Service 04 · Keep it running

AI workflow maintenance and optimization

Maintenance, quality tuning, and improvement work, with someone accountable when the output looks off.

Book a 30-minute call
Why this exists

Why AI workflows need maintaining

An AI workflow is not something you install and leave. Ordinary changes move its output.

New client types arrive with documents laid out differently.

The workflow was tuned on what came before.

Suppliers change their forms.

Extraction that was reliable starts missing a field.

The team starts using it for adjacent work.

Work it was not tested on.

Model providers release updates.

Output that was consistent shifts in tone or structure.

Your own standards change.

What counted as a good draft last year is not what you want now.

None of these stop it working. They move the output gradually — from reliable, to occasionally needing an edit, to needing a careful check every time. At that point people quietly return to doing the work by hand, and the conclusion becomes that AI did not suit the business.

Maintenance is what prevents that.

Levels

Three levels of AI workflow support

LiteStandardAdvanced
CoversA single workflowA sequence several people depend onWorkflows the business runs on
Fixes and quality tuningIncludedIncludedIncluded
Monthly written check-inIncludedIncludedIncluded
Monthly review call—45 minutes45 minutes
Quality sample review—MonthlyDeeper, tracked over time
Continuous improvement work—Minor changesAgainst a prioritized backlog
Performance reporting——Monthly
Multi-team support——Included
Quarterly review——Included
Response priorityStandardStandardPriority
The limits

Outside this engagement

  • New workflows — quoted as implementation work.
  • Rebuilds and redesigns.
  • Out-of-hours or on-call response.
  • Unlimited changes.

Anything outside scope is quoted before the work is done, never after.

Terms

Monthly and rolling. Thirty days’ notice either side. Hours reset monthly and do not carry over. Available after an implementation. Move between levels at any point, effective the following month.

If several months pass with little to do, we will say so and suggest moving down a level.

Common questions about AI workflow maintenance

Why does an AI workflow need maintaining at all?
Ordinary changes move its output. New client types arrive with documents laid out differently, suppliers change their forms, model providers release updates, and your own standards shift. None of these stop it working — they move the output gradually from reliable to needing a careful check every time.
Can we start a retainer without having had an implementation?
The retainer is available after an implementation. Maintaining a workflow we did not build means learning it first, which is scoped separately.
What happens if there is very little to do one month?
We will say so and suggest moving down a level. Hours reset monthly and do not carry over.
How quickly do you respond when something looks wrong?
Standard response on Lite and Standard, priority on Advanced. Out-of-hours and on-call response are not included at any level.
Can we change level?
At any point, effective the following month. Monthly and rolling, thirty days’ notice either side.
Next stepTell us what your team keeps redoing.Thirty minutes, no preparation needed. We will tell you honestly whether AI would help or whether simpler automation would do the job.
Book a 30-minute call