When NOT to hire an AI agents services company (honest list)

Every vendor's most useful article is the one that talks the reader out of buying. We turn down 15–20% of inbound at Logitelia and the reasons cluster cleanly. This article catalogues those reasons so prospective buyers can self-diagnose before any sales call. It is the companion to our pillar guide on how to choose an AI agents services company.

1. The process is not documented and not done consistently by humans

An AI agent automates a process. If the process exists only in three people's heads and each of them does it differently, the agent will codify the worst version — or it will produce an average that satisfies no one. The vendor cannot fix this for you; the workflow has to exist before it can be automated.

Fix first: spend 1–2 weeks documenting the current process, run it manually against the documentation for 2–3 weeks, and reconcile the gaps between what is written and what is done. Then talk to a vendor.

2. Volume is below 50 transactions per month per workflow

Below that threshold, the cost of building and operating the workflow exceeds the cost of a human doing it manually. The evaluation set is too small to be statistically meaningful — you cannot tell whether the agent is good or bad, you only know whether the last five runs were good or bad. The vendor's per-workflow fee divided by 30 transactions is a unit cost that no business case survives.

Exception: if the work is high-skill-low-frequency and the bottleneck is the human's availability rather than their hours, the unit-cost math may still work. Even then, often a specialist freelancer is the better answer.

3. No internal owner with 2–4 hours per week

The vendor cannot own an outcome inside your business without an internal sponsor who removes blockers, opens accesses, decides edge cases, and translates results to leadership. Engagements without an internal owner fail regardless of vendor quality. This is the failure mode that wrecked the 2018 RPA wave and it is repeating.

The internal owner does not need to be senior; they need to be reliable. A capable operations associate who shows up to the weekly review and has authority to make small decisions is better than a VP who attends quarterly.

4. Workflow is regulated to a point that every step needs a human signature

Some workflows — certain medical, legal, regulated-finance processes — require human sign-off at every step by statute. AI agents help on the margins (draft the document, surface the relevant case, prepare the file) but the savings are small relative to the build cost. Worth revisiting in 18–24 months as auditability of agent decisions matures.

This is a "not yet" rather than a "never." The economics will shift.

5. You are buying because the board pressed you to "do something about AI"

This budget goes to whichever vendor pitches best, the project produces a demo not an outcome, and the next year's board conversation is about why AI did not work. Worse, the failed project becomes the company's internal evidence that "AI does not work for us," delaying the real adoption by 18–24 months.

If you must spend the budget, spend it on internal AI literacy training and on a small pilot in a low-stakes workflow your team genuinely wants to fix. That at least produces learning. A vendor engagement built on board pressure rarely does.

6. The workflow changes weekly

An AI agent codifies a workflow. If the workflow itself is changing every week — because the product is in heavy flux, because the market is shifting your offer, because your team is still figuring out the right process — the agent will be perpetually out of date and the vendor's tuning capacity will be consumed by churn rather than improvement.

Wait until the workflow has been stable for 6–8 weeks. Then automate.

7. A SaaS tool already solves 80% of the problem

Many "we need a custom AI agent" projects could be a €99/month SaaS subscription. Before hiring a services company, audit the SaaS market for your problem. Modern category leaders in sales (Apollo, Outreach), support (Intercom Fin, Zendesk), finance (Ramp, Pleo), and content (Surfer, Jasper) ship strong AI features that cover the common 80%.

The right time to hire a services company is when you have hit the 80% with SaaS and need the remaining 20% built around your specific workflow. Buying services to do what SaaS already does is paying enterprise prices for commodity functionality.

8. The expected ROI is less than 2× annualised on fees

Managed AI services should return 3–6× annualised on fees within the first year for back-office automation, and often higher for revenue-side work. Below 2×, the business case is fragile — one slow quarter or one unexpected cost line and the engagement is in the red. The right answer is either a smaller scope (cheaper engagement, same ROI), a different workflow (higher ROI), or no engagement (the value is not there).

The framework for this math is in AI agents ROI calculation.

9. The vendor would need access to systems you cannot expose

If the workflow requires the vendor's agent to touch a system with extreme sensitivity that you cannot wrap in a sandbox — production payment rails, the master CRM with no audit trail, the EHR in a healthcare context — the engagement is harder than it needs to be. Sometimes the right answer is to spend 3–6 months building the access layer first (read-only sandbox, audit trail, scoped credentials), then engage. Sometimes the right answer is in-house only.

10. Your team is hostile or skeptical of AI

An AI agent only delivers value if humans on your team adopt it. If your support team will refuse to use agent-drafted replies, if your sales team will refuse to act on agent-scored leads, if your finance team will treat agent-coded invoices as suspect — the agent will run, the dashboards will look healthy, and nothing will change in the metrics that matter.

Fix first: do the change-management work. Bring the team into the design of the workflow. Show them where the agent is good and where the human still has to judge. Skip this step and the engagement produces beautiful logs and no business impact.

What to do instead, by case

For each case above, there is usually a better next step than "hire a vendor."

• Process not documented: spend 2 weeks documenting. Then come back.
• Volume too low: keep manual, or hire a fractional specialist 4–8 hours/week.
• No internal owner: assign one. Sometimes the answer is to hire one person, not a vendor.
• Heavily regulated workflow: pilot only the unregulated sub-steps, with a vendor that has done this before.
• Board-pressure buying: spend on AI literacy training and a small internal pilot instead.
• Workflow changes weekly: stabilise the workflow, then automate.
• SaaS solves 80%: buy the SaaS, layer services on the remaining 20% later.
• ROI below 2×: re-scope, re-target, or do not engage.

Vendors who push past these "no" signals are not your friends. Vendors who recognise them and tell you to come back later are the ones worth remembering when the time is right.

The "wait six months" diagnosis

Many "do not hire" diagnoses are really "do not hire yet." Six months of preparation often turns a doomed engagement into a successful one. The preparatory work that matters:

Document the current process. Two pages, with the inputs, the steps, the decision points, the outputs, the exception handling. This document becomes the brief for the vendor and the ground truth for evaluation.

Pick an internal owner and protect 2–4 hours/week of their time. The owner does not need to know AI. They need authority, attention, and reliability. Most companies that struggle with vendor engagements struggle because no one was given the slack to make it work.

Run the workflow manually against the documentation for 3–4 weeks. This finds the gaps between what is written and what is done. It also produces the dataset the vendor's evaluation suite will need.

Stabilise the workflow for 6–8 weeks. If the process is in flux, freeze it long enough that the vendor's first build does not become immediate technical debt.

Companies that do this preparation get a 2–3× better outcome from the same vendor engagement than companies that do not. The preparation is the engagement's hidden first phase.

Where Logitelia fits

Logitelia turns down inbound that fits the above descriptions. That is not virtue; it is self-interest — engagements that start broken tend to end broken, and that is bad for both sides. If you read this list and recognised your situation, the right next step is the work that comes before a vendor engagement, not the vendor engagement itself.

If you read this list and none of it applies, the pillar guide for the buyer's decision is How to choose an AI agents services company in 2026. For the no-code-vs-managed comparison see AI agents services vs no-code automation. If you want a 30-minute call to sanity-check whether your situation is "buy now" or "wait six months," book an intro call — we will tell you straight.