Anatomy of a 90-Day NATIVE Engagement

Why this is written illustratively

Most of our engagement work lives under NDA. Enterprises do not typically grant permission to name them in a public case study, especially while a transformation is mid-flight. Rather than either fabricate a customer or wait a year to publish anything, we write engagement walkthroughs the way a surgeon might describe a procedure — representative, drawn from composite experience, structured so the pattern is legible even though the patient is not.

If you are evaluating ScaledNative for a residency and would prefer a live reference call with a named practitioner and a current client, we can usually arrange one under NDA. Contact us and we will find a match.

Pre-engagement: what we need from you

Before day 1, the enterprise supplies three things. First, a problem statement narrow enough to be finished in 90 days — not “become AI-native” but “rebuild the underwriting adjudication path to run as an agent loop with human review on exceptions.” Second, access to a production-representative environment: codebase, data sample, and at least one engineer who can merge PRs. Third, a named executive sponsor who will block and tackle on security, legal, and procurement when residency practitioners hit a wall.

We do not start until all three are in place. Residencies that begin without an executive sponsor end at week 6 trying to get a VPN credential provisioned.

Week 1–2 · Navigate

The residency lead and a staff practitioner spend the first two weeks on discovery. Not workshops — systems archaeology. We read the codebase, trace the production data flow, interview the engineers who actually merge the critical paths, and map the implicit operating model (how work moves from idea to production, who bottlenecks it, what the unwritten rules are).

Artifact: a Navigate Memo, usually 8–12 pages. It states the problem back to the sponsor in our words, identifies the three or four constraints that actually bind, and proposes the narrowest intervention that moves the system. We do not ship the Memo as a consulting slide deck. It is a living markdown file checked into the engagement repo.

If the sponsor disagrees with the Memo, we stop. A residency where the enterprise and the practitioners disagree about what the problem is will burn 90 days and produce a deck.

Week 3–4 · Architect

With the Memo approved, we design the system. Model selection, data flow, evaluation harness, human-in-the-loop surface, rollback plan, compliance posture. Architecture decisions are written as ADRs and reviewed with the enterprise's platform and security leads before any code ships.

Artifact: a committed ADR set, an evaluation harness skeleton, and a first production-shaped prototype running against synthetic data. Not a demo — the actual scaffolding that will become the production system by week 9.

Week 5–7 · Transform

This is the build. Practitioners pair with the enterprise's engineers daily. Every PR ships into a staging branch the enterprise controls. We push hard to keep the review cadence tight — a residency where the enterprise takes three weeks to review a PR loses the week of delivery it was supposed to catch up.

Artifact: a working system behind a feature flag, with a running evaluation harness showing regression on every commit. At this point the enterprise's own engineers should be reviewing PRs on the critical paths, not just observing.

Week 8–9 · Integrate

The system lands in production behind a flag, rolled out to an internal audience first, then a small external cohort. Observability, alerting, and the on-call runbook are written jointly with the enterprise's SRE team. Compliance, legal, and the DPO sign off before external rollout.

Artifact: a live production system, a written runbook, an evaluation dashboard the enterprise owns, and a signed-off rollout plan.

Week 10–11 · Validate

The system runs in production while we observe, tune, and harden. Failure modes get added to the eval harness. Edge cases get codified. The enterprise's engineers are now shipping changes without us pairing on every commit.

Artifact: two consecutive weeks of the enterprise shipping improvements unassisted. If this doesn't happen, we haven't finished the transfer.

Week 12+ · Evolve (handover)

The last two weeks are governance, not code. We write the operating model — how the enterprise will evolve this system without us. Hiring profile for the next two engineers. Evaluation cadence. Model upgrade policy. Rollback thresholds. Incident postmortem template. A 30-minute weekly cadence the team will run after we leave.

Artifact: an Operating Model Memo. This is the most valuable document we produce. It is how the enterprise avoids backsliding into RFP-driven procurement six months later.

What you own on day 91

• A production system owned by your engineers, not us.
• An evaluation harness running on every commit.
• A written operating model describing how you will evolve the system.
• At least two of your engineers who shipped this alongside us and can extend it.
• A credentialed residency lead (SNCP) who has signed the final PR and will take your reference call.

You do not own a slide deck describing a future transformation. You do not own a new dependency on us. If you call us again, it is because you have a different problem, not because the last one is still running.

Where engagements usually break down

Three failure modes, in roughly decreasing order of frequency.

1. Scope drift from the sponsor. A new executive is hired in week 4. The problem statement becomes “and while you're at it, also rebuild onboarding.” We push back hard. Residencies that swallow scope creep do not finish.
2. Review latency. A PR waits two weeks for review because the enterprise's engineers are on other priorities. We flag this at week 3 and escalate to the sponsor at week 4 if it doesn't resolve.
3. Procurement cutting compute or data access. Rare, but recoverable — the sponsor is the unblocker here. If the sponsor cannot unblock it, the engagement was mispriced.

None of these are fatal if they're named early. Most of them are fatal if they're not.