Boutique, startup or large consultancy: how to choose an AI provider
Choosing an artificial intelligence provider is becoming increasingly complex in a rapidly expanding market where a wide variety of players, solutions, and proposals coexist.
Faced with the same need, an organization might receive proposals from a specialized startup, a boutique tech company, or a large consulting firm. All of these may offer technically sound approaches, but at the same time, they may present significant differences in terms of time, scope, and budget.
Technology has reduced some of the distances that previously clearly separated these players and made it much more difficult to determine which proposal is truly comparable to another.
Adding to this complexity is the pressure to move forward. AI is a recurring topic in discussions among boards of directors, shareholders, and business units, all of whom expect increasingly faster results. Those leading in technology and data face the challenge of responding to this demand without losing sight of everything else that... A solution will need to be able to integrate, scale, be governed, and remain relevant to the business..
Choosing an AI provider isn't just about who promises to develop a solution faster or at a lower price. The decision requires understanding the true needs of the use case, what each proposal includes, and what will happen to the solution after implementation.
In this article we analyze why there can be such significant differences between proposals that seemingly address the same need, what aspects are not shown in the price and implementation timeframe, what the main selection criteria are, and why choosing a technology partner involves looking beyond the immediate project.
Why are the same things being traded today at prices that differ by 10x?
The market has changed. As he explains Daniel Menal, Head of Data & AI at IT Patagonia, Today, an organization can sit down with a large-scale company and receive a proposal to solve a specific AI case And the next day, you might encounter a startup of two or three specialists that proposes a technically sound solution for a seemingly similar need, but in a fraction of the time or budget.
The differences that once made it easier to distinguish between types of providers began to diminish. The democratization of artificial intelligence tools and capabilities allowed small, highly specialized teams to rapidly develop solutions that, until a few years ago, required considerably larger structures.
This creates a situation that can be summarized with an idea that is especially useful for those who need to compare proposals: in many cases, everyone can be right.
A startup may be right when it claims to have the capacity to implement a case in a few weeks, a boutique firm when it proposes a highly specialized team, and a larger company may also be right when It incorporates into the proposal additional capabilities for integration, governance, observability, security, or continuity.
The difference lies in which part of the challenge each one is addressing. If the goal is to experiment quickly and validate a hypothesis, a limited solution might be exactly what the organization needs.
But if that development is to later become an integrated business capability, other questions will need to be answered to determine how to choose an AI provider. These include: How will it scale? How will it be controlled? How will the cost of operating it be measured? How will the return on investment be evaluated? What governance will it require?
Comparing only the final price can therefore lead to putting proposals in the same table that seem equivalent but contemplate very different scopes.
Therefore, the decision shouldn't be limited to considering how much it costs to implement the use case. It's also necessary to understand what we are buying for that price and to what extent is each supplier responsible?.
What speed doesn't tell you: scalability, governance, and maintenance
Speed matters. In fact, one of the main challenges is finding a balance between responding quickly to business needs and building a foundation capable of supporting what comes next.
The pressure is there. Boards of directors want to see results, shareholders expect progress, and business units are increasingly aware of the possibilities of AI. For technology and data leaders, simply stating that a project will take several years may no longer be a viable option.
But showing value quickly and building solely for the short term are two different things. A solution can resolve a use case in a few weeks and perfectly meet its initial objective, but the challenge arises when the organization decides to take it further.. This is where observability, integration, data governance, security, operating costs, adoption, maintenance, and the ability to evolve come into play.
Daniel points out that some projects even reach production and then lose momentum because they weren't accompanied by an adoption strategy or the necessary capabilities to sustain them. Therefore, the speed of implementation alone doesn't guarantee the long-term viability of a solution.
This analysis should be part of the TCO (Total Cost of Ownership), since in addition to the initial development, the costs of infrastructure, model consumption, integrations, maintenance, observability, governance, support and evolution should be considered when determining how to choose an AI provider.
A proposal that is ten times cheaper may still be the best alternative, but the comparison needs to be made on equivalent scopes.
5 criteria beyond price and delivery time
There is no single type of provider that is always superior to another. A startup can bring agility and specialization, while a boutique firm can combine specific expertise with a flexible structure, and a larger consultancy can add experience in integration, governance, and operations within complex organizations.
The choice should be based on the use case and the capabilities that the project will need during its evolution and at what point. Let's look at some criteria to consider:
1. Understanding the business problem
The conversation should begin with the need that needs to be addressed, not with a specific tool or model. A provider needs to understand what outcome the business expects, what the KPIs are, the urgency of the situation, and how the project's value will be measured.
This also involves understanding internal expectations. An initiative whose objective is to quickly validate an opportunity should not be evaluated using the same criteria as a solution intended to be integrated into a critical process.
The right partner is not necessarily the one who proposes the most technology, but the one who best understands what the organization needs to achieve with it.
2. Ability to combine speed and scalability
A key lesson lies in understanding that Speed and solidity should not necessarily be considered as alternatives. Organizations need to show results in increasingly shorter timeframes, but without losing sight of what will happen if the project is successful.
Therefore, when comparing proposals, it is advisable to analyze which components will allow evolution from the first use case, what changes would be necessary to increase users or transactions, and what part of what has been built can be reused.
Scaling up doesn't mean over-dimensioning from day one. It means preventing success from forcing you to start all over again from scratch.
