Skip to content

AI Agent Implementation Cost for Companies: 2026 Cost Guide

CFOs and CTOs need numbers. This guide covers small and mid-size operations on Tolky's published plans and enterprise operations on request, with explicit assumptions and a 24-month TCO.

Marlos Carmo

Founder of Tolky

Updated on November 15, 2026

10 min read
AI Agent Implementation Cost for Companies: 2026 Cost Guide

Share

The question CFOs, CEOs and procurement managers ask most when evaluating AI platforms for customer service is direct: how much does it cost? And the answer they most often get is frustrating: "it depends". To understand the return and not just the cost, the AI automation ROI framework uses the same assumptions as this guide.

It depends on volume, integrations, flow complexity and support level. All of that is true, but it is also a way to avoid numbers that allow comparison. This guide does the opposite: it starts from what Tolky publishes at /precos, states every assumption and only uses market ranges when they are labeled as a scenario.

Hand-drawn cost and revenue chart: the cost of an AI agent must be read together with the return projection, not just the license

What Tolky publishes today

The plans and values below come from the pricing page (checked in November 2026; see the page for current values). The page shows amounts in reais only; in this guide, dollar figures use US$ 1 = R$ 5.50 as a reference rate (R$ 500 is about US$ 91; R$ 2,500 about US$ 455; R$ 1.00 about US$ 0.18; R$ 2.50 about US$ 0.45).

PlanMonthly feeConversations includedAdditional conversationNotes
IndividualFree150/monthnot stated1 user, web chat; no WhatsApp; no credit card
StarterR$ 500/month500/monthR$ 1.002 users, WhatsApp included
ProfessionalR$ 2,500/month1,000/month (2,000 option)R$ 2.5010 users, WhatsApp included
EnterpriseOn requestCustomOn requestIncludes, among others, custom reports and weekly Advisor notices

Points that change the math:

  • A conversation is a contact who sent at least one message in the month, whether resolved by AI or escalated to a human. Billing is per conversation, not per "resolved problem".
  • WhatsApp fees charged by Meta are outside the plan. Estimate them with the WhatsApp pricing simulator.
  • Tolky credits (used for features such as transcription and web search) are included in each plan and renew monthly: 100 on Individual, 500 on Starter and 1,000 on Professional.
  • Setup in minutes, with implementation support. The page does not publish a setup fee. Cancel anytime.

The six cost variables of any AI implementation

Whatever the vendor, the total cost of an AI agent in customer service has six variables. Ignoring any of them produces a budget that surprises you during the project.

1. Platform licensing

The most common pricing models in the market are per user/agent (predictable, but it does not reflect operations with lots of automation and few humans), per conversation (aligned with usage, but variable in seasonal peaks) and per resolution (aligned with value, but it requires a precise definition of "resolution", a common source of dispute). Tolky charges a monthly fee plus additional conversations, as in the table above.

2. Initial implementation

Includes configuring agents and flows, building the knowledge base, training the team and managing the project. On the published plans, setup is quick and the page lists no fee; what usually weighs is your internal time (organizing documentation, validating answers). As a project estimate, we use 60 internal hours at R$ 80/hour (R$ 4,800, one time, about US$ 870) in scenario A. In enterprise deployments with multiple agents and legacy systems, the effort rises a lot and is defined in a proposal.

3. Integrations with existing systems

This is where most surprises appear. As project estimates, with no external source: integrations with modern platforms and well-documented APIs usually take 1 to 2 weeks; legacy systems, 4 to 8 weeks, with cost mostly in engineering. Also keep a reserve for maintenance when integrated systems change.

4. Ongoing maintenance and management

An AI platform in production does not maintain itself: knowledge-base curation, review of conversation samples and flow adjustments. In this guide, we use R$ 4,000 per month of internal hours (assumption, about US$ 730). The vendor's support level also weighs in enterprise contracts.

5. Language model costs

Each interaction consumes language-model processing, and that cost can be built into the monthly fee or charged separately, depending on the vendor. Model prices change often, so this guide gives no per-interaction values: ask each vendor whether model cost is included and how it is billed. At Tolky, the pricing page states the credits included for features such as transcription and search; for the rest, ask the team.

6. Migration and exit

Before signing, ask: can the data be exported in a usable format? Can the knowledge base be migrated? What is the estimated cost of switching? Portability terms are a contractual matter: ask the sales team what applies to your case.

Scenario A: small or mid-size operation on the published plans

Illustrative scenario (not a promise of results). Assumptions: 6,000 tickets per month; human cost of R$ 18 per ticket (assumption, from an agent costing R$ 5,000 to R$ 6,900 per month handling 250 to 350 simple tickets per month, or R$ 14 to R$ 28 per ticket; R$ 18 is about US$ 3.30); Professional plan; Meta at R$ 0.50 per conversation (illustrative assumption, use the simulator); R$ 4,000 per month of internal hours; internal implementation of R$ 4,800.

