How to Start
A practical guide for CEOs and PE partners.
Executive summary · ~30 sec read
If you’re a CEO or PE partner ready to act: this is what action looks like in the next 30 days. Three engagement options — AI Readiness Diagnostic ($25K–$75K, 2–4 weeks) → Wedge Prototype ($150K–$400K, 8–12 weeks) → Company Brain Build ($500K–$2M+ for v1; $3M–$10M total over 18–24 months). Plus a workflow scoring matrix, a vendor comparison (where Glean, Copilot, Salesforce Agentforce land), the buyer FAQ, and the right rollout pattern for PE firms managing portfolios. No hidden pricing. No deck-only consulting.
Who this is for
For the operator who has read the manifesto and the playbook and is asking the only question that matters: what do we do this week? Three readers:
- The CEO, COO, or CHRO with budget, tired of pilots that don’t move EBITDA.
- The PE operating partner deploying this pattern across a portfolio rather than one portco at a time.
- The board-level sponsor trying to decide what to do.
If you’re earlier — still exploring whether AI matters or building literacy — the manifesto and playbook are the starting points. Come back when ready to commit a wedge.
The CEO’s First 30 Days
Week-by-week sequence for a substrate-first build instead of another scattered pilot.
| Week | Action | Output |
|---|---|---|
| Week 1 | Appoint an executive sponsor (CEO, or a COO direct report with org pull). Identify three candidate workflows where the knowledge-fragmentation gap is most painful. Freeze any new AI pilot procurement. | A named owner; three candidates; a hold on new tool buys. |
| Week 2 | Run knowledge-fragmentation interviews with department heads. Where does the work actually live? Where does it leak between systems? What lives only in heads? | Tagged inventory of fragmentation by function. |
| Week 3 | Score the three candidate workflows on volume, knowledge intensity, measurable value, risk, data access, and human owner (matrix in §5.5). | A scored shortlist with one clear winner. |
| Week 4 | Pick one wedge. Approve the prototype budget. Brief the leadership team on the substrate-first build sequence so the People Track starts on day one, not month six. | Approved budget; named wedge; aligned leadership. |
Four weeks, four artifacts. No code, no vendor, no headcount beyond a sponsor. The first 30 days replace “should we do AI?” with “which workflow first, and who owns the prototype?”
What you stop doing. Funding more disconnected AI pilots. Buying enterprise AI seats without a workflow target. Treating this as a CIO-led IT project. Each is a way of looking busy without committing to a build sequence that compounds.
What you tell your board. “We’re moving from scattered pilots to a substrate-first build. First measurable P&L impact in 6–9 months on one workflow. Total program ~$3M–$10M over 18–24 months for a mid-market company, with embedded engineers transferring capability to an internal team by year three. Next decision: wedge approval at end of week four.”
That sentence is the governance frame. If the board pushes back on timeline or spend, the conversation to have is the one in the playbook — not the one about another pilot.
Do You Even Need a Company Brain?
Not every company is ready for a Company Brain in 2026. Two lists, one scorecard, a clear answer.
You probably do need a Company Brain if:
- Critical decisions live in Slack, email, meetings, and people’s heads.
- Employees ask the same context questions repeatedly.
- New hires take 3+ months to become productive.
- AI pilots fail because they lack company-specific context.
- Customer / account history is scattered across multiple tools.
- Leadership cannot cleanly explain how work gets done across functions.
You probably do not need it yet if:
- Your goal is individual productivity gains (Copilot is enough).
- Your workflows are already structured and well-documented.
- Your biggest gap is basic AI literacy across the team.
- You don’t have an executive sponsor.
- You can’t yet name a workflow where better context would create measurable value.
A short readiness scorecard. Six yes/no questions. Your recommendation appears as you answer.
Click Yes or No for each question. The score and recommendation update live.
1. Do critical decisions live primarily in Slack, email, meetings, or people’s heads?
2. Do employees ask the same context questions repeatedly?
3. Do new hires take 3+ months to become productive?
4. Have prior AI pilots failed because they lacked company-specific context?
5. Is customer / account history scattered across multiple tools?
6. Does leadership struggle to explain how work actually gets done across functions?
Your readiness score
— / 6
Answer all 6 questions to see your recommendation.
