Services

43 Small Requests a Year Cost More Than a Full Redesign: The Batch Economics of Outsourced Work

2026.08.24 · 6 views
43 Small Requests a Year Cost More Than a Full Redesign: The Batch Economics of Outsourced Work

Every request carries a fixed entry cost that has nothing to do with its size. Once average work time drops below it, slicing finer costs more — with the batching formula, a four-lane intake model, and a 90-day playbook.

Share:

A 60-person parts manufacturer raised 43 ERP work orders in 2025 — the smallest was "print the customer tax ID on the shipping note" — totalling NT$516,000. A full rebuild had been quoted at NT$450,000 two years earlier. Why did a year of nothing big cost more?

The answer is not the rate. It is the number of intakes.

Three industry myths

Myth 1: Small requests should cost small money. The floor on a quote is not labour hours but the cost of entry: reloading context, preparing an environment, re-running regression, booking a deploy, acceptance, invoicing — none of it scales with the size of the change.

Myth 2: Doing it immediately is good service. Queue-jumping buys perceived speed and sells throughput; the interrupted item pays its context cost twice.

Myth 3: Batching means slower. Under Little's Law, lead time equals work-in-progress divided by throughput — what slows you is items in flight, not the batch window.

The contrarian read: DORA's case for small batches assumes deployment cost approaches zero. With CI automation and one-click deploys, F may be 0.3 hours; in most SME custom systems F is 3–6 hours. Two paths: spend money to drive F down, or batch.

The formula and the four-lane model

Cost per intake = fixed cost F + variable hours V. Efficiency ratio = V ÷ (F + V); below 50%, half the money did not buy a feature.

Batching benefit = (n − 1) × F × merge rate m, where m = (n − k) ÷ (n − 1), n = annual requests, k = batch windows.

Worked example (F = 3.5 h, NT$2,000/h, V = 2.5 h, n = 43)

ItemOne intake per requestMonthly batches (k = 12)
Fixed cost43 × 3.5 = 150.5 h12 × 3.5 = 42 h
Variable hours107.5 h100 h (shared regression)
Total258 h / NT$516,000142 h / NT$284,000

m = (43 − 12) ÷ (43 − 1) = 0.738, so the benefit is 42 × 3.5 × 0.738 = 108.5 h ≈ NT$217,000; with shared variable savings the annual gap is NT$232,000, or 45%. This is economic order quantity: high ordering cost means larger batches.

When not to batch: the test is whether the loss per day of delay exceeds F. Four categories take the jump lane — security patches, payment failures, compliance deadlines, revenue-blocking bugs.

LaneTakesCadencePricing
HotfixThe four categories above4 h response, 24 h fixWithin monthly allowance
Monthly batchV < FFreeze date + delivery weekF charged once per batch
Quarterly mini-projectF ≤ V ≤ 3F, separate acceptanceOnce per quarterQuoted separately
Annual redesignV > 3FAnnual planningProject-based

Three contrasting scenarios

10-person e-commerce60-person manufacturer200-person multi-site
n1855120
F2.0 h5.5 h4.0 h
RateNT$1,800/hNT$2,200/hNT$2,000/h
LanesBiweekly batch + HotfixMonthly batchMonthly batch + quarterly (k = 16)
Annual saving~NT$29,000~NT$520,000~NT$780,000 net
VerdictNot worth slowing campaignsBleeding without batchingDowntime events 120 → 16

Same formula, opposite conclusions.

The hidden cost list (what F is made of)

  • Reloading context (old specs, business rules): 0.5–1.5 h
  • Branching and environment prep (seed data, test accounts): 0.3–0.8 h
  • Re-running regression: 0.5–2.0 h (0.2 h with automation)
  • Deploy window and downtime coordination: 0.3–1.0 h (30–50% out-of-hours premium)
  • Acceptance meeting plus invoicing admin: 1.5 person-hours + 0.3–0.5 h per ticket
  • Scheduling fragmentation: sub-4-hour tickets do not fit gaps, 20–40% premium

Total 2.0–6.5 hours, median around 3.5.

Vendor KPI scorecard (10 dimensions × 0–3, max 30)

#DimensionWhat a 3 looks like
1F transparencyComponents and hours itemised
2Minimum billing unit0.5 hour (half-day scores 1)
3Automated regressionTop 10 flows every batch
4Deployment automationOne-click deploy with rollback
5Batch calendarFreeze and delivery dates published
6Hotfix SLAWritten, four categories defined
7Merge discountF charged once per batch
8Batched invoicingOne monthly invoice with detail
9Says "not now"Proactively recommends deferring
10Quality trackingPublishes defects per batch

Reading it: 24+ is worth a long-term deal; 17–23 run one or two batches first; below 17, change vendor.

How we map to this, and where we are the wrong fit

  • One-off project: the annual redesign and quarterly mini-project lanes, for V > 3F.
  • Monthly retainer (NT$30,000–80,000/month): the monthly batch lane, with a freeze date, two Hotfix slots, F charged once per batch, one invoice.
  • Advisory (from NT$25,000/month, 4 hours): for clients with their own IT — intake governance and driving F down.
  • Full outsourcing: we absorb F and keep automating it, for companies with no IT.

Wrong fit: fewer than 12 requests a year and all genuinely urgent (no batch base); daily-cadence operations (build in-house); nobody internally holding prioritisation authority.

The transition playbook

Month 1 — inventory and definition

  • Pull 12 months of tickets and compute your real n, F, V and efficiency ratio
  • Define the four lanes and SLAs in the maintenance addendum
  • Pick the window (freeze second Wednesday, deliver week 3) and build the intake form

Months 2–3 — run two batches

  • Complete two full batches with no exceptions
  • Build a minimum automated regression suite covering the top 10 flows
  • Exercise the Hotfix lane to prove the SLA; move to monthly invoicing and one acceptance meeting

Day 90 review

Measured F vs baseline (20% lower), batch hit rate (≥ 80%), queue jumps (≤ 2 per month), average lead time, defects per batch, cost per request.

Decision checklist

  • ☐ How many tickets last year?
  • ☐ Average cost and hours per ticket?
  • ☐ Is that average under 3.5 hours?
  • ☐ A size-independent base fee in the quote?
  • ☐ A separate acceptance meeting per ticket?
  • ☐ A separate invoice per ticket?
  • ☐ Is "urgent" defined in writing?
  • ☐ More than 24 queue jumps last year?
  • ☐ Automated regression tests?
  • ☐ One-click deployment, or manual?

Fewer than six ticks means you are paying the same fixed cost repeatedly.

FAQ

Will genuinely urgent work get stuck once we batch?

No, provided the Hotfix lane is defined first: batching replaces "everything is urgent" with "only these four categories are urgent." Once those four are in the addendum, queue jumps usually drop from 5–8 a month to 1–2.

Our vendor charges a base fee on every ticket. Is that fair?

Fair, but it must be transparent — the base fee is F. What is not fair: components that cannot be explained, charging per ticket after merging, and a half-day minimum with no willingness to consolidate.

Is a monthly retainer the same thing as batching?

No. A retainer is a pricing model; batching is intake discipline. With no freeze date, no intake form and no Hotfix definition, F still eats 35–45% of the hours.

Want to know what your F actually is?

Send us last year's ticket list. In a free 30-minute consultation we will calculate your F, efficiency ratio, batching benefit and a first-draft four-lane split. If the maths does not favour batching, we will say so.

Share: