Goose Group: Applied AI for Operational Systems

Working document — March 2026

Thesis

Most mid-market companies have 3-5 internal processes that run on people manually moving data between systems. These processes are too specific for off-the-shelf SaaS and too "small" for big consultancies to care about. But they're now cheap and fast to solve with custom software because AI makes unstructured-to-structured data conversion trivial, and cloud primitives make event-driven pipelines commodity infrastructure.

The companies where this matters most are the ones whose differentiation is their service — where the people are the product, and the operational friction is what's holding those people back.

What we do

Build custom internal systems that sit on top of (and integrate with) a company's existing tools. These systems replace manual processes — email triage, spreadsheet tracking, copy-paste workflows — with structured pipelines and human-in-the-loop interfaces.

The AI components are specific and bounded: extraction from unstructured input, classification, routing, drafting, scoring. The rest is software engineering — queues, dashboards, notifications, integrations, state machines. AI makes it economical. The product is the operational system.

The human-in-the-loop dial is the key design pattern. Start with humans reviewing everything. As the system proves itself, the team dials down their involvement to where their judgment actually matters. They decide the pace.

The change pattern

This follows the same structural pattern as infrastructure migrations: incremental, capability-unlocking, and compounding.

Step 1 — Consolidate. Take a fragmented process and put it on real infrastructure. Multiple channels into one system. Shared visibility. Nothing gets dropped. This solves immediate pain and gives the team a single place to work from.

Step 2 — Add intelligence. Now that the process runs on proper infrastructure, AI capabilities are easy to layer in. Auto-extraction, completeness scoring, smart escalation, draft responses. Each one is a small capability that compounds. (Capabilities, not systems.)

Step 3 — Shift the balance. The team stops approving routine decisions and focuses on exceptions. "We approve the draft 9/10 times" becomes "we only review the complex ones." The system handles more, the team handles what matters. (Friction reduction as north star.)

Each step funds and justifies the next. The first build reveals adjacent problems worth solving.

Proof point: Motz365

Built for a $100M field services company (The Motz Group) through Substantial. Service requests arrived through voicemail, email, SMS, and phone calls to individual employees. The scheduler worked from email, spreadsheets, and memory.

What exists now: unified intake across four channels, AI transcription and extraction, completeness scoring, shared operator queue, automated customer confirmation, time-based escalation engine. Built in weeks by one person. The team uses it daily.

This is the template — not the intake/triage mechanics specifically, but the pattern: fragmented manual process → composable pipeline on real infrastructure → HITL interface → intelligence layered in incrementally.

GG's role vs. implementation

The expensive part of this work is figuring out what to build — sitting with the team, understanding the real process, designing the system, making product decisions. That's GG.

Implementation is where Spatialedge fits. They have 135 engineers and 25 PhDs who already build data pipelines, ML models, and AI systems for major banks, telecoms, and airlines. Their engineers think in terms of data flow, event processing, and pipeline orchestration — which is exactly what these operational systems are under the hood. Teaching ML engineers to build HITL web interfaces (dashboards, queues, approval flows) is a much shorter leap than teaching CRUD web developers to think about data architecture and systems design.

The rates work too. Spatialedge's current pricing ($20k/week) leaves significant room in engagements priced at market rates, which funds GG's product leadership layer without squeezing anyone.

The Spatialedge opportunity is bidirectional

GG → Spatialedge's engineers: We bring the product leadership and customer relationships. Their MLE's already know how to build notebooks, pipelines, and data processing systems. Extending that capability to build Motz-like operational systems (web UI + HITL workflows + AI pipeline) is a natural progression of what they already do, not a new skill set.

Spatialedge's existing customers → GG: Their current clients buy data engineering and ML pipelines. Those same companies almost certainly have manual operational processes that could benefit from this kind of work. A bank that hired Spatialedge to build a predictive attrition model also has intake processes, exception handling workflows, and coordination problems running on email and spreadsheets. GG's product leadership + Spatialedge's implementation = expanded scope on existing accounts.

This is the real leverage: GG makes Spatialedge's engagements bigger and stickier, and Spatialedge gives GG an implementation team that already understands the hard parts (data architecture, pipeline orchestration, ML integration) at rates that make the economics work.