3. Governance and integration with the existing ecosystem
AI adds a new technological layer, but it doesn't eliminate everything the organization already needs to manage. Data, quality, architecture, security, privacy, regulations, processes, and integrations remain part of the equation.
Here it is important to consider the risk of linking together small projects intended solely to address urgent business needs without building a common structure to sustain them. This criterion is particularly important for a AI-first strategy.
It is worth noting at this point that the provider should be able to explain not only how it will resolve the immediate case, but also how that solution will coexist with the existing architecture, what governance it will require, and how it will contribute to developing capabilities that can be leveraged in future initiatives.
4. Experience to accompany what comes next
Implementing the use case is only one part of the journey. Once in production, it will be necessary to monitor its performance, control costs, manage changes, resolve incidents, and adapt the solution to new needs.
As AI is gaining more prominence in customer relations and critical processes., Additional challenges also appear.
Security, agent behavior, response quality, and new control mechanisms can become central aspects of the operation. Therefore, when evaluating a provider, it's important to know not only what they can deliver today, but also what experience they have in resolving the challenges that arise after implementation.
Knowledge transfer is also important. If AI begins to become a strategic capability, the organization needs to develop its own knowledge and to prevent the operation from depending entirely on a third party.
5. TCO and project continuity
The initial cost represents only a portion of the investment. Infrastructure, licenses, consumption, integration, maintenance, observability, security, governance, and support can significantly alter the equation over the lifecycle.
Analyzing the TCO allows you to incorporate this perspective and also evaluate the dependency generated by the solution, through questions such as the following: What knowledge remains within the organization? Can another team continue development? Which components are proprietary? What would happen if the supplier changes? How much will it cost to sustain the project when its use increases?
The cheapest proposal on day zero will not necessarily be the lowest cost solution when the solution has been operating for several years.
Why are there increasingly more than one supplier at the same table?
One of the most significant transformations in the IT market is occurring in projects that previously involved a client and a consultancy, and today may bring together the client with two or three different providers.
The reason is not necessarily a lack of trust, but a consequence of complexity. AI and data challenges They involve more and more dimensions, and by definition, no single provider has the best answer for all of them.
A startup can provide a niche solution with enormous speed, a boutique can add specialized knowledge, and a large company can contribute with governance, architecture, security, or industrialization capabilities.
Instead of asking themselves what type of vendor should take on the entire project, some organizations are starting to ask themselves what combination of capabilities they need to best solve it.. This approach also allows for the incorporation of different perspectives into decisions that will have a long-term impact.
But working with multiple vendors doesn't eliminate the need for a holistic view. On the contrary, it makes it even more important. Data standards, architecture, governance, integrations, and business objectives all need to align with a common strategy. Otherwise, the organization may accelerate multiple initiatives and end up building a fragmented ecosystem.
In other words, working with multiple partners can add value, but outsourcing the overall view does not.
Choosing to respond today without compromising what comes next
Those who lead Data & AI strategies They face a difficult-to-avoid tension. On the one hand, they need to listen to the business and respond to a specific demand for speed. On the other hand, they are responsible for building capabilities that must be sustainable once the initial urgency has passed. The key is not to ignore either of these two dimensions when evaluating how to choose an AI provider.
An organization needs to be able to quickly demonstrate value, but at the same time, it must focus on building the structure that allows it to continue growing. Governance, security, quality, architecture, and the capacity to evolve shouldn't only become apparent when the first projects begin to encounter difficulties.
This also changes how you choose an AI provider. The best partner won't necessarily be the biggest, the most specialized, the cheapest, or the fastest. It will be the one whose capabilities better address the problem the organization needs to solve and the path that solution will take if it works.
In some cases it will be a startup, in others a boutique or a larger-scale consultancy, and increasingly it may be a combination of several partners working on a common strategy.
The focus should be on solving the immediate need without losing sight of the overall picture, because an AI case should not be evaluated solely by how quickly it gets to production, but also by the capacity it leaves in place to continue transforming the business.
Scorecard for comparing proposals
When two or more providers offer to solve seemingly the same use case with significant differences in price and time, comparing only those variables provides an incomplete view.
Before deciding how to choose an AI provider, it is advisable to bring the proposals to a common base that allows you to understand what each company is really offering.
With this objective, the scorecard should consider, among other aspects:
- Understanding the problem and aligning it with business KPIs.
- Ability to deliver value in a timely manner that meets the need.
- Experience and specialization relevant to the use case.
- Ability to evolve from a pilot to production and scale.
- Architecture and integration with the existing ecosystem.
- Governance, observability, security, and risk management.
- Support and subsequent maintenance.
- Transfer of knowledge and capabilities to internal teams.
- Operating costs and TCO.
- Technological dependence and possibilities for continuity.
- Ability to collaborate with other teams and suppliers.
Not all criteria need to be given equal weight. A low-criticality experiment might prioritize speed, cost, and specialization. A solution that will be involved in core processes, use sensitive information, or have direct contact with customers will likely need to place much greater emphasis on integration, governance, security, and continuity.
That's where the true value of a scorecard lies: not in determining who is the best supplier in the abstract, but in determining who has the most suitable capabilities for the project that the organization needs to solve.
Steps to follow
If your organization is evaluating different alternatives for an artificial intelligence project, you can also Schedule a meeting with our Data & AI team to analyze the use case and the criteria that should guide the selection.
