engagement

engagement 015 · 2026

Clearing a three-year backlog in thirty days

Three years of internal product requests — dispatch, quoting, exception handling, client portal — stacked up because no one had encoded what the rules actually were. We found the two invariants, built the harness, and cleared the backlog in a month.

client
A US logistics services company (80 people)
sector
Logistics · B2B services
duration
30 days
30 days
to clear a 3-yr backlog
< 7 days
request → usable prototype (was ~45)
93/100
client satisfaction (was 70)
2
invariants encoded

An 80-person logistics company serves several distinct B2B client types. Over three years, internal product requests had accumulated — dispatch tooling, quoting flows, exception handling, a client portal — each raised by a business team that genuinely needed it, each stalled because the development process had no shared specification layer. Requests took twelve to eighteen weeks from raised to usable. Most never got there.

We started not with the backlog but with the rules underneath it. What does every quote have to carry, regardless of client type or carrier? What exceptions must never be resolved without a human dispatcher’s judgment? Two invariants emerged. Everything else — the feature prioritization, the prototype sequencing, the client-facing configuration — was judgment. That is the arc: one meta-rule (the operation runs on encodable rules), two invariants (the company’s private instances of that rule), and judgment freed to do its actual job.

Two invariants, one clear boundary

The first invariant: any exception affecting a service-level agreement must route to a human dispatcher. Not a queue. Not a heuristic. A person. The second: every quote must record cost, lead time, carrier constraints, and the client commitment together in one place — not split across three tools, not reconstructed from email.

Those two rules had never been written down. Once they were, the backlog became a sequencing problem, not a judgment problem.

“For the first time, what the business team asked for didn’t fall into a black hole.” — COO, anonymized

The harness

We did not deliver a prioritization framework. We delivered a running system: a spec-to-prototype pipeline that takes each internal request through a decision trace — request, specification, prototype, live — visible to both the product and business teams. The trace is the audit record; the record is what ends re-litigation of the same decisions.

Over a 30-day concentrated effort in 2026 Q1, the ~120–150 open P0 and P1 internal requests were cleared. Median time from internal request to usable prototype dropped from roughly 45 days to under 7. Client satisfaction, measured by client-success survey over a 60-day post-go-live window, moved from 70 to 93 out of 100. The burn-down was not a sprint; it was what happens when two rules are written down and the judgment that needs a person finally has room to breathe.

how it ran

  1. 01

    Discovery

    Mapped ~120–150 open P0/P1 requests. Found the two rules doing the real work — and the one call that genuinely needed a person, which had been blocking everything else. 1 week.

  2. 02

    Specification

    Wrote the two invariants as machine-readable rules, defined the decision trace each request must carry, and sequenced the backlog by dependency. 1 week.

  3. 03

    Harness

    Shipped the running system: spec-to-prototype pipeline, SLA exception routing to human dispatchers, and the quote record that captures cost, lead time, carrier constraints, and client commitment together. 2 weeks.