When can I start growing the business?

As a summary of the previous chapters, the most important ideas are:

Here are the resources needed for success and how you can calculate your costs.

For AI integration in B2B, costs are not a black box. Even when a provider offers a single price, the project is inevitably split into Setup (integration, data, rules, testing) and Run (operation, monitoring, adjustments).

We present them separately for clarity and so that the budget correctly reflects the reality of the implementation. For specialists, Setup is CAPEX (investment), and Run is OPEX (operations).

First, we present the cost structure (human resources, external services, and the two phases). Then we include a cost matrix. You can, therefore, get a pen ready.

We saw in the implementation stages in Ch. 4.1 that sales, marketing, finance, and IT departments can be involved in AI adoption. For the project to support business growth, not only technical roles are important. That's why we start with the business roles.

In the Build version, requirements for data and ML engineering increase significantly. In the Managed version, the focus is on configuring external services, data flows, and business rule control.

Even companies with strong internal teams frequently outsource some activities:

Case Study: Financial sentiment analysis with GenAI & Google Cloud

From a financial perspective, costs can be grouped into two phases.

With a good understanding of your own company and a few easily confirmed assumptions about the IT market, you can calculate an estimated cost range using the tables below.

The CFO can view AI as a digital employee. It has a recruitment and onboarding cost (Setup = CAPEX) and a monthly salary (Run = OPEX), going on to work 24/7 for the company, under continuous health and working condition monitoring.

We group the percentages below according to company type:

• SMB / Generic: Small catalog, relatively simple data, standard ERP.
• Mid-market / Distribution: Large catalog, long transaction history, uncleaned data, legacy ERP, complex pricing logic.
• Enterprise: Internal data team, maximum security and customization requirements.

1. Setup: Initial Construction

Where are the architecture and development days consumed?

2. Run: Operation, maintenance, experimentation

For a long-term cost evolution vision, we believe that as AI tools become more powerful, raw execution costs (data cleansing and implementation) will decrease as a percentage.

Value and, implicitly, the budget will move towards analysis and architecture. The advantage will belong to those who know what to build and how to integrate the result into business workflows.

In brief:

Sales with AI - Quoting Software for sales team efficiency

We explain Google's policies and guarantees, along with the quality methodology of OPTI Software, when using hybrid architecture for AI in sales.

The proposed architecture is based on four fundamental promises from Google, which are public.

Google's policy is explicit and contractual: data remains the company's property. Data is stored and processed in the region chosen by the customer out of those available.

Sales history and pricing strategies are not used to train other customers' models (without explicit permission). Currently, for Vertex AI Search, the location can be EU multi-region, while Vertex AI Search for commerce has a global location only.

Google supports the principle of "Grounding in Enterprise Truth" and provides tools for minimizing AI hallucinations.

The core data consists of what the company holds in its systems (ERP, CRM, e-commerce). In a hybrid architecture, AI results are additionally filtered and enriched with "warm" data (just-in-time).

Google promotes an open architecture, "Open Cloud":

If you implement the Build (custom) version on Google Cloud, you can export all business logic, data, and models, reducing significantly the risk of customer lock-in.

If you use Google Cloud via the Managed version, the pre-trained model cannot be exported. However, data, events, and configurations can be exported. Migration would involve training a new model or adopting a pre-trained model from another vendor (e.g., Amazon Personalize in AWS).

In the vast majority of use cases, the cost model is variable ("pay-as-you-go"), based on actual resource consumption (per calls/resource), rather than fixed licenses.

Founded in 2005, OPTI Software has implemented multiple technologies, vendors, and channels. In our AI projects, the partnership with Google for cloud and HubSpot for CRM (e.g., Breeze agents) elegantly resolves the tension between innovation (selling more) and control (not selling at a loss of margin or brand image).

Our 20 years of experience in custom development are the foundation for successfully integrating AI with existing software.

The quality principles recommended in this guide do not depend exclusively on one vendor. OPTI Software additionally includes mandatory checks for every project.

To specify the Data Contract from Ch. 2.5.2 and achieve operational clarity, we perform eight checks:

Case Study: Analytics in BigQuery for the gaming industry

In the Managed version, Google offers ServingConfigs to control recommendations. As in the Build (custom) version, these are useful for defining business rules (guardrails) for company safety.

In B2B applications or middleware, our procedures always consider:

Image Diagram

        AI INTENTION                       SAFETY FILTERS                         SUCCESS

┌───────────────────────────┐       ┌───────────────────────────┐       ┌───────────────────────────┐
│ "Recommend product        │       │ 1. Stock filter (vs. ERP) │       │                           │
│  or discount"             │ ─────>│ 2. Margin filter          │ ─────>│ VALID OFFER               │
│                           │       │ 3. Other business rules   │       │                           │
└───────────────────────────┘       │ 4. Permissions limit      │       └───────────────────────────┘
                                    └─────────────┬─────────────┘
                                                  │
                                                  ▼
                                    ┌───────────────────────────┐
                                    │ BLOCKED: "Exceeds minimum │
                                    │ allowed margin"           │
                                    └───────────────────────────┘

                

Based on the principle of grounding in company truth, we recommend:

For decision transparency, we can use Google's Explainable AI tools, augmented with dynamic business rule explanations, providing agents with not just a recommendation output, but the reasoning behind it. While there are limits at the moment, the argument for the recommendation is available (partially) in the Managed version as well, not just in Build.

An AI system might only explain "score 0.95" or "primary reason: X", whereas a hybrid system can also explain reaching a stock limit via guardrails.

Image Diagram

       HISTORICAL DATA                        AI MODEL                      HYBRID EXPLAINABILITY

┌───────────────────────────┐       ┌───────────────────────────┐       ┌───────────────────────────┐
│ Previous                  │       │                           │       │ RECOMMENDATION:           │
│ transactions              │ ─────>│ Vertex AI Search          │ ─────>│ "Installation kit"        │
│ (BigQuery/SQL)            │       │ for commerce              │       │                           │
└───────────────────────────┘       └─────────────┬─────────────┘       │ Score: High               │
                                                  │                     │ Reason: "In stock and     │
                                                  ▼                     │ fits this customer"       │
                                    ┌───────────────────────────┐       └─────────────▲─────────────┘
                                    │ Explainability            │                     │
                                    │ Layer                     │ ────────────────────┘
                                    └───────────────────────────┘

                

Company applications can be configured to display easy-to-understand hybrid labels (e.g., Best seller in stock, Frequently bought together by 102 people). Consequently, trust in AI will increase. When the sales agent can say to the customer "The system noticed that you bought X 2 months ago and now you might also need Y", they become a consultant.

In brief:

Do you want an estimate for your company? OPTI Software can provide a free audit (Setup vs Run).

×