The scorecard is short because the diagnosis is rarely subtle. Companies that need a Company Brain feel the symptoms across every function, every week. Building the substrate ahead of demand doesn’t compound — it sits idle.
Three Ways to Start
The full transformation runs $3M–$10M over 18–24 months. Most companies aren’t ready to commit on day one. Three engagement paths, each scoped to where you actually are.
| Path | For whom | Timeline | Cost | Output |
|---|---|---|---|---|
| AI Readiness Diagnostic | Unsure CEO / PE partner who needs the map before the build | 2–4 weeks | $25K–$75K | Workflow map, fragmentation map, AI readiness score, top 5 workflow opportunities, recommended wedge, build/buy/hire plan, 90-day execution roadmap, board-ready summary |
| Wedge Prototype | Company with one obvious workflow pain and an executive sponsor | 8–12 weeks | $150K–$400K | Working internal agent on a single high-volume workflow, baseline + measured impact, decision on whether to scale to full Company Brain build |
| Company Brain Build | Company committed to AI-first transformation | 6–9 months v1; 18–24 months full | $500K–$2M v1; $3M–$10M total | Production substrate, 1–3 production workflows, embedded engineering team, capability transfer to internal team |
No “Enterprise” tier with hidden pricing. The ranges are real. Diligence-grade modeling against your company comes after the diagnostic — that’s the point of the diagnostic.
The three paths are sequential, not alternatives. Most engagements run Diagnostic → Wedge Prototype → Company Brain Build, with go/no-go between each phase. Companies that already know their wedge can skip the diagnostic; ones that have run a successful internal pilot can enter at the Build tier.
The right starting point is the smallest one that answers your real question. Which workflow first → diagnostic. Will this actually move the metric → wedge prototype. How do we operate as an AI-first company → full build.
How to Pick Your First Workflow
The wedge decision is the most consequential call in the first 30 days. Pick well and the prototype produces a P&L outcome that funds the rest. Pick poorly and the program stalls. Six criteria, scored 1–5.
| Criterion | Question to ask | High score (5) means |
|---|---|---|
| Volume | Does this happen hundreds or thousands of times per month? | High volume — small per-event lift compounds. |
| Knowledge intensity | Does the work depend on scattered context that an LLM without your data can’t answer? | Yes — substrate value is highest here. |
| Measurable value | Can we tie this to revenue, margin, cycle time, churn, or headcount leverage? | Direct, quantifiable line to P&L. |
| Risk | Can errors be reviewed before customer impact? | Internal-first or human-in-the-loop is feasible. |
| Data access | Is the needed context available or capturable? | Sources are reachable; permissions clear. |
| Human owner | Is there a business leader who already owns the metric? | Named, accountable, willing. |
20+ total signals a strong wedge candidate. Lower than that, and the prototype is fighting the wrong constraint — usually missing data access or owner, both of which surface in a diagnostic before they sink a build.
Common high-score workflows by function:
- Sales
- call prep, proposal generation, account-management recommendations, win/loss analysis.
- Customer service
- triage and routing, refund eligibility checks, mediation, knowledge-base drafting.
- Legal / contracts
- redline review, clause comparison against precedent, risk flagging.
- Finance / ops
- variance analysis, vendor-spend analysis, close-cycle automation.
- Compliance
- filing review, anomaly detection, audit prep.
- HR
- onboarding curation, policy Q&A, internal mobility match.
- Engineering
- incident triage, runbook generation, on-call summary.
The right wedge is the one with the highest matrix score and an existing owner already frustrated with how the work happens today. Frustration converts to adoption faster than any change-management plan and survives the first messy weeks when the agent gets things wrong before it gets them right.
