AI agents services pricing models compared (subscription, outcome, project)

AI agents services pricing in 2026 is a moving target. Two years ago every vendor charged hourly because no one knew what the work was. Today the better vendors charge flat per workflow because they have shipped enough to predict the cost. The transition is happening at different speeds across categories — boutiques have moved, consultancies have not, freelancers vary widely.

This is the companion to our pillar guide on how to choose an AI agents services company. We covered the basics there; this article goes deeper on each pricing model, with the trade-offs and the failure modes you should expect.

Model 1: Productized subscription (flat monthly per workflow)

The dominant model for managed AI agents services in 2026. You pay a fixed monthly fee per workflow regardless of volume (within a usage envelope). The fee covers platform, agent runs, model costs, operator time, weekly review, monthly tuning.

Typical ranges (EU, 2026). Light back-office workflow: €2,000–€3,500/month. Medium-complexity revenue or operations workflow: €3,500–€6,000/month. High-touch revenue-side workflow (outbound prospecting at scale, complex content production): €5,000–€8,000/month. Bundles of 2–3 workflows usually price at a 10–20% discount.

What it includes. All agent runs within a stated cap (e.g. "up to 5,000 runs/month"; overages billed at a published rate), model inference costs, operator time, weekly written summary, monthly business review, integration maintenance, evaluation suite upkeep, observability portal access.

Why it is well-aligned. The vendor's incentive is to keep the workflow running well so you do not churn. Their margin comes from running it efficiently (model cost optimisation, prompt tuning, batch processing) — improvements that benefit you too. Cost predictability for you is high; you can put a single line in your annual plan.

What can go wrong. Volume creep — your workflow grows and the usage cap is breached, triggering overages or a renegotiation. Scope creep — you ask for "small tweaks" that accumulate into a custom workflow not covered by the subscription. The defence is a clean scope of work in an exhibit, and an honest renegotiation when the workflow has genuinely changed.

Model 2: Project (build-and-handoff or build-and-support)

One-time fee to scope, build, and ship a workflow. Sometimes followed by a smaller monthly retainer for support.

Typical ranges (EU, 2026). Boutique single-workflow build: €15,000–€80,000. Consultancy single-workflow build: €100,000–€500,000. Support retainer after go-live: €1,500–€5,000/month (boutique) or €10,000–€40,000/month (consultancy).

What it includes. Discovery, design, build, testing, deployment, training, documentation, sometimes 30 days of post-launch warranty. After warranty, you either go on the retainer or you operate it yourself.

When it fits. The workflow is one-off — a one-time data migration, a specialised research project with a defined endpoint, a custom integration with a system the vendor's productized service does not support. You have an internal team capable of operating the workflow after handoff.

When it does not fit. The workflow is ongoing operational work. Project pricing for ongoing work creates the wrong incentive: the vendor is paid to deliver, not to keep it working. The fastest way to finish a project is to ship something that passes acceptance criteria and then never look at it again. A small post-launch retainer barely changes that calculus.

Model 3: Outcome-based (pay-for-results)

The vendor takes a share of incremental value: a bounty per qualified lead, a percentage of collected invoices above a baseline, a fee per resolved support ticket above a threshold.

Typical ranges (EU, 2026). €30–€200 per qualified lead, 5–15% of incremental collected invoices, €1–€5 per resolved ticket above baseline. Almost always paired with a monthly floor of €1,500–€5,000 so the vendor does not abandon you in slow quarters.

Why it sounds beautiful. Perfect alignment in theory: the vendor only wins if you win. No risk for you on slow months. Vendor is incentivised to optimise relentlessly.

Why it breaks in practice. Attribution disputes are constant. "Was that lead the agent's or the form's?" "Was that invoice paid because of the dunning sequence or because the customer was going to pay anyway?" The baseline shifts as your business changes and the vendor wants it locked, while you want it floating. The vendor cherry-picks easy clients and you become the harder one. By month nine you are renegotiating every quarter and the relationship is mostly meta-conversation about the contract.

