Inside a Nine-Agent Business Development Platform

Picture a managing director at a large consulting firm on a Sunday evening. They follow 30 client and target accounts and perhaps 100 executives across them. Somewhere in the past week, one of those executives changed jobs, one company told analysts it was reviewing a cost program, and another filed for a major expansion.

Any one of those is a reason to call. The managing director will hear about maybe one of them, late, from a colleague.

This article walks through a nine-agent business development platform our founding team built to close that gap for the partners of a global consulting firm: the architecture, why it uses several agents instead of one, what the weekly digest looks like, what it cost against the SDR team it replaced, and what we learned building it.

What the Platform Does

The platform watches the accounts and executives the firm's partners care about, notices when something changes, researches what the change means and delivers a weekly digest to each managing director.

The published facts about it are specific:

• Nine agents watch executive moves, earnings calls, news, filings and major events.
• One orchestration layer builds a weekly digest per managing director.
• 3,000+ qualified leads a month come out of it.
• 1,400+ executives are tracked continuously.
• 10+ signals are researched per prospect.

Before it existed, that watching and research work sat with a six-person SDR team.

The Architecture, Layer by Layer

We do not publish the name and job of every one of the 9 agents. What we can describe is the pattern they follow, which is the part worth copying anyway. The platform works in 4 layers.

Fig. 1The four layers of the platform

1. Watching

   Agents each monitor one type of source for changes at tracked accounts and executives.

   Executive movesEarnings callsNewsFilingsMajor events
2. Research

   Each change is expanded with context: who is involved, what it means, what was said.

   10+ signals per prospectSources linked
3. Scoring

   Changes are ranked by relevance to each partner's accounts and practice.

   Account fitTimingPartner focus
4. Digest

   One orchestration layer assembles a weekly digest for each managing director.

   WeeklyPer partnerRanked

1. Watching

Each watching agent is responsible for one type of source. One type of agent looks at executive moves. Another reads earnings calls. Others watch news, regulatory and company filings, and major events such as acquisitions or restructurings. Each knows where its source lives, how often it updates and what counts as a change.

2. Research

A raw change is not useful on its own. "A new CFO started" matters more when you know where the CFO came from and what they did there. The research layer adds that context and keeps a link to every source, so nothing in the digest is a claim without evidence.

3. Scoring

Across 1,400+ tracked executives and their companies, a lot changes every week. Most do not matter to any given partner. Scoring ranks each change by how closely it touches that partner's accounts and practice, and how urgent it is. This is the layer that turns a firehose into a short read.

4. Digest

The orchestration layer pulls the scored, researched changes together and writes one digest per managing director, every week. Same system, different digest for each reader.

Why Multi-Agent Instead of One Big Agent

It is tempting to build one large agent with a long prompt that says "watch everything and tell me what matters." We did not, for 4 reasons.

• Sources behave differently. An earnings call transcript arrives quarterly and runs to many pages. An executive move is one line that can appear any day. Filings follow regulatory formats. One agent trying to handle all of them does each badly.
• Narrow jobs can be tested. When an agent has one job, you can check its output against the source and know whether it works. A single agent doing everything fails in ways that are hard to trace.
• Failures stay contained. If a news source changes its layout and one agent breaks, the other 8 keep running. The digest arrives with one section thinner, not empty.
• Each part can improve on its own. A better way to read earnings calls can ship without touching the agent that tracks executive moves.

Here is the tradeoff in a table.

The cost of the multi-agent approach is coordination: something has to combine the outputs, remove duplicates and decide what goes in which digest. That is exactly the job of the orchestration layer.

The Weekly Digest

The digest is the only part of the system most partners ever see, so it carries the whole value. If it is long, vague or wrong, nobody reads it, and the 9 agents behind it might as well not exist.

Here is an illustrative layout of what a digest like this contains. The companies and changes are examples, not real data.

Fig. 2Weekly digest, managing director view

Weekly digest, managing director viewIllustrative layout. Companies and people are examples, not real data.

This week at a glance 1

6 changes across 4 of your tracked accounts.

2 rated act this week, 4 rated worth knowing.

Act this week 2

Northwind Logistics: new CFO started Monday, previously led a cost program at a peer firm. Source: company announcement.

Fabrikam Health: earnings call flagged a supply chain review in the next 2 quarters. Source: call transcript, page 7.

Worth knowing 3

Contoso Energy filed for a new facility. Source: public filing.

Adatum Retail named in sector news on store closures. Source: news, 2 outlets.

Suggested next step 4

Northwind: congratulate the CFO and share the cost benchmark from your last client review.

