You don't need to pick the winning product
The AI tool market changes every quarter. New models, new products, new pricing. If your readiness report recommends a specific vendor, it will be outdated before it reaches leadership.
Instead, think in categories. Each category solves a different class of problem. Once you know which category fits your workflow, you can evaluate specific tools within that category — and swap them out as the market evolves.
There are four categories that matter for financial services firms. We will walk through each, then you will match your priority workflows from Module 4 to the right category.
What is the best approach to tool selection in a fast-moving AI market?
Category 1 — AI Assistants
What they are: Conversational AI tools that you interact with directly — Claude, ChatGPT, Gemini. You paste text, upload documents, ask questions, and get responses.
What they solve: Individual productivity. Document analysis, research synthesis, drafting, summarisation, brainstorming.
When to use: When the task is ad-hoc, the input is a document or a question, and the output is text. Think of it as having a very fast, very well-read junior analyst available 24/7.
Financial services examples:
- Summarise a 150-page CIM and flag key risks
- Compare management commentary across four quarterly earnings transcripts
- Draft a first-pass investment memo from uploaded due diligence materials
- Review an LPA and extract key economic terms into a table
Limitations: The work stays in the chat window. You copy-paste output into your real tools. There is no connection to your firm's systems — Bloomberg, your CRM, your document management system. It is a powerful standalone tool, but it is standalone.
Category 2 — AI-Powered Work
What they are: Tools where AI operates more autonomously — reading files, writing documents, executing multi-step tasks, running on a schedule. Claude's "Cowork" feature is an example: you delegate a task, and the AI works through it, producing deliverables you review.
What they solve: Tasks that are too complex for a single prompt. Multi-step workflows where the AI needs to read several documents, cross-reference information, and produce a structured output.
When to use: When you would normally assign a task to a junior team member and review their work product. The AI reads the inputs, does the analysis, and produces a draft for your review.
Financial services examples:
- "Review these five portfolio company reports and produce a consolidated quarterly update with performance flags"
- "Compare this term sheet against our standard terms and produce a redline summary"
- "Read these three broker reports and produce a consensus-vs-divergence analysis"
Limitations: Still operates within the AI platform. The files you give it are the only files it sees. It cannot reach into your portfolio management system or pull live market data — that requires the next category.
Category 3 — Enterprise Integration
What they are: Protocols and platforms that connect AI to your firm's existing systems — databases, document repositories, CRMs, compliance tools. The Model Context Protocol (MCP) is the leading standard: it lets AI tools read from and write to your enterprise systems through a controlled, auditable interface.
What they solve: The "last mile" problem. AI is smart, but it is useless if it cannot access your data. Enterprise integration means the AI can pull live portfolio data, query your document management system, or check compliance rules — all within a governed framework.
When to use: When the workflow requires data that lives in your firm's systems, not in uploaded documents. When you need AI to act on real-time information, not static files.
Financial services examples:
- AI pulls the latest NAV data from your portfolio system and flags positions that breach concentration limits
- AI queries your CRM to identify which LPs are approaching re-up decisions and drafts personalised outreach
- AI connects to your compliance database to check whether a proposed trade conflicts with your restricted list
Limitations: Requires technical implementation. Someone needs to build the MCP connectors to your systems. This is where IT involvement becomes essential — and where governance frameworks matter most, because AI is now touching live data.
Why does enterprise integration require stronger governance than standalone AI assistants?
Category 4 — Team Infrastructure
What they are: Shared resources that make AI consistent and scalable across your firm — prompt libraries, custom skills, plugins, and standardised workflows. Think of these as the "playbooks" that ensure every team member gets quality results, not just the one person who figured out how to prompt well.
What they solve: The "one person is 3x productive, the team stays the same" problem. Without shared infrastructure, AI adoption is individual and inconsistent. With it, every team member benefits from the best prompts, workflows, and configurations the firm has developed.
When to use: When more than one person needs to do the same type of AI-assisted work. When you want consistent quality regardless of who is using the tool.
Financial services examples:
- A shared prompt library with tested templates for DD memos, earnings analysis, and compliance reviews
- Custom skills that encode your firm's specific analytical frameworks — "Run our standard credit assessment" as a one-click action
- Plugins that connect AI to your specific data sources and formatting standards
- Onboarding workflows that get new analysts productive with AI in their first week
Limitations: Requires someone to build and maintain the library. Prompts go stale as workflows change. This is an ongoing investment, not a one-time setup.
The matching exercise
Pull up your VALUE-scored workflows from Module 4. For each of your top three workflows, ask yourself these questions:
Does this workflow mainly involve reading and analysing documents I can upload? If yes → Category 1 (AI Assistant) or Category 2 (AI-Powered Work), depending on complexity.
Does the workflow require multiple steps, cross-referencing, or producing structured deliverables? If yes → Category 2 (AI-Powered Work). Single-prompt tasks stay in Category 1; multi-step tasks need delegation.
Does the workflow require data that lives in our firm's systems — not in files I can upload? If yes → Category 3 (Enterprise Integration). You will need MCP connectors or equivalent.
Do multiple people on the team do this same workflow? If yes → Category 4 (Team Infrastructure) on top of whatever category you chose above. Build the prompt, share it, and make it a standard.
Here is a quick reference:
| Workflow Type | Category | Example |
|---|---|---|
| Analyse an uploaded document | 1 — AI Assistant | Summarise a CIM |
| Multi-step analysis across documents | 2 — AI-Powered Work | Quarterly portfolio review |
| Requires live firm data | 3 — Enterprise Integration | Compliance checks against restricted list |
| Done by multiple team members | 4 — Team Infrastructure | Standardised DD memo template |
Self-assessment — current tool usage
Before you can recommend tools, you need to know where your firm is today. Be honest — this goes into your readiness report.
Which best describes your firm's current AI tool usage?
The build vs buy decision
Financial services firms face a specific version of this question. The answer is almost always: buy the platform, build the configuration.
Buy:
- The AI model itself (Claude, GPT, etc.) — you are not training your own model
- The enterprise platform (security, compliance, audit trails) — these take years to build properly
- MCP connectors for common systems (Salesforce, standard databases) — no need to reinvent these
Build:
- Prompt libraries specific to your firm's workflows and standards
- Custom skills that encode your analytical frameworks
- MCP connectors for proprietary or niche internal systems
- Governance rules specific to your regulatory environment
The firms that try to build everything from scratch spend 18 months and millions of dollars recreating what enterprise AI platforms already provide. The firms that buy everything off the shelf get generic results that do not match their workflows. The sweet spot is buying the infrastructure and building the intelligence layer on top.
A managing director asks: 'Should we build our own AI platform?' What is the right response?
What to bring forward
You now have three things to carry into Module 6:
- A category match for each of your priority workflows — you know whether each one needs an AI assistant, AI-powered work, enterprise integration, or team infrastructure.
- An honest assessment of where your firm is today — which categories you already use and which are gaps.
- A build-vs-buy perspective — buy the platform, build the configuration.
Write these down. They go directly into Section 3 of your readiness report in Module 7.
Module 5 — Knowledge Check
What is the key difference between Category 1 (AI Assistants) and Category 3 (Enterprise Integration)?
When should you invest in Category 4 (Team Infrastructure)?
What is the recommended build vs buy approach for financial services firms?
