Your AI doesn't know your firm. Connect it to the tools that do.

Four AI workflows, four stages of the deal, one idea: AI gets useful the moment it can reach your pipeline, your relationships, and your data. Here are all four, with the prompts.

Everyone at your firm is already using AI. Most of it happens in a blank chat window that has never seen your pipeline or a single one of your deals. The prompts might be perfect, but the answers come back confident and ungrounded, with no way to tell whether any of it is real.

That's the gap these workflows close. We built four of them with leaders across the deal lifecycle: Grata, Crunchbase, AlphaSense, and Lumonic. Different stages, different data, but the same move underneath.

The move: connect the model to where the work lives

An LLM on its own gives you half an answer. It doesn't know which deals you've won, who at your firm already knows the founder, or what your portfolio reported last quarter. Connecting it through MCP, a standard way to plug a model into the systems you use, changes what it can do. Now it can run a real search, read your CRM, and reason across both.

The strongest version pairs two connectors as a loop. A partner brings the outside picture of live market and company data, and Affinity brings the other half: the relationship and deal context your firm generates just by doing business, captured automatically without anyone logging it. One side tells you what's out there while the other tells you what you already have a path to. That loop is fullest where a partner connects directly to Affinity, as Grata and Crunchbase do, and the whole thing runs in one chat. Where a partner doesn't, its data still complements what Affinity already holds about your firm, informing the same decision from the other side.

To get started with Affinity MCP, follow our step-by-step guide here.

Four workflows, four stages of the deal

Each workflow is self-contained, with the exact prompts. Start wherever your next deal is.

Proprietary dealflow, with Grata: from a one-line thesis to a warm intro

Type one sentence like "I want to take advantage of rising demand for data center builds" and watch it become five searches running in parallel, a ranked read on the sub-verticals, and a filtered list of bootstrapped companies in the states with the most build activity. Then Grata scores each one by seller intent: who's researching a transaction right now, the signal a company is genuinely in market. Grata and Affinity connect directly, so the search results and your firm's relationships live in one place. Affinity checks the list against your collective network, surfaces the one company you already have a warm connection to, and drafts the outreach from the full history of that relationship. Thesis to intro without leaving the chat.

→ Read the workflow and copy the prompts: Proprietary dealflow with Grata

Deal sourcing, with Crunchbase: from a hunch to an investment memo

The obvious sectors are crowded and everyone's fighting over the same names. This workflow goes the other way. It finds the sub-categories heating up before they're obvious, then narrows to the specific companies inside them. One prompt chain takes you from "which B2B SaaS categories are accelerating?" down through micro-industries, to the startups most likely to raise next, to a full investment memo with a built-in bear case, exported as a one-pager you can hand to your committee. Crunchbase and Affinity connect directly, which powers the sharpest part: feed your last twelve months of wins into the search and your own track record becomes the query, grounded in who you already know.

→ Read the workflow and copy the prompts: Deal sourcing with Crunchbase

Diligence, with AlphaSense: from a blank prompt to a data-room red-flag list

Diligence has always taken weeks, and most of that time goes to synthesis, not access. This workflow compresses it. A plain-language prompt on an industry returns a sourced market read, every claim traceable back to the document it came from. From there you can generate a committee-ready slide, pull precedent-transaction comps into Excel, and run channel checks across expert transcripts. The newest piece: import an entire virtual data room into a workspace and ask it what you'd otherwise spend days finding: the red flags, the customer-concentration risk, the documents that are missing. Pair that diligence trail with the relationship context in Affinity and you know two things at once: whether the company holds up, and who at your firm can get you in.

→ Read the workflow and copy the prompts: Diligence with AlphaSense

Portfolio monitoring, with Lumonic: a portfolio view that isn't a quarter stale

The work doesn't stop at close; it just stops being visible. By Lumonic's count, most private-credit firms still rebuild their portfolio picture by hand in spreadsheets every quarter, which means every decision runs on data that's already out of date. This workflow keeps it current. Lumonic (a PitchBook company) holds what your companies reported; PitchBook holds what the market is doing. Together they build a blended valuation model, flag any company that breaches a covenant or KPI, and stand up a monitor that re-prices positions as conditions move, with every number one click from its source cell for audit. The market side of this loop runs on PitchBook, which Affinity connects with directly, so the data pricing your portfolio is the same picture grounding your next deal.

→ Read the workflow and copy the prompts: Portfolio monitoring with Lumonic

What every workflow has in common

Four partners, four stages, and the same pattern each time. Every number traced back to a source you could click. Work that used to take weeks came back in minutes. The model surfaced, ranked, and drafted; the judgment stayed with the person. And each one points the same direction: a partner's outside view paired with the deal and relationship history Affinity already holds, so a team works off one current picture instead of four stale tabs, most completely when the two connect directly.

Setting it up

These workflows run in the AI assistant your firm already uses. Across the series the demos spanned Claude, ChatGPT, and the partners' own platforms, with connectors switched on, and setup takes a couple of minutes. To get started with the Affinity MCP, follow our step-by-step guide here.

Where to go from here

Copy the prompts from any of the four workflows above and run one on a live deal this week. When you want the loop built around your own pipeline, book time with our team and we'll set it up for your firm.

FAQ

Is using AI in a chat window enough for deal work?

On its own, an LLM doesn't know your pipeline or who at your firm has already met the founder, so its answers stay generic and hard to trust. Connecting it to your systems is what makes the output specific to your firm and traceable to a source.

How does Affinity fit with tools like Grata, Crunchbase, AlphaSense, and Lumonic?

The partner supplies the outside picture (market, company, or portfolio data), and Affinity supplies the inside picture: the relationship and deal context your firm captures automatically. Grata and Crunchbase connect with Affinity directly; Lumonic's market data comes through PitchBook, which Affinity also connects with. Where a partner connects to Affinity, the two form a loop in which every deal you close makes the next one easier to find; where it doesn't, the partner's data still complements what Affinity holds about your firm.

Which AI model do the workflows use?

No single one. Across the series the demos ran in Claude, ChatGPT, and the partners' own platforms. Because MCP is model-agnostic, these workflows run in whichever AI assistant your firm already uses.

How do you connect a CRM to AI tools like Claude or ChatGPT?

Connect it through an MCP connector in your AI assistant. In Claude, for example, open Settings, go to Connectors, add a custom connector, and paste the CRM's MCP address, then log in as usual. Once connected, the assistant can read your CRM and act on it in plain language, and the same approach works across assistants because MCP is model-agnostic.