Engineering · · 4 min read · Lukas Ceponis

What a care retainer should include, priced from $1,500 a month

Most maintenance retainers in this industry are a promise shaped like a shrug: "we will be around if something breaks." The client pays monthly, nothing visible happens, and after a while the line item looks like waste. So they cancel, and the systems rot in exactly the way the retainer was meant to prevent.

We operate our own voice receptionist on our business line plus six production systems for clients, and that has taught us the problem runs both ways. Maintenance matters, and most retainers fail to prove it. So here is what a care retainer should contain, concretely enough that you can hold any vendor to it, us included.

Monitoring that pages a human

AI automations fail silently by default. A workflow that errors stops. A phone assistant whose telephony credential expired stops answering. Nobody gets a message unless someone built the message. The first thing a retainer must include is instrumentation: health checks on every workflow, error alerts routed to a person, and usage tracking so a sudden drop to zero runs (the classic quiet death) triggers an alert. If a vendor cannot tell you, specifically, how they would learn your system broke at 2 a.m. on a Saturday, they are waiting for you to notice.

A fix window in the contract

"We respond quickly" is a mood, not a term. A retainer states, in the agreement, how fast a reported or detected failure gets fixed. Ours is 24 to 48 hours depending on tier. The number matters less than its existence: a vendor willing to commit to a window has thought about their capacity, and a vendor who will not is telling you your outage queues behind whatever project pays them more this week.

One improvement a month

This separates maintenance from insurance. Systems that only get fixed decay relative to the business around them. The business adds a service, changes a policy, hires new staff, and the automation still reflects last year. A good retainer ships one deliberate improvement every month: a new intent the phone assistant handles, a workflow extended to cover a case that used to need a human, an eval set grown by last month's edge cases. Small and named. Over a year that is twelve compounding upgrades.

The one-page report

Almost everyone skips this, and it is the part that keeps retainers alive. Every month, one page in plain language: runs completed (calls answered, tickets triaged, documents produced), a conservative estimate of hours saved against doing the same volume by hand, what broke and how long the fix took, and what ships next. Incidents go on the page. An honest three-incident month builds more trust than a fictional zero.

It has to fit on one page that takes three minutes to read, because a dashboard link does not get opened.

Quiet looks like dead

Businesses switch off working systems because nobody shows them the systems are working. An automation answers a few hundred calls a month without fuss, and from the owner's chair fuss-free is indistinguishable from broken. When budget review comes the invisible line item loses. We have watched profitable automations get cancelled for this reason and the fix was never technical. The report makes the value visible on a monthly rhythm so the renewal decision is made against evidence, which protects the client from paying for nothing and protects a decent vendor from being fired for doing the job quietly.

What it costs, and when to skip it

Our care plans are $1,500, $2,500 and $4,000 a month by tier, inside a market that runs $500 to $8,500 a month across the 20 agency price lists we compared. Every tier includes the four pieces above: monitoring with alerts, the 24 to 48 hour fix window, one shipped improvement a month, and a monthly report the owner actually reads. Tiers change response speed and how many systems are covered.

A retainer is a waste of money in two situations, and we say so on the scoping call. If you have a single workflow with no AI in it, the 30-day bug warranty on the build plus one monitoring alert is enough, and paying $1,500 a month to watch a Zap is silly. If you have a technical person in-house who is willing to own the system, buy a few hours of their time for the runbook instead. Care plans pay off when the system talks to customers, changes often, or would cost you money by Tuesday if it stopped on Saturday.

Two questions for any vendor

  1. "Show me a sample monthly report." A vendor who produces them has one within reach and will share a redacted copy. One who has never been asked will improvise, which tells you the reporting does not exist.
  2. "What happened the last time something broke at night?" Listen for a specific story: what alerted them, when, what the fix was, what they told the client. Vendors with monitoring have these stories on tap. Vendors without it describe their process in the hypothetical tense.

Unmaintained automations do fail, and the expensive kind of maintenance is the emergency kind, by which point the question has become whether to fix or rebuild the thing at $750 to $10,000 instead of a monthly fee.

If you have automations in production and nobody can currently answer those two questions about them, that is the gap our Care retainer is built to close.

Have a system that needs this treatment?