM/coder split, since both people do both halves over time.
Starting points (first read, to confirm with each person)
| person | 4a | 4b | first aim |
|---|---|---|---|
| Ola | 2? (product) | 0 | engagement lead on JoPACC (Jordan Payments); João as the model to ask |
| Luccas | 2? | 0 | not a 4b focus for now (developing in 2/2a); evaluate 4a depth |
| João | 2? (Canada) | 2 (NGED, MoEI) | 4b level 3: turn how he runs NGED and MoEI into the kit (update and acceptance templates), and let Ola shadow him |
| Daniela | 2? (Arc PoC) | ? | the Swiss deal: harden her own PoC (4a.4) and lead the client (4b) |
| Monika | 1? | 2?-like skills | 4a level 1: direct agents on one feature with a builder reviewing |
| Osahon | ? | ? | agent fluency in 4a first |
Kit: what exists, what to build
From training module 4.
| kit | supports | status |
|---|---|---|
| PortalJS skill chain end-to-end, tested on 3 past engagements | 4a | ? |
| PoC-to-production hardening checklist | 4a.4 | ? |
| client review & acceptance workflow tied to the SOW | 4b.4 | ? |
| change-request intake feeding 3a change control | 4b.3 | ? |
| kickoff and plan template | 4b.1 | not in training yet |
| weekly client update template | 4b.2 | not in training yet |
Open questions
- Which engagements are the first practice ground for 4b, and does the Swiss deal count for Daniela?
- Luccas and 4a: how do we evaluate whether he's more effective there?
- Does pairing need a cost cap? Two people on one engagement costs more until the learning pays off.