Every automation business case has a launch date. Almost none have a maintenance line. The omission isn't malicious — it's inherited from how companies think about projects: build, ship, done. But workflow automations live inside a moving environment, and the environment doesn't sign off on your project plan.
01What actually breaks
- ▸Upstream drift. The CRM adds a required field. The vendor portal changes its export format. The email template the extraction relied on gets redesigned by marketing. Nothing announced; things just stop parsing.
- ▸Volume shift. The workflow that ran forty times a week runs four hundred after the new product launches. Queues back up, rate limits bite, the review pattern that worked at low volume becomes a bottleneck.
- ▸Model updates. Providers deprecate versions and upgrade defaults. Usually outputs improve; occasionally a prompt that relied on specific behavior needs retuning. Without evals, you find out from users.
- ▸Process evolution. The business changes the workflow itself — new approval step, new region, new pricing tier. The automation encodes the old process until someone updates it, and every day between is quiet wrongness.
- ▸Knowledge staleness. Retrieval-based systems degrade as their source documents age. The assistant answering from the 2024 policy manual isn't broken — it's worse: confidently outdated.
02The real cost shape
The good news: maintenance for a well-built workflow automation is not a full-time job. It's a rhythm — the weekly review (fifteen minutes), a monthly look at the metrics, and a few hours when something upstream changes. The bad news: it's a rhythm someone must own. Unowned, each small break becomes permanent, users quietly revert to manual, and six months later the automation is a line item nobody can explain.
Decay is rarely dramatic. It's one team member finding one workaround, once. Then it's culture.
03Three ownership models
- ▸Internal ownership. Your team runs the rhythm. Works when you have engineers with capacity and context — the catch is that automation upkeep loses every priority fight against the product roadmap, and "we'll get to it" is where automations go to rot.
- ▸Continuous operation. The builder stays attached — monitoring, drift review, and improvement on a rhythm. Costs more than abandonment and less than rebuilding; this is the model Taldyn runs, because it's the only one we've seen compound instead of decay.
- ▸The handoff with teeth. If a true handoff is the plan, it needs: documented failure modes, runnable evals, alerting the receiving team actually watches, and a named owner with maintenance hours in their actual schedule. A handoff without those isn't a transfer of ownership — it's an abandonment with paperwork.
04The quarterly health check
Between the weekly rhythm and the incident play sits the quarterly health check — thirty minutes per automation, four questions, on the calendar or it doesn't happen.
- ▢Usage: is volume tracking the business — and if the business grew while runs stayed flat, who is quietly working around the system, and why?
- ▢Quality: what do the override rate and the miss log say this quarter versus last — improving, stable, or drifting?
- ▢Dependencies: what changed upstream this quarter — formats, APIs, models, the process itself — and did the automation get updated or just survive?
- ▢Verdict: keep, change, or kill — decided explicitly, with the owner's name on the decision. Zombie automations that nobody uses but everybody maintains cost more than their invoice; they cost the credibility of the next build.
05Budgeting it honestly
Whatever the model, put a number on it before launch. A serviceable heuristic for workflow automations: expect ongoing attention measured in hours per month, not per week — but never zero. Write it into the business case next to the build cost. An automation with a maintenance line survives budget season; one without becomes "that thing that broke" in eighteen months.
OPERATOR NOTE — Nobody budgets for watering the plants, either. That's why the office ficus is plastic. Decide which kind of automation you're buying.
Put this thinking to work.
A 30-minute strategy call with an operator — we'll map your first deployment path, not send a deck.