Why Tools Alone Won’t Make You AI-First
By 2026, every enterprise software vendor sells an “AI” SKU. Most are useful. None of them, on their own, makes a company AI-first.
| Option | What it solves | What it does not solve |
|---|---|---|
| Microsoft Copilot | Individual productivity in Office | Workflow redesign, institutional memory, agent automation |
| Glean | Enterprise search across SaaS | Decision curation, agent platform, operating-model change |
| Salesforce Agentforce | CRM-bounded automation | Cross-system context, non-CRM workflows |
| Notion AI | Document and workspace assistance | Permissioned company-wide substrate, regulated data, deep system integration |
| ServiceNow Now Assist | IT and ops workflow automation grounded in CMDB | Cross-functional knowledge layer, customer-facing workflows |
| Custom agents (LangChain, etc.) | Workflow automation | Durable, queryable knowledge layer (unless you build it intentionally) |
| The Anthropic / Blackstone / Goldman JV | Forward-deployed engineers redesigning workflows around Claude | The substrate underneath the workflows |
| Company Brain | Institutional memory + agent substrate that everything else plugs into | Comes with implementation cost; isn’t off-the-shelf |
Read the table as a stack, not a bake-off. Each tool moves a specific number. None produces institutional memory that compounds across functions, and none rewires the operating model.
Tools are components. The substrate is the system. Most companies are buying components and hoping they assemble. They don’t.
The companies whose AI work shows up in earnings calls rather than press releases built the substrate first and let the components plug into it.
What Synaptrix Does / Doesn’t Do
Where we come from. Synaptrix is built by operators who’ve shipped this playbook at production scale. Outcomes delivered inside operating companies:
- Engineering throughput up 18% YTD with the same headcount
- Product cycles compressed from 10–12 weeks to 3–4 weeks
- Agentic systems in production taking actions in underlying systems (not chatbots)
- ~28% infrastructure cost reduction maintained through the transformation
We do the work because we’ve already done it.
Specific proof points — companies, dates, references — are part of direct conversations under NDA.
What we do:
- AI Readiness Diagnostics (the entry tier in §5.4)
- Company Brain architecture (graph + integrations + curation jobs)
- Embedded engineering teams (3–8 forward-deployed engineers, 6–24 months)
- First-wedge implementation with measured P&L impact
- Executive operating-model design and People Track program management
- Capability transfer — explicit goal of phasing out vendor presence by month 18–24
What we don’t do:
- Generic AI training without a workflow target
- Deck-only transformation programs. Diagnostic engagements produce an execution plan; build engagements ship production code.
- Tool resale dressed up as transformation
- Customer-facing agent rollouts before the internal substrate is proven
- Lock-in. The substrate is the company’s IP — your repos, your infrastructure, your team can run it without us.
Who works on your account: named senior engineers and an executive sponsor — not a rotating bench of analysts.
Buyer FAQ
The questions sophisticated buyers actually ask. Direct answers.
“We bought Copilot already. Is this still relevant?”
Yes. Copilot is individual productivity; the Company Brain is institutional memory. They complement, not substitute. Copilot makes each employee 20% faster at existing work. The Brain redesigns the work.
“Can our IT team do this?”
Some of it, eventually. Foundations are off-the-shelf; substrate work is custom; the People Track is leadership. IT can run the platform once built, but most IT teams haven’t built one before — that’s where embedded engineers compress the timeline.
“Do we really need a knowledge graph? Can’t we just use vector search?”
Vector search retrieves; it doesn’t reason about relationships. “Which contracts mention vendor X, were signed by Sarah, and have an auto-renewal expiring this quarter?” is a graph query — vector retrieval can’t answer it cleanly. Most production substrates use both: graph for entities, vectors for unstructured retrieval.
“What if our employees don’t document anything?”
The GIGO question (playbook §4 Step 1). The audit and build are the documentation. Klarna ingested undocumented Slack threads through curation, not pre-cleaning. Don’t wait for documentation — build the substrate that creates it.
“How do you handle permissions?”
Permission-aware retrieval inherits from source systems (Slack, Salesforce, Drive). HR/legal/M&A/board sources gated by default. Every query logged; every answer cites sources. Detailed treatment in playbook §4 Step 2.
