How does a bank decide whether to build or buy an AI assistant for its advisors?
Four things: which activity to support first for advisors, by when a result is needed, who will maintain the service after launch, and how the pilot is measured. Lined up, the answers make the choice much clearer. For a reference on timing, the standard Streetbeat for Finance plan takes an assistant from pilot to production in 5-6 months.
Which activity should the bank support first for advisors?
Just one, frequent enough to matter and well-defined enough to measure, chosen together with the advisors: they know where support weighs most in their day. Starting from "an assistant for advisors" leads to a project with no boundaries; starting from meeting preparation or the notes to enter in the CRM gives it a beginning and an end. The choice decides which data is needed and what gets measured after 3 months.
| Activity | Data needed | What the advisor decides |
|---|---|---|
| Meeting preparation | CRM, portfolios, previous notes | The summary and the points to discuss |
| Post-meeting follow-up | Meeting notes, CRM | The actions and the message to the client |
| Proactive outreach | Portfolios, market data | Whom to contact and for what reason |
By when is a result needed?
Often the date comes from the board of directors, and it changes the answer. A vendor can be asked for the schedule from the start: the Streetbeat plan provides for a pilot with real usage of 3 months, followed by about 8-13 weeks before production. With in-house development, the schedule depends on what the bank already has: access to models, data connections, controls and interface.
Who will maintain the service after launch?
This is a question that tends to be postponed. Launch does not end the work: model providers change versions and terms, market data must be kept current, users ask for support. Streetbeat orchestrates 4 model providers and gives access to more than 170 financial datasets: a bank building in-house has to manage the equivalent with its own team, for as long as the service is in use.
| Component | If the bank builds | If the bank buys |
|---|---|---|
| Models and their versions | The in-house team | To be allocated in the vendor contract |
| Connections to CRM and portfolios | The in-house team | To be allocated in the vendor contract |
| Licensed market data | The bank, under its own contracts | To be verified: included or separate |
| User support | The in-house team | Service levels to be set in the contract |
| Review rules and permissions | The bank | The bank |
Which rules apply either way?
The rules on model risk do not change with the path chosen. The Federal Reserve's SR 26-2, dated April 17, 2026, replaces the 2011 guidance and also covers models purchased from third parties, but it does not cover generative AI: for an assistant of this kind, the bank defines its own governance framework, together with the other applicable requirements, such as third-party vendor management. Buying shifts the technical work, not the accountability.
Which numbers should the pilot be asked for?
The same ones for both paths, defined before starting. At a bank with 3,000 advisors using Streetbeat, over three months 85% of advisors were active and 99.64% of interactions were error-free, user errors included; the cost of the model tokens alone was between €0.01 and €0.05 per interaction. These are the results of a single institution and do not predict those of another, but they show the level of detail to ask for from an in-house project as well.
| Measure | How it should be defined |
|---|---|
| Active advisors | Out of how many enabled, and over what period |
| Errors | With what definition, user errors included |
| Cost per interaction | The model only, or also infrastructure and oversight |
| Time to production | When advisors use it every day, not when the demo starts |
Which path do the answers point to?
There is no rule, but there is a direction: what matters most is who is on the team and how close the deadline is. In either case, the choice is tested on a single workflow, in a 3-month pilot, before extending it to the network.
| If the bank... | Direction |
|---|---|
| Has an AI team that has already taken services to production | Build, at least the first workflow |
| Has developers but no dedicated AI team | Buy the platform and keep data and rules in-house |
| Must show a result within a year | Buy, with a pilot on a single workflow |
| Sees the assistant itself as its competitive advantage | Build, budgeting for maintenance |
About Streetbeat
Streetbeat helps organizations use AI to grow revenue, not costs. Streetbeat OS, the AI operating system for companies, is the foundation of Streetbeat for Finance, Streetbeat for Impact and Brain50 by Streetbeat.
Streetbeat for Finance
Streetbeat for Finance brings AI to banks, wealth and asset managers, insurers and real estate investment firms. It includes the AI Suite for Advisors, the Portfolio Engine, the AI Suite for Retail and the AI Suite for Operations, deployed on-premise or in the client's cloud and configured around the institution's workflows, data and controls. See Streetbeat for Finance.