A headline price for AI implementation is hard to interpret without its scope. A document assistant, a voice workflow and a system that writes to several business tools have different integration, testing and support needs. This guide explains the cost components, the assumptions a quote should state and a way to model payback with your own data.
How the main solution types compare
A comparable quote separates discovery, build, integration, deployment, handover and operation. The table describes cost structure, not Ace Digital pricing or a market benchmark.
| Solution | Cost pattern | Inputs a quote should state |
|---|---|---|
| FAQ chatbot | Build, hosting and content maintenance | Questions, channels, languages and handover |
| Retrieval or RAG assistant | Build, model usage and evaluation | Document volume, access rules and test set |
| Automated outreach workflow | Build, data, sending infrastructure and operation | Markets, contacts, inboxes and integrations |
| AI voice agent | Build plus usage linked to call minutes | Call volume, duration, logic and telephony |
| Automated reporting | Build, connectors and maintenance | Data sources, refresh rate and data quality |
| Integrated system | Build and operation across several components | Interfaces, permissions, failure paths and support |
The five factors that move the number
Two projects in the same category can have very different scope. These factors commonly change the estimate:
- Number of integrations. Every connection to a CRM, calendar, ERP or legacy system is development time. A well-documented modern CRM is quick; a bespoke internal system that nobody has touched in five years is not.
- Flow complexity. A linear three-step flow is fast. Ten conditional branches with business rules that only two people in the company fully understand is where the hours go.
- Number of channels. Web, WhatsApp, email, voice. Each additional channel is a separate implementation, not a checkbox.
- Data quality. If your customer records are clean and in one place, enrichment and routing are cheap. If they are spread across three spreadsheets and an inbox, the cleanup is part of the project and it is not free.
- Whether it has to survive contact with real volume. A demo that works for one user and a system that handles a Monday morning are different pieces of engineering: error handling, retries, rate limits and monitoring are most of the difference.
How to work out your own payback
Do not take anyone's ROI claim at face value. The arithmetic is simple enough to do yourself:
- Count the hours. Pick one process and estimate, honestly, how many hours a week your team spends on it. Ask the person who actually does it, not the person who manages them.
- Put a cost on the hour. Fully loaded, not salary divided by hours. Add employer costs and the opportunity cost of what that person is not doing.
- Multiply by the number of weeks the process actually operates each year. State that assumption in the model.
- Divide the project price by the monthly net benefit to estimate payback time. Net benefit should subtract usage, maintenance, supervision and exception handling. Decide the acceptable period from your own cash, risk and alternative investments rather than a universal threshold.
What the cost looks like after launch
The build price is not the whole picture, and any agency that presents it as such is being economical with the truth. Ongoing costs fall into three buckets:
- Platform and model usage. Usually the smallest line for text-based systems. Voice is the expensive exception: per-minute call costs are real and worth modelling before you commit.
- Maintenance. Things you integrate with change their APIs. Budget for someone to notice and fix that, whether it is us on a retainer or your own team with the documentation we hand over.
- Iteration. Reserve time to review observed failures and improve the workflow after launch.
Red flags that you are about to overpay
- A billing model with no explanation of who carries scope and estimation risk. Fixed price can suit stable scope; time-based billing can suit discovery or changing requirements.
- A quote with no written scope. If the deliverable is not specific enough to argue about, it is not specific enough to pay for.
- No written answer on account ownership, configuration access, data export and handover.
- A percentage of revenue or savings without an agreed baseline, attribution method, exclusions and audit rights.
- Refusal to state the unit rates and volume assumptions used to estimate ongoing cost.
What we do
Our published English-language offer is Ace Reception 24 for independent hotels. The price is USD 1,990 setup. The first month is included; USD 290 per month starts on the day-30 report, with no minimum term. Work starts around two weeks after payment. Connecting a booking system is quoted separately. These conditions apply to that defined offer, not to AI projects in general.