1. 1A count first, so a partner knows in 3 seconds whether to read on.
2. 2Ranked items with a reason to act now, each tied to a source.
3. 3Lower-priority changes kept short. Context, not a task.
4. 4A suggested move the partner can take or ignore. The system never contacts anyone.

A few design choices matter more than they look:

• It is weekly. Partners do not want a daily stream of alerts. A weekly read fits how they plan their calls.
• It is per person. Each managing director gets their own accounts and executives, ranked for them.
• Every line has a source. A partner about to call a CFO needs to trust the fact they are opening with.
• It suggests, it never acts. The system does not email anyone. It hands the partner a reason and a starting point.

How Partners Use It

The most telling result is not a lead count. Managing directors now open $2M to $5M pitches by demoing the system itself.

That changes what the platform is. It started as a way to replace research hours. It became something the partners sell with, because showing a prospect that you track their executives, their earnings calls and their filings every week says more about how you will work for them than any slide.

Day to day, the pattern a digest like this supports is simple: read it at the start of the week, pick the 2 or 3 items worth acting on, and open each conversation on something real. The same thinking runs through our guide to executive movement tracking, which covers what to do in the first week after a job change.

What It Cost Against the SDR Team

The platform replaced $480K a year of six-person SDR team cost, at 99% lower cost.

For context on that number, one US SDR commonly costs about $100k to $200k a year fully loaded, and ramp is commonly 3 to 6 months. A six-person team doing manual research carries that cost every year whether or not the research gets read.

The system also does things the team could not do at any headcount: track 1,400+ executives continuously and research 10+ signals per prospect without anyone falling behind. You can see the published figures for this and our other systems on the results page, and a wider cost comparison in AI Sales Team vs Hiring an SDR.

Lessons From Building It

These are the lessons the architecture itself teaches. They hold for anyone building a system like this.

1. Start from the reader

The digest decides everything upstream. Once you know what a managing director will actually read in 10 minutes, you know which sources matter and how much research each change needs.

2. Sources before scale

Each watching agent is only as good as the source it reads. Picking reliable sources pays back more than clever prompts do.

3. Every claim needs a link

Partners act on the digest in front of clients. A single wrong fact puts the whole system's credibility at risk, so a source on every line is part of the design, not a nice extra.

4. Ranking is the product

Collecting changes is the easy part. Deciding which handful of changes a specific partner should see is where the work goes. The signals in CRM with decay play shows one way to keep older signals from crowding out fresh ones.

5. Keep people in charge of contact

The system finds reasons to reach out. Partners decide whether and how. A tool that never speaks for them is one partners can comfortably put in front of a prospect.

How to Start With a Smaller Version

You do not need 9 agents on day 1. The layered pattern works at any size, and the safest way to build it is one layer of value at a time.

1. Pick one source and one reader

Choose the source most closely tied to how your senior people win work. For many firms that is executive moves, because a new leader often reviews advisers and vendors. Choose one partner or founder who will read the output every week and tell you honestly what is useful.

2. Ship a plain digest

Before adding scoring, send that reader a simple weekly list of changes at their accounts, each with a source. Ask them to mark what they acted on and what they ignored. That feedback is the start of your scoring rules.

3. Add research, then scoring

Once the reader trusts the list, add a line of context to each change, then rank the changes using what they acted on. Each step should make the digest shorter and more useful, never longer.

4. Add sources and readers

Only when one digest works for one reader do you add a second source or a second reader. Each new source is its own agent with its own checks, which is the reason the pattern scales.

This is the order we would follow for any new build, and it is the same approach we take in our GTM engineering work: one working layer, proven with a real reader, before the next one.

When This Pattern Is the Wrong Choice

• You sell to thousands of small accounts. A digest per partner makes sense when each senior person owns a few dozen relationships. For high-volume, low-value deals, a signal-based outbound engine fits better. See Signal-Based Outbound.
• Nobody will read the output. If partners will not spend 10 minutes a week on a digest, no architecture fixes that.
• You have one source that matters. If job changes are your only trigger, you need one good agent, not nine.

FAQs

Build Your Account Intelligence Platform With Oriel

Oriel builds systems like this for B2B companies above $100k/month in revenue where senior people own the relationships. We pick the sources that matter for your accounts, build the watching, research and scoring layers, and deliver a digest your senior people will actually read, inside tools you own.

On a 30-minute call, tell us who your senior people sell to and what they wish they knew sooner. We will show you which sources we would watch first and what a first digest could look like. If you are not ready for this yet, we will tell you that too.