Business development AI agent for consulting firms: the Bob pattern

Business development in a boutique consulting firm rarely fails because nobody knows sales matter. It fails because relationship follow-up is fragmented across memory, inboxes, calls, notes, LinkedIn, and good intentions.

A business-development AI agent should not pretend to be the rainmaker. That would be a category error. The agent’s job is to keep the commercial thread alive so the human can spend more time in the moments where trust is won.

That is the Bob pattern.

What Bob watches

Bob starts with the firm's existing traces:

• meeting notes
• CRM records
• partner notes
• proposal drafts
• past introductions
• public company signals
• dormant accounts
• promised next steps

The first job is not to send more messages. The first job is to reconstruct context.

Useful context looks like this:

• who the relationship owner is
• why the relationship mattered
• what was discussed last time
• what was promised
• what changed since then
• what the next respectful action could be

Without that context, outreach sounds generic. With it, the partner can decide quickly.

What Bob produces

A good Bob output is not a wall of leads. It is a short operating queue.

For each opportunity, Bob should show:

• the person or company
• the relationship owner
• the last meaningful interaction
• the reason this is worth attention now
• the suggested next action
• the source evidence
• the confidence level
• the approval status

If Bob drafts a message, it should be a draft. The human approves it before it leaves.

What Bob should not do

Bob should not:

• invent familiarity
• overstate relevance
• send cold outreach without approval
• attribute revenue without evidence
• speak in the partner’s name without review
• chase every weak signal as if it were a deal

That last point matters. The goal is not volume. The goal is to catch high-quality follow-up that would otherwise disappear.

The approval gate

Every client-affecting action needs a human gate.

A simple Bob approval flow:

1. Bob identifies an opportunity.
2. Bob cites the source context.
3. Bob proposes the next action.
4. The partner approves, edits, rejects, or snoozes.
5. Bob records the decision.
6. If approved, Bob drafts or prepares the action.
7. The human sends or authorizes the final send.

This keeps the agent useful without letting it become reckless.

The learning loop

The agent improves from corrections.

If the partner rejects a suggestion because the timing is wrong, Bob should learn the timing rule. If the partner edits the tone, Bob should learn the voice. If an opportunity moves forward, Bob should learn which signals mattered.

This is where managed agents beat one-off prompts. The system does not just produce text. It accumulates operating memory.

What to measure

Bob should be measured against business-development operations, not against fantasy automation claims.

Good metrics:

• qualified opportunities surfaced
• approved follow-ups drafted
• dormant relationships reactivated
• time from signal to partner review
• follow-up delay after meetings
• pipeline influenced by Bob-supported actions

Be careful with revenue attribution. Bob can influence a deal without owning it. The human relationship still matters.

Where Bob fits in the AI Jungle stack

Bob is one PAIDA-style role: one managed digital employee around one business function. In a larger MAIDA cockpit, Bob can coordinate with EVA for meeting follow-through, INO for firm memory, and So-Fi for market signals.

That is the operating model we care about: small role first, measured loop, then coordination.

If this sounds close to your firm’s bottleneck, look at Bob or start with the AI Jungle Assessment. The first question is not "can an agent sell for us?" The better question is "which commercial thread are we already dropping?"