Blog
Perspective

When a Business Function Becomes a Software Owner

Business teams are using AI to replace parts of expensive vendor tools. An app can work well and still be hard for the team to own.

We’re seeing more business teams use AI to build their own version of a vendor tool that costs the company six figures a year. Not the whole product. Just the part they actually use.

The build is the quick part. Here is what tends to happen next, as an illustrative composite. The names, company, and figures are fictional.

One app, one year

A 40-person advisory firm pays about $150,000 a year for a client reporting platform. The research team uses it for one job: quarterly reports that combine each client’s portfolio from the CRM with market data from a data provider. In March, Daniel, a senior analyst, spends two weekends building a replacement with an AI coding assistant and shares it with the team by link. The April reporting cycle takes two days instead of five. By June, the managing partner is asking whether to cancel the renewal.

In July, the data provider renames the field that holds benchmark returns. Daniel’s app can’t find it, so it falls back to the last value it saved. The app doesn’t crash. It produces a full set of clean, well-formatted reports, and the benchmark chart in every one shows last quarter’s numbers. The review before sending focused on layout and totals, and no one noticed that the benchmark chart used stale data.

A client spots it before the firm does. The fix takes twenty minutes. Working out which reports went out wrong, resending them, and calling each client takes a week. In October, Daniel is on leave for the next cycle, and nobody else knows how the app picks each client’s peer group, so the team leaves the setting alone and checks every report by hand.

In November, compliance asks two questions: who approved the change to how benchmarks are calculated, and who can see client portfolio data. The honest answer to the first is a chat thread. The answer to the second is anyone with the link.

Daniel was right to build it. The app is faster, and better suited to this workflow than the platform was. It works. The team just can’t own it yet.

What the app was missing

Each of those problems has a known fix. None of them is exotic.

What happenedWhat it neededIn practice
A renamed field filled every report with last quarter’s numbersCheck incoming dataBefore any report is built, reject a missing or unexpected benchmark field, confirm the data is dated this quarter, and never silently reuse stale values.
A week to find which reports were wrongKeep a record of each runSave the data pulled, the version of the app, and every report it produced.
Nobody could explain a setting while Daniel was awayWrite down the rulesDescribe how peer groups and benchmarks are chosen, in plain English, next to where they are set.
The approval record was a chat threadRecord who approved each changeTest each change on a known-good reporting period, and log who approved it and when.
Anyone with the link could see client portfoliosLimit who can see the dataCompany sign-in, with access limited to the people who work on each client.

What would have caught July

The app should reject a missing or unexpected benchmark field and never silently reuse stale data, and any change to the app should be tested on a known-good reporting period before release.

Every reporting cycle

Input CRM and market data
Validate inputs Benchmark field as expected, data current
Draft Client reports drafted
Send Owner reviews and sends
July stops here: missing benchmark field, stop and alert the owner.

When the app or its rules change

Test Rerun a known-good reporting period
Investigate Explain every difference
Release Owner approves the release
Two separate approvals: the owner sends each cycle’s reports, and separately approves each release of a change to the app.

Where WisBot comes in

Daniel’s app is worth keeping. What it needs is the work around it, and that is what WisBot takes on.

The code stays yours. WisBot copies the repository, with its history, into a new repository in the team’s own GitHub account and works on its own branch. The original is untouched.

Nothing changes without a yes. The owner agrees a plan before anything is built, and approves each release before it goes live.

For Daniel’s app, the plan would include checking incoming data and comparing results after changes. The July fix would be tested in a private preview against a known-good reporting period before release. The peer-group rules and run history would help another authorized team member operate the next cycle while Daniel is away. And when compliance asks who approved a change, the answer is a record, not a chat thread.

WisBot helps teams build and operate workflow-specific software, including systems they have already started. Read the docs to see how the process works.

Built something your team now depends on? Bring it into WisBot, or book a demo to talk it through.