Automation field notes

Build, Buy, or Use the Software You Already Have?

Compare a simpler manual process, an existing software feature, a connector platform and custom code on first-year cost, review labor and who will operate the workflow.

The short answer: use the smallest sufficient path

Do not commission a new integration until a simpler process and the features in software you already pay for have been tested on the same real task. Compare four paths: simplify the manual handoff, use a native feature, connect applications through a workflow platform, or build custom software. A connector is an integration, not a substitute for choosing an owner; custom code is justified only when the simpler paths cannot meet a required control or when its measured benefit exceeds its operating cost. The worked example below chooses an existing feature and declines both a connector and a custom build.

  • Write the required outcome and prohibited actions before evaluating products; a draft bill for human review is not permission to pay it.
  • Reject any path that cannot meet required access, approval, audit or recovery controls, even if its estimated cost is lowest.
  • Compare the same annual volume, employee cost and review policy across every path.

Define one workflow and check the native capability first

For an illustrative accounts-payable decision, assume 20 incoming bills per week for 50 working weeks: 1,000 bills a year. The desired result is a reviewable bill record; an AP employee still checks vendor, amount, invoice number and duplicates before approval. Intuit documents bill image/PDF upload into QuickBooks Online's For review flow, where a person can verify extracted details. That does not establish that the feature is enabled in every account, that every invoice arrives in a supported format, or that duplicate controls and approvals meet this buyer's needs. Test representative bills in the actual subscription before assigning the native row a zero incremental license cost.

  • Manual simplification: require a consistent supplier submission channel and a complete bill before an employee enters it.
  • Existing feature: upload supported files and review the proposed bill inside the accounting system already in use.
  • Connector: move files or fields between systems, then stage a bill for the same AP approval.
  • Custom build: own a bespoke intake, mapping and exception path when supported features cannot satisfy a necessary requirement.

Further reading: Review Intuit's current bill-upload instructions·Invoice automation purchase decision

Cost the same year, not just the subscription

Use first-year total cost = setup labor or fee + annual human handling and review + 12 months of incremental runtime or licensing + annual monitoring, fixes and support. Handling time is included in every option, including the manual baseline. Setup includes testing, access, training and the fallback; support includes connector breakage, changed fields and incident investigation, not ordinary per-bill review. For this purely hypothetical calculation, use a $30 loaded employee-hour, not a GLCO rate or a vendor quote. Baseline entry and correction take an assumed six minutes per bill: 1,000 × 6 ÷ 60 × $30 = $3,000 a year. Labor reductions represent potentially released capacity, not automatic cash savings.

  • Price a required software-plan upgrade as incremental runtime even when the company already subscribes to the product.
  • Estimate platform usage using actual steps and monthly volume; Zapier describes tasks, Make credits and n8n hosted plans workflow executions, so their allowances are not interchangeable.
  • Keep an expected correction or loss allowance separate when failure severity differs between paths; do not bury consequential mistakes in an average handling minute.

Further reading: Existing automation cost guide·Existing ROI method

Worked first-year comparison: the native feature wins

Every number in this table is an invented planning assumption for the same 1,000 bills, not a current product rate, implementation offer or observed GLCO project. Manual simplification takes four minutes per bill after a two-hour setup; the native feature takes three minutes after six hours; a connector takes two and a half minutes after 24 hours; custom code takes two minutes after 80 hours. Runtime budgets of $20 and $25 monthly in the last two rows are placeholders to replace with actual quotes and measured usage. Support hours are two, 12 and 24 respectively for the native, connector and custom paths; manual simplification has no separate support allowance here.

Hypothetical first-year dollars at 1,000 bills and $30 per employee-hour; runtime figures are assumed budgets, not vendor prices
PathSetupAnnual handling or reviewAnnual runtimeAnnual supportYear-one total
Simplify manual intake$60$2,000$0$0$2,060
Use existing bill feature$180$1,500$0 assumed$60$1,740
Connector platform$720$1,250$240 assumed$360$2,570
Custom build$2,400$1,000$300 assumed$720$4,420
  • Manual: $60 setup + $2,000 handling = $2,060, saving $940 of assumed capacity versus the original process.
  • Native: $180 setup + $1,500 review + $0 incremental runtime + $60 support = $1,740, the lowest modeled total.
  • Connector: $720 setup + $1,250 review + $240 runtime + $360 support = $2,570.
  • Custom: $2,400 setup + $1,000 review + $300 runtime + $720 support = $4,420, or $1,420 more than the original $3,000 process.

What would overturn that no-build decision?

The native option is cheaper here because its extra minute of review relative to custom costs only $500 annually, far less than the custom path's additional setup and support. In the second year, without repeat setup and assuming unchanged volume and costs, the modeled native path is $1,560 and custom is $2,020. A necessary second-system write, unsupported document type, missing approval control or mandatory audit requirement could nevertheless make the native path ineligible. Then document that failed test and compare a supported connector with custom code; do not assume custom wins because it saves the most minutes per bill. Conversely, if the native workflow requires a costly plan upgrade or more correction than modeled, put those actual costs into its row.

  • Run 10–20 permissioned, representative bills through the candidate feature and time correction and approval, including duplicate and unreadable-file cases.
  • Calculate the extra annual benefit needed to recover any higher setup plus recurring cost over the expected ownership period.
  • If volume is low or native review remains near manual effort, retain simplified manual intake rather than paying for a new integration.

Separate software execution from the accountable operator

A platform can move data but cannot decide whether a suspected duplicate is safe to post. Name an internal AP approver and a technical operator for each path, decide who receives alerts and who checks the source document after a failed run, and require a safe retry rule that cannot create a second bill. A connector adds dependencies on two systems and their permissions; custom code adds hosting, deployment, monitoring and developer continuity. If the provider builds either route, specify which support incidents it handles and which business decisions remain with your employee. No path should silently convert a reviewable draft into a payment instruction.

  • Failure boundary: on missing vendor, uncertain amount or duplicate number, hold for AP review rather than writing an authoritative record.
  • Recovery boundary: retain the original file, a transaction identifier and a manual entry route when a destination or integration is unavailable.
  • Ownership boundary: the business controls production accounts and billing; a contractor's build fee does not buy indefinite operations.

Further reading: Handoff and export drill

Turn the comparison into a scoped decision

Vendor features, plan availability and usage rates can change. Intuit's support page describes account availability limits, while the provider pricing pages below describe different billing units; these primary pages were checked on September 24, 2026. The dollars and minutes in this article are hypotheses, not published rates or measured customer outcomes. Replace them with your subscription terms, sample-run usage, timed review, failure costs and expected term. If the smallest sufficient path is the native feature or a cleaner manual process, that is a valid decision with no new build. If a required gap remains, bring the failed test, volumes and approval boundary to GLCO for a scoped conversation at /contact; no pilot fee, timeline or result is promised here.

  • Bring a redacted example input and the expected approved output.
  • Bring the current subscription and any proposed plan upgrade or platform estimate.
  • Name the employee who will approve records and the person who will respond to an incident.

Further reading: Discuss a bounded workflow with GLCO·Read the process service scope

Helpful sources