“How soon should we expect ROI?”
First measurable workflow impact in 6–9 months on a well-scoped wedge. Full operating-model EBITDA lift in 18–24 months. Observed ranges, not promises — variance comes from data quality, sponsorship, and wedge choice. Anyone selling fixed ROI hasn’t done it.
“Is this only for large companies?”
There’s a floor. Below ~50 employees, fragmentation is too small to justify the build. 50–500 is the sweet spot for ROI velocity. 500–2,000 is where this becomes essential. Above 2,000 the math always works; the question is sequencing.
“Who owns the IP if you build it?”
You do. Substrate, curation rules, integrations, agent code — all your IP, in your repo, on your infrastructure. We’re not licensing a product back. By month 18–24, your team runs it.
“What if we fire you in month 24?”
The Brain keeps working. You own the code, the data, the curation rules, and (by then) the team. The vendor relationship phases to advisory, not full operations.
“What does the team look like in year 3?”
~1 AI/product lead, 2 backend/data engineers, 1 agent engineer, 0.5 FTE CHRO for curation governance. ~$1.2M–$2M/year for a 500-person company. Detail in playbook §4 Step 5.
“How is this different from McKinsey / BCG / Bain?”
They write strategy and hand you a deck. We build the system and hand you a working substrate. Either of them can complement on operating-model redesign (§4 Step 6), but they don’t ship engineering, and you need engineering.
“What should PE firms do differently from individual companies?”
Standardize the substrate pattern first, then deploy across portfolio. One PE firm running the 6 steps across 20–80 portcos generates compounding operating alpha no individual portco can match.
For PE Firms: The Right Rollout Pattern
The biggest mistake PE firms make with portfolio AI: pushing the same stack across every portco at once. The right pattern is staged, opinionated, selective.
The 6-move PE rollout pattern:
- Pick 1–2 portcos as pilots. Highest-readiness (§5.3 scorecard), strongest operating-partner trust with the CEO. Pilots prove the pattern before scaling.
- Run diagnostics across 3–5 portcos in parallel. Identify recurring workflows (sales call prep, support triage, contract review, finance ops are usually portfolio-wide) and map fragmentation patterns. Separate where a shared substrate compounds from where portco-specific stays portco-specific.
- Build a reusable reference architecture. Standardize the substrate at the pattern level — graph schema, integration connectors for common SaaS (Slack, Salesforce, Jira, Notion, Google Workspace), curation cron architecture, eval framework, security model. Portcos instantiate the pattern, not rebuild it.
- Keep source systems and workflows portco-specific. Portcos still have their own CRM, ERP, regulated data, and customer base. The reference architecture is the pattern; each portco runs its own instance configured for its industry and stage.
- Centralize at the PE-firm level: playbooks, vendor management, eval patterns, security baseline, transformation governance. This is where compounding alpha lives. One Anthropic enterprise contract, one security baseline, one change-management playbook across 30 portcos. Cost amortizes; learning compounds.
- Scale only after one portco shows measurable workflow impact. The first portco that proves the wedge becomes the case study. Others opt in via operating-partner conversations, not top-down mandate. Operating partner credibility > corporate-mandate fatigue.
Funding model. PE firm funds diagnostics across portcos (low risk, high information); the build is funded by the portco (their P&L impact). The firm captures operating alpha at the portfolio level through the standardized pattern, not by controlling portco budgets.
What gets centralized vs. portco-specific:
| Centralized at PE level | Portco-specific |
|---|---|
| Reference architecture | Source systems and workflows |
| Vendor contracts (LLM APIs, vector DBs, eval tools) | Wedge workflow selection |
| Security baseline and audit framework | Internal team structure |
| Transformation playbook + change-management materials | Operating-model redesign decisions |
| Cross-portco eval and benchmark patterns | P&L attribution and accountability |
Talk to us
If your company is ready to start — or if you’re a PE partner thinking about portfolio rollout — talk to us. The first conversation is a 45-minute call to figure out which of the three engagement paths fits, or whether you’re not ready yet (and we’ll tell you).