Skip to content
AdmasJournal
All stories

Opportunity investigations

Before you build a business app: prove one workflow is worth fixing

A practical evidence process for agencies to choose one business workflow, establish a baseline, define a measurable pilot, and avoid building the wrong app.

“Build a CRM,” “make an invoice app,” or “add an AI assistant” sounds like a starting point. It is actually a proposed solution.

Before an agency estimates features or opens a coding tool, it needs a clearer answer: Which repeated business job is failing, and what observable result would make a change worthwhile?

This guide offers a proposed evidence process. It is not a universal formula, and it does not prove that a workflow will produce demand, savings, or profit.

Replace the app idea with a research question

Start with an assumption and rewrite it as something the team can investigate.

  • We need a CRM → Where does a real lead wait, disappear, or require duplicate entry today?
  • Invoicing is too slow → What happened in the last five completed invoices, and which step delayed issue or payment?
  • Customers want an AI host → Which approved questions repeat during a real session, who handles exceptions, and would a buyer run a second session?

The first two questions investigate work. The last one also separates an internal product idea from buyer evidence. A feature request, founder idea, or synthetic test phrase should remain a hypothesis until actual users support it.

GOV.UK's research guidance recommends turning unsupported assumptions and opinions into research questions. That reduces the risk of treating confidence as evidence.

Reconstruct the last real case

Ask the operator to walk through the most recent completed example. Avoid asking only what software they want.

Record:

  • Trigger: What event started the job?
  • Actor: Who performed each step?
  • Current path: Which messages, spreadsheets, calls, approvals, and systems were used?
  • Finished state: What proved the job was complete?
  • Wait or failure: Where did information stop, repeat, or require correction?
  • Exception: What happened when the normal route did not apply?
  • Consequence: What observable delay, rework, missed opportunity, or support burden followed?
  • Access: Which records, permissions, devices, or third-party systems would a pilot need?

Ask “how did you do that?” and “why did that step happen?” A stakeholder's first explanation can describe the official process; the recent case shows the process that actually ran.

Do not assume every human step is waste. A review may protect the company from an incorrect quote, duplicate payment, unsafe answer, or unauthorized action. Understand it before deciding whether software should remove, support, or preserve it.

Map the whole service, not only the screen

Capture four parts of the service:

  • User steps: What the customer or employee tries to finish
  • Visible service: Forms, emails, calls, pages, decisions, and status messages
  • Backstage work: Review, correction, routing, reconciliation, and exception handling
  • Support systems: Identity, database, files, billing, vendors, policies, and ongoing operations

Mark every handoff between people and systems. Those boundaries often explain missing information, duplicate entry, unclear ownership, and delays.

Establish the baseline

Before promising improvement, measure the current route where practical.

For a small workflow, the baseline might include:

  • completed cases in the last 30 or 90 days;
  • median and worst-case completion time;
  • cases returned for missing information;
  • manual corrections;
  • support contacts;
  • operator time for the normal and exception paths.

Use only data the organization can collect responsibly. Label estimates as estimates. A few examples can reveal the process, but they do not establish a reliable average.

GOV.UK's performance guidance recommends judging changes against a baseline and combining metrics with user research. One snapshot or a polished dashboard is not enough.

Define one complete pilot outcome

“Make it better” cannot pass or fail. Choose one result that both the current process and the pilot can produce.

Examples:

  • A complete inquiry becomes an approved quote with required items and a named reviewer.
  • Accepted work becomes an invoice whose issue and payment status can be found without searching separate spreadsheets.
  • A client request reaches the correct owner, with missing information identified before work begins.

For the selected outcome, record:

  • starting event;
  • finished state;
  • normal and exception routes;
  • required evidence;
  • maximum acceptable completion time;
  • authorized data and integrations;
  • the person who decides whether the pilot passes.

Digital.gov recommends connecting the intended outcome to quantitative or qualitative measures, a collection method, an owner, a benchmark, and a review cadence. Choose three or four measures the team can actually collect.

Use five gates before building

Record supported, uncertain, or blocked for each gate. Do not hide a critical blocker inside a total score.

1. Repetition

Does the job happen often enough to observe and improve? A rare job may still matter when failure is costly, but it needs a different justification and stronger controls.

2. Evidence

Can the team show recent cases, friction, alternatives, and a baseline? Opinions are leads, not measurements by themselves.

3. Bounded result

Can a small pilot finish one useful business outcome rather than stop at a mockup? If an inaccessible system or undecided policy blocks the finish, resolve that first.

4. Ownership

Who owns access, exceptions, corrections, support, and later changes? A buyer who cannot name an operator is not ready for handoff.

5. Continuing economics

Separate implementation from monthly database, model, messaging, payment, monitoring, and support costs. Compare those costs with an outcome the buyer can observe. Do not turn an estimate into promised savings.

The SBA adds the market questions: Is there demand? Which alternatives already exist? What do customers pay? Broad research can establish context; direct research is needed for audience-specific answers.

Proceed, research, or stop

Proceed to a bounded pilot when the team has:

  • a named operator and buyer;
  • recent examples of a repeated job;
  • one complete outcome;
  • authorized access to required systems;
  • a pass/fail acceptance record;
  • disclosed one-time and recurring costs;
  • an owner for exceptions and continuing support.

Research when important evidence is uncertain. Stop when the pilot cannot complete the business job, lacks an accountable owner, or depends on access the team does not have. Avoiding an unready build is a useful result.

Copy this evidence brief

Investigate a pilot for [named user] who needs to complete [one repeated job] when [trigger] occurs. Reconstruct at least [number] recent cases. Record the current steps, systems, completion time, corrections, exceptions, and observable consequence. The pilot begins with [starting state] and passes only when [finished state] is produced with [required evidence]. Preserve [human review or exception route]. Use only [authorized data and services]. State implementation limits, recurring services, operator duties, and exact acceptance checks. Do not add unrelated features before this workflow passes.

You can use that brief in Admas or another development process. Admas can preserve project context, edit and run the software, show verification evidence, and return portable source. Those are delivery capabilities. They do not prove that the chosen workflow matters, that customers will pay, or that the application will create profit.

The useful order is: question the assumption, reconstruct the work, establish a baseline, define one finished outcome, and let observed use decide what grows.

Method and evidence

Original proposed buyer-research method, reviewed November 16, 2026 against GOV.UK, Digital.gov, SBA and public Admas product documentation. CRM, invoicing and AI-host examples are illustrative hypotheses. No customer interviews, measured demand, application benchmarks, savings or profit results are reported.

Put the idea to work

Bring your next useful problem.

Describe a workflow, inspect the result, and keep improving it in Admas.

Start a project

Keep exploring.

All stories

A useful read, once a week

Get the Field Notes.

One investigation, a practical takeaway, and an idea worth testing. Free. Leave whenever you like.

Confirm your address before receiving editions. Privacy notice.