# NVG8

> NVG8 is an operator-led practice that turns the repeatable, manual work inside a revenue function into instrumented systems the client owns. It is led by Josh Weckesser, drawing on twenty years in go-to-market roles across technology and hardware, with fractional CRO/COO and RevOps partners on delivery. NVG8 operates a client's go-to-market as fractional leadership, then builds the systems that keep it running, across Sales, Marketing, Customer Success, RevOps, and Partnerships, on the client's existing stack (Salesforce, Slack, Google Cloud, AWS, Microsoft, Anthropic, OpenAI, Gemini).

Most AI for GTM is built by people who never carried the number. NVG8 works in sequence: Operate (fractional CRO, COO, or Chief Partner Officer running the motion alongside the team), Build (the repeatable work found while operating, turned into AI workflows, data pipelines, and copilots, sometimes starting with measurement such as scorecards and pipeline analysis), and Hand off (the client owns the systems, documented and versioned). Teams that already know what they want automated can start with the build.

The site has four main pages: [Operate](https://nvg8.ai/operate), [Build](https://nvg8.ai/build), [Field notes](https://nvg8.ai/field-notes), and [About](https://nvg8.ai/about).

## What we build

The [Build](https://nvg8.ai/build) page lists the workflows NVG8 builds, grouped by GTM team: Sales, Marketing, Customer Success, RevOps, and Partnerships.

- **Sales**: opportunity scoring & enrichment, pre-meeting briefings, meeting capture to action items, discovery-to-qualification briefs
- **Marketing**: LinkedIn research & posting workflows, iterative content generation, market & competitive intelligence, content-grade retrieval pipelines
- **Customer Success**: QBR & renewal prep briefings, recurring executive updates, quarterly & annual reviews, semantic search across account history
- **RevOps**: spreadsheet ↔ CRM sync, record deduplication, stakeholder attachment automation, Slack assistants for internal knowledge
- **Partnerships**: warm-path co-sell mapping, partner ecosystem intelligence, recurring partner updates, annual partner reviews

## How we work

- **Built on your stack.** Workflows fit your data and your team's ergonomics, not the other way around.
- **Owned by your team.** Code, prompts, and data stay with the client; versioned and documented.
- **Practical, not theatrical.** Narrow solutions that earn back hours every week before scope widens; pilots in weeks, measurable lift before quarter-end.

## Operate: fractional GTM leadership

NVG8 embeds as a fractional CRO, COO, or Chief Partner Officer to run sales, partnerships, and RevOps, not just build the tooling for it. Go-to-market moves through three stages, whether the motion is direct, through partners, or both: Establish (a go-to-market plan with a named target list, the first repeatable sales motion, a partner map if partners are part of the motion, the operating framework), Grow (one operating blueprint, the first systems that stretch a small team, a revenue plan per segment, channel, or partner tier), and Scale (resource allocation across motions, market expansion, the board story). Engagements run three to twelve months at up to two dedicated days per week, priced on duration and days.

See [Operate](https://nvg8.ai/operate) for the full description.

## How to work with us

Engagements start with a call, not a package. There are no tiers, cohorts, or seats.

1. **A call about the work.** Thirty minutes on what the team does by hand today and where it hurts. No deck.
2. **One workflow, scoped.** The narrowest workflow that earns back real hours, written up as a specification with the data it touches and what it costs to build. The client owns that spec either way.
3. **Build, then widen.** First workflow ships in weeks, not quarters, and has to earn its keep before scope widens. Code, prompts, and data stay with the client.

See [Build](https://nvg8.ai/build) for the full description. Contact is by email.

## Contact

- Email: info@nvg8.ai
- LinkedIn: https://linkedin.com/company/nvg8
- Website: https://nvg8.ai

## Field notes

Case studies and notes from inside the work. Honest accounts of what shipped, what didn't, and why.

- [Most partner plans stop at the