The first serious target for the builder shift is not the polished SaaS stack.

It is spreadsheet ops.

Every company has them: business-critical processes running through spreadsheets, copy/paste work, manual lookups, Slack nudges, CSV exports, calendar reminders, and one person who knows which tab is the real one.

These systems are usually not called systems. They are called trackers, models, templates, logs, planning docs, handoff sheets, or "the file." That naming hides their importance. In practice, many are production workflows without product ownership.

AI makes these workflows easier to replace, but only if operators stop treating them as embarrassing messes and start treating them as maps of unmet software demand.

Spreadsheet ops are signals

A spreadsheet process usually means one of three things.

First, the existing SaaS tool does not support the workflow well enough. The team has created a sidecar system to compensate.

Second, the process changes too often to justify traditional software work. The spreadsheet gives the team flexibility while the operating model is still evolving.

Third, nobody has taken ownership of the workflow as a product. The business tolerates the manual load because it is distributed across people and meetings.

All three are useful signals.

The mistake is jumping straight from "spreadsheet" to "app." Some spreadsheet ops should become automations. Some should become better reports. Some should become vendor configuration. Some should be killed because the process itself is bad. Some should stay as spreadsheets because the volume is low and the cost of productizing them would exceed the drag.

Look for repeated judgment, not just repeated tasks

The best targets are not merely repetitive. They combine repeated steps with repeated judgment.

Examples:

  • a sales ops analyst reviews account changes, applies territory rules, and updates CRM fields;
  • finance reconciles invoice exceptions and decides which ones need escalation;
  • customer success flags renewal risk from usage, sentiment, support history, and executive sponsor changes;
  • product ops gathers launch readiness across teams and identifies blockers;
  • recruiting coordinators check candidate stage, interviewer availability, and follow-up quality.

These workflows are painful because the logic lives in human heads and scattered artifacts. AI can help extract, summarize, classify, draft, route, and recommend. Traditional automation can handle deterministic steps. A small internal interface can give users a better way to act.

That combination is usually more valuable than another dashboard.

Start by documenting the actual workflow

Do not begin with a requirements meeting.

Watch the work.

Document:

  • trigger: what starts the process;
  • inputs: systems, documents, messages, files, human knowledge;
  • decisions: what judgment is applied;
  • outputs: records, approvals, tasks, communications;
  • exceptions: where the process breaks;
  • frequency: how often it happens;
  • volume: how many cases;
  • risk: what happens if it is wrong;
  • users: who touches it and who depends on it;
  • owner: who is accountable after launch.

This is not requirements theater. It is operational archaeology.

Replace the workflow in layers

A good spreadsheet-ops replacement usually has layers.

First, stabilize the data. Stop relying on pasted exports if an API or scheduled sync can provide the same input.

Second, separate deterministic logic from judgment. Rules, validations, calculations, and routing should not be hidden inside a prompt if they can be explicit.

Third, use AI where ambiguity exists: summarization, classification, extraction, comparison, drafting, and recommendation.

Fourth, create a review surface. Users need to see evidence, override decisions, and correct errors.

Fifth, write back carefully. Updating systems of record is higher risk than producing an internal recommendation.

The operator's rule

Do not ask, "Which spreadsheet can we automate?"

Ask, "Which spreadsheet is secretly operating the business?"

The best early wins are workflows with real volume, visible pain, clear ownership, bounded risk, and enough repetition to justify a maintained internal tool.

Spreadsheet ops are not a shame file. They are the backlog.