Economics

Goal: A sustainable, high-margin, high-leverage business that's small on purpose. Autonomy in what we work on and who we work with. Interesting problems, not volume.

What that looks like: 2-3 active engagements at any time. Engagements priced at $80-150k/month, with GG's share covering product leadership (30-40% of engagement value). Implementation cost subcontracted to partners.

Revenue paths:

GG-sourced deals       We own the customer. Price the engagement.
                       Sub implementation to partners at their rates.
                       Margin is in the product leadership layer.

Partner-sourced deals  Spatialedge (or others) bring the customer.
                       GG does paid product leadership on their paper.
                       Or: commission on deals we help close/position.

Retainer model         Partner wants us as their customer-facing arm
                       (like current Substantial arrangement, but better).
                       $30-50k/mo retainer + success fees.

Key difference from Substantial: On GG-sourced deals, we own the customer relationship. We're not subcontractors.

GTM

The problem: Everyone is shouting AI. We're not trying to get people to use Claude more. We're building custom software people use every day.

The hook we need to figure out: What's the conversation starter that doesn't sound like every other AI consultancy? Current thinking:

Not "let us help you with AI"
Not "digital transformation"
Not "AI strategy"

More like:
"What does your team do manually that drives them crazy?"
"How do your customers reach you today?"
"What's the process that works because of one person,
 and what happens when they're on holiday?"

The sell is the prototype. Two weeks embedded, then a working thing they can click on. Not a deck. Not a roadmap. Software with their data in it. That's the close.

For 2-3 customers, not 100. This is network-driven. Alex's relationships, Mike's relationships, Juan/Spatialedge's existing accounts. The fastest path might be: ask Spatialedge which of their current clients has operational teams drowning in manual work. GG does discovery and product leadership, Spatialedge implements, the deal size doubles for everyone.

Platform account teams as a channel

Anthropic, AWS, Databricks, and similar all have the same problem: their account teams sit with customers who want to do something real with the platform, and the default referral is Accenture or Deloitte — which means 6 months of discovery, a massive SOW, and a team of 30 before anything gets built. The customer either balks at the price or gets buried in process.

GG is the alternative those account teams need. Small, fast, opinionated, actually builds things. When an AWS account team has a customer saying "we want to use Bedrock + Step Functions to automate our intake process" — they need someone who's already done it, not a consultancy that will spend 3 months scoping it.

Why this works for account teams:

They need customer success stories              We build them, fast
They need platform consumption                  Our systems run on their infrastructure
They need references that aren't Fortune 500    We work mid-market
They need partners who won't embarrass them     We're small enough to care about every engagement

Why this works for us: These account teams are talking to exactly the customers we want, every day. They're a warm introduction channel — not a sales partnership with contracts and tiers, just "hey, talk to these guys, they've done this before." We help their customer, the customer consumes more platform, the account team hits their number. Everyone wins.

This is a real channel worth investing in: get on the radar of 5-10 account teams at Anthropic/AWS/Databricks in our geography, show them the Motz365 case study, and let the referrals come.

Integration reality

These systems don't replace CRMs or ERPs. They sit alongside them and connect via APIs. Most modern platforms (Dynamics, Salesforce, SAP, ServiceNow) have solid integration surfaces. The pattern is: ingest from existing systems, add intelligence and workflow, write back. Customers who just invested in Dynamics aren't the wrong fit — they're often the best fit, because the gap between "we have a CRM" and "our process actually works" is exactly what we fill.

Open questions

Industry specificity vs. horizontal. The pattern is industry-agnostic. But selling is easier with "we do this for companies like yours." Do we need 2-3 reference verticals, or is the Motz case study + the pattern description enough for network-driven sales?

Spatialedge relationship structure. Agent agreement? Subcontractor? Joint venture on specific deals? Need to figure out what works for both sides before the call with Retief and Jacques.

Pricing validation. $80-150k/month feels right for mid-market (way under Deloitte, way above current Substantial rates). Need to test with 1-2 real conversations.

Alex's time allocation. How much of Alex is on GG vs. Substantial vs. other things? Same question for Mike. The model only works if we can commit real time to discovery and product leadership.