In developmentThe Keystone platform is being built and is not yet in use on client engagements. Companies, buyers and figures in the walkthrough are illustrative and fictional; hours, volumes and metrics are design targets, not results.

Product design v2 · Nov 2026 In development, not yet live Example companies and figures are fictional; metrics are design targets
Keystone · The deal engine

Agents work around the clock. Bankers make the calls and close the deals.

Keystone is a digital deal team made up of ten types of AI agents. Around the clock and across markets, they find sellers, find buyers, match them, run outreach, draft materials, manage the process and answer diligence. Bankers do three things: meet clients, make judgment calls and sign off. We do not sell this software. We sell completed transactions.

10agent types, each owning one part of the work
7×24sourcing, follow-up and drafting continue overnight and on weekends
≈260 hrsof banker time per deal (vs. ~1,850 traditionally)
16 / yrclosings targeted per 3-person pod (vs. ~4 for a traditional 4-person team)
The digital deal team

Ten agent types take on the work of analysts, associates and deal coordinators

Each agent type has a defined role, its own tools and data, and a set level of autonomy. The Deal Captain breaks work into tasks, assigns them, tracks progress and hands anything that needs judgment to a banker. The "overnight output" figures are design targets for one night of a pod running five live mandates.

Deal Captain

Breaks every deal into tasks for the other agents, tracks progress and deadlines, and gives bankers a to-do brief each morning.

Tools: task queue, calendar, deal board
Overnight: status updates on 5 deals + morning brief
Automatic
Sourcer

Continuously scans private companies in the US and abroad, builds and updates profiles, and flags signs an owner may sell.

Data: state registrations, licenses and government contracts, job postings, company websites, trade associations, commercial databases
Overnight: ~48,000 company updates, 37 new signals
Automatic
Buyer Intel

Tracks every PE fund, strategic acquirer and family office: acquisition history, dry powder and preferences, to infer each one's buy box.

Data: deal announcements, fundraising data, news in multiple languages, annual reports of overseas companies
Overnight: ~2,100 buyer profiles updated
Automatic
Matchmaker

Scores every seller against every buyer on strategic fit, price range and likelihood of closing.

Data: our own deal records, buyer profiles, seller profiles
Overnight: a buyer shortlist for every new seller
Automatic
Outreach

Drafts personalised emails and follow-up sequences, books meetings into bankers' calendars and logs every contact.

Tools: email, calendar, CRM
Overnight: ~120 drafts awaiting approval, 18 follow-ups on approved templates
First email approved
Analyst

Three-method valuation, financial model, teaser and CIM, with every figure linked to its source document.

Tools: spreadsheet and document generation, comparable-transaction database
Overnight: first drafts of 3 CIM sections
Sign-off before release
Process

Runs the buyer funnel: sends NDAs, grants data-room access, chases bids and reports to the seller weekly.

Tools: data room, e-signature, email
Overnight: 6 NDAs out for signature, 4 buyer reminders
Automatic within approved process
Diligence

Reads the general ledger and bank statements to draft the QoE, and drafts data-room Q&A.

Data: ledgers, bank statements, contracts, tax returns
Overnight: 41 Q&A drafts
CPA review
Documents

Drafts and marks up NDAs, LOIs and purchase agreements, and compares each side's changes.

Data: clause library, prior agreements
Overnight: 2 agreement mark-ups
Attorney sign-off
Compliance

Pre-send review, conflict checks, numbered archiving of all external communications, and client KYC.

Rules: FINRA 2210, information barriers, watch and restricted lists
Overnight: every draft pre-checked, queued for the CCO each morning
Automatic pre-check
Product walkthrough · about 70 seconds

One deal, from an overnight signal to signed letters of intent

Example: Ridgeline HVAC Services (fictional), a commercial HVAC services company in the US Southeast with $24M revenue and $3.8M EBITDA, whose founder is preparing to retire.

Keystone · Deal RDG-0127 · Ridgeline HVAC (example)

Question 1 · Where deals come from

Agents go out and find both sellers and buyers, rather than waiting for introductions

Deal flow at a traditional boutique depends on partners' networks, and leaves when they do. Keystone turns origination into a pipeline that never stops: sellers found on the left, a buyer universe built on the right, matching in the middle, 24 hours a day.