ComponentPer month12 months24 months
Professional plan: R$ 2,500 + 5,000 additional × R$ 2.50R$ 15,000R$ 180,000R$ 360,000
Meta fees (6,000 × R$ 0.50, assumption)R$ 3,000R$ 36,000R$ 72,000
Internal hours (assumption)R$ 4,000R$ 48,000R$ 96,000
Internal implementation (60 h × R$ 80, one time)R$ 4,800R$ 4,800
TotalR$ 22,000 (about US$ 4,000)R$ 268,800 (about US$ 48,900)R$ 532,800 (about US$ 96,900)

The return comes from avoided human cost: 6,000 tickets × R$ 18 = R$ 108,000 per month at baseline. With a deflection ramp in the first two months (25% and 40%) and then stable:

Deflection after the rampHuman cost avoided in 24 monthsTotal cost in 24 monthsNet result24-month ROI
35% (conservative; ramp 15%, 25%)R$ 874,800R$ 532,800R$ 342,00064%
52% (base; same as the internal figure at 90 days)R$ 1,305,720R$ 532,800R$ 772,920145%
65% (optimistic; ramp through month 6, same as the internal figure at 6 months)R$ 1,589,760R$ 532,800R$ 1,056,960198%

Monthly break-even is a deflection of about 20% (R$ 22,000 ÷ R$ 18 = 1,222 tickets). Not included: lower AHT on tickets that stay with humans, avoided churn and the loss from reopens (if 10% of deflected tickets go back to humans, the base scenario loses about R$ 5,600 per month). The ROI framework details those blocks.

Smaller operation (Starter). With 1,500 conversations per month, Starter would cost R$ 500 + 1,000 additional × R$ 1.00 = R$ 1,500, plus R$ 750 of Meta fees (assumption) and R$ 1,000 of internal hours, or R$ 3,250 per month. With 52% deflection and R$ 18 per ticket, avoided human cost is 780 × R$ 18 = R$ 14,040, and the net result is about R$ 10,790 per month. Check current terms at /precos before using these numbers.

Scenario B: enterprise on request (market ranges, not Tolky prices)

For higher-volume operations, deep integrations or corporate requirements such as SAML SSO, the Enterprise plan is on request. The ranges below are enterprise project estimates from the market, with no external source and unrelated to Tolky's price table; they help you understand the structure of a TCO, not compare prices:

ComponentYear 1Year 224-month total
Licensing (enterprise platform, example)R$ 240,000R$ 252,000R$ 492,000
Initial implementationR$ 80,000R$ 80,000
Integrations (4 systems)R$ 40,000R$ 8,000 (maintenance)R$ 48,000
Language modelsR$ 36,000R$ 40,000R$ 76,000
Internal management (20% of 1 senior analyst)R$ 36,000R$ 38,000R$ 74,000
Premium supportR$ 60,000R$ 60,000R$ 120,000
TotalR$ 492,000R$ 398,000R$ 890,000 (about US$ 161,800)

In this example, the license is 55% of TCO (492 ÷ 890), and evaluating platforms by license alone understates cost. Larger enterprise contracts can have licenses that alone exceed R$ 900,000 per year; the range of a real enterprise project depends on a proposal.

Return in scenario B, with the same R$ 18 per ticket base. An operation of 5,000 interactions per month and 52% deflection: 2,600 tickets × R$ 18 = R$ 46,800 per month, or R$ 1,123,200 over 24 months. The 2,400 tickets that reach humans have a 20% reduction in resolution time (assumption): 2,400 × R$ 18 × 20% = R$ 8,640 per month, or R$ 207,360 over 24 months. Total return is R$ 1,330,560 against a TCO of R$ 890,000: net result of R$ 440,560 and a 24-month ROI of about 50%, not counting avoided churn. It is lower than scenario A because the platform cost assumption is much higher; that is why comparing scenarios requires the same cost-per-ticket base.

Warning signs in commercial proposals

  • Pricing by license only, with no clarity on language-model cost. Ask before signing. The guide on how to choose an enterprise AI automation platform has the full checklist.
  • "Free" or very cheap implementation can mean something important (knowledge base, flows, training) is not included. The cost shows up later in internal hours or poor answers.
  • No SLA for integration maintenance.
  • A contract with no mention of data portability. Resolve it before signing.
  • Resolution rate promised without context. "Resolves 80% automatically" needs context: in which operation, with which knowledge base, with which integrations? Ask for references with a profile similar to yours.

How Tolky positions itself on cost

Tolky's proposal is predictable cost per conversation: it starts free (the Individual plan, no credit card), has published plans with a monthly fee and conversations included, setup in minutes (no published setup fee) and scales to Enterprise on request when the operation requires it. Meta's costs are charged separately by Meta and can be estimated in the simulator.

Tolky internal data (period, sample and definition available on request): average deflection of 52% in the first 90 days, 65% after six months and median payback of 47 days. Scenario A above is a simulation with explicit assumptions, not an average of customers.


The question "how much does it cost?" has an honest answer in two parts: for small and mid-size operations, it fits in a public plan table plus Meta's fees; for enterprise operations, it depends on a proposal, and the TCO must be built with the real data of your environment.

Want to calculate your case? See the plans, estimate Meta fees in the WhatsApp pricing simulator and measure your operation in the agent efficiency test. For Enterprise, talk to our team.

Share

Tags

  • AI agent implementation cost
  • enterprise AI platform price
  • how much does AI automation cost
  • corporate AI budget
  • B2B intelligent agent investment

About the author

Marlos Carmo

Founder of Tolky

Marlos Carmo is an AI entrepreneur and founder of Tolky, the conversational-era infrastructure and AI CRM that unifies intelligent service, multi-channel support (such as WhatsApp and voice), live CRM, and operational intelligence in a single ecosystem. He is a finalist for the SXSW Innovation Awards and a member of Francesco's Economy, a global network of young entrepreneurs focused on innovation and social impact. He works connecting Artificial Intelligence and digital transformation in projects for large organizations.

Next step

Want to bring this to your operation?

Tati, Tolky's AI assistant, understands your scenario in a few questions, suggests the best path and connects you with someone from the team, with full context.

Read also