When it fits anyway. A narrowly instrumented funnel with no other touchpoints (rare). A collections workflow where the baseline is genuinely measurable (more plausible). A specialised content workflow where output is the unit of value and quantity is the only dimension. In every other case, treat as a trap.

Model 4: Hybrid (subscription plus small variable)

Flat base fee plus a small performance kicker tied to one measurable outcome. Becoming the dominant model among the better boutiques because it gets the best of subscription (predictability, alignment) without the worst of outcome (attribution wars).

Typical structure. €3,000–€5,000/month base. Plus €30 per qualified opportunity above a baseline of 30/month. Or €500 bonus per quarter for hitting a single KPI (e.g. DSO reduction). Variable component caps at 20–30% of base to avoid attribution disputes becoming the whole conversation.

Why it works. The base covers the vendor's cost and gives you predictability. The variable signals to the vendor what you care about and gives them upside for delivering it. The cap on variable keeps the conversation healthy when the metric moves for reasons neither party controls.

What to watch for. The variable metric should be one number, not three. The baseline should be agreed in advance and reset annually, not negotiated every quarter. The attribution rule for the variable should be simple enough to explain in one sentence — "any opportunity created in HubSpot tagged ai-agent in the source field" — or the dispute curve is inevitable.

Hidden costs to add to every quote

Regardless of pricing model, the published number is rarely the all-in cost. Budget for these.

Your internal owner's time. 2–6 hours per week. At a fully loaded €80/hour, that is €1,000–€2,000/month in opportunity cost.

One-off integration work. €3,000–€15,000 in year one for the connectors your stack does not have out of the box. Some vendors absorb this; many do not.

Model API overages. If the subscription has a usage cap and you exceed it, overages can be 30–50% of the base fee in a heavy month. Ask for the cap and the overage rate up front.

Migration cost on exit. If your workflow is portable, near zero. If not, 6–10 weeks of internal time plus a new vendor's onboarding fee.

What to ignore

Two pricing structures show up that you should treat as red flags rather than serious offers.

"Per seat" pricing for managed services. An AI agent is not a seat. Per-seat pricing imported from SaaS is either lazy or a way to inflate cost as you grow.

"Per token" or "per API call" pass-through with markup. If the vendor is charging you cost-plus on model usage with a multiplier, you are paying their margin on a commodity that is dropping in price every quarter. A flat fee absorbs that volatility and is more aligned.

How prices have moved in the last 18 months

Two trends are worth knowing as a buyer. First, model inference costs have dropped 40–70% on like-for-like quality since early 2024. Vendors that have not adjusted their flat fees in 12+ months are either pocketing the difference or running on margins thinner than they should be. Ask any vendor what model and inference cost lives behind their monthly fee — the answer tells you whether they are running a real platform or expensively reselling a model API.

Second, productized subscription pricing has compressed into a tighter range as the category has matured. A workflow that cost €8,000–€12,000/month from a boutique in 2024 is closer to €4,000–€7,000 in 2026 for the same outcome. Vendors still charging the older numbers should justify it with platform features or domain depth — sometimes they can, often they cannot.

How to compare quotes apples-to-apples

Normalise every quote to an annual all-in cost for a single workflow at expected volume. Include everything: subscription, project fee amortised over 12 months, expected variable, integration work, sub-processor pass-throughs, your internal owner time.

Then divide by a unit that matches the workflow: per qualified lead, per processed invoice, per resolved ticket, per article published. The cost-per-unit comparison is much more honest than the headline monthly fee and usually narrows the field quickly.

For the deeper framework on pricing and ROI, see Managed AI agents pricing guide and AI agents ROI calculation.

Where Logitelia fits

Logitelia prices flat per workflow with no per-seat tax, no per-token markup, and a clean overage rate when volumes exceed the cap. For complex engagements we offer a hybrid structure with a small KPI-tied variable, capped at 25% of base. If you want to compare a Logitelia quote against your other options, book an intro call — we will share the line items and the cost-per-unit math.

For the broader framework, read the pillar How to choose an AI agents services company in 2026. For the deliverables side of the conversation, see What does an AI agents services company actually deliver?.