Seller sources Buyer sources Referrals and inboundAdvisors, lenders and owner inquiries Global public-data scanRegistries, licenses, hiring, news Banker relationshipsPast mandates, unclosed deals InboundWeb, content, adviser referrals PE funds and platformsAdd-on buy boxes US strategic acquirersConsolidation, regional expansion Overseas strategic acquirersJapan, Europe, SE Asia, Middle East Family offices, independentsLong hold, owner can stay on Keystone matching Seller profiles × buyer profiles Fit · price range · likelihood to close Sharper with every deal we close Runs 24/7 Sell-side mandates Owners selling; success fee at closing Fees set out in the engagement letter Buy-side mandates Off-market targets for PE and strategics Monthly retainer + success fee Conflict wall
Question 2 · Serving sellers and buyers

Yes. Each side feeds the other, and in any one deal we act for one side only

The more sell-side mandates we hold, the more buyers will pay for access to our targets; the more buyers we serve, the more sellers trust us to get a strong price. That network effect is something a software company cannot build. The price is strict control of conflicts.

One side per transaction

Once we represent a seller, we charge no buy-side fee to any client acquiring that company. A buy-side client may bid as an ordinary buyer, on the same terms as every other bidder.

Agents flag conflicts; people resolve them

Every match is checked automatically against both sides' engagements. A conflict stops the process and goes to the Chief Compliance Officer, who either discloses it and obtains both parties' written consent or declines one side.

Information stays separate

A seller's confidential information is open only to that deal team, and a buy-side client's criteria are never shown to sellers.

Question 3 · How a few people run many deals

Banker time per deal falls from about 1,850 hours to about 260

The savings come from analysis, drafting, buyer outreach and process management, which move to agents. Negotiation and closing barely change: that is what clients pay for. Figures are design targets; traditional hours are our estimate of common industry practice and will be calibrated against hours measured on our own engagements.

Banker hours per sell-side deal
Traditional boutique (est.)Keystone (target)

A traditional boutique deal team

MDVPAssociateAnalyst
~4 closings / yr · 4 people4–6 mandates in progress at once

A Keystone pod

MDVPReview associateTen agent types
~16 closings / yr · 3 people~6,000 hours a year for 3 people ÷ ~260 hours per mandate ≈ 23 mandates, ~70% closing

At firm level, 15 bankers (5 pods) could close about 80 deals a year, which would take about 80 people in the traditional model. This is capacity, not a financial forecast; the revenue projections in the operating plan stay conservative.

Question 4 · Where agents stop

Internal work is automatic, anything external goes through a person, and judgment and signatures stay with people

Agents bring reach, expertise and stamina: they search sources worldwide in many languages, remember every past deal and keep working overnight. Their boundaries are set by regulation and client trust, not by what the technology can do.

Agents do automatically

  • Research, profile and update companies and buyers worldwide
  • Draft valuations, models, CIMs, Q&A and agreements
  • Score matches, rank buyers, pre-check conflicts
  • Send follow-ups and book meetings on approved templates and sequences
  • Send NDAs, manage data-room access, chase deadlines
  • Draft the daily brief and weekly client reports

Needs a person's approval

  • The first email to every new contact
  • Any material or data sent outside the firm
  • Valuation ranges and pricing advice
  • Every QoE adjustment (CPA)
  • Anything involving a conflict of interest (CCO)

Agents never

  • Negotiate price or make commitments for a client
  • Sign any agreement
  • Answer legal questions or give investment advice
  • Handle client funds
  • Place calls as a person or pose as a banker
Build plan

First build the ability to find deals, then the ability to run them

Our bankers run deals by hand today. The first agents to go live are Sourcer, Buyer Intel, Matchmaker and Analyst. They will be proven on our own engagements, with measured results, before wider rollout.

Jan 2027

Find deals

  • Sourcer, Buyer Intel, Matchmaker
  • Analyst (valuation, CIM)
  • Live on our own engagements
First milestone
Apr 2027

Run the process

  • Outreach, Process, Compliance
  • Deal Captain and morning brief
  • All of our engagements on the platform
Target
H2 2027

Diligence

  • Diligence agent
  • Referral partners connected to sourcing
Planned
2028

Both sides at scale

  • Documents agent
  • Buy-side mandates at scale
Planned
2029+

Cross-border network

  • Overseas buyer coverage
  • Shared deal data across markets
Vision
How we will prove it

Four metrics we will measure once the platform is live

These are design targets, not results to date. We will measure them on our own engagements and report results only once they are measured.

Banker hours per deal
1,850260 hrs
Time recorded per deal, all reviews included
Closings per 3-person pod per year
416
Counted at signed definitive agreement
Sell-side mandates sourced by the platform
> 50%
First identified by an agent, not a personal introduction
Share of AI drafts changed by people
< 25%
By words changed; a measure of draft quality

Struck-through figures are our estimates of common manual practice and will be calibrated against actual hours on our current engagements. We describe a capability as in use only once it is live; agents in development are always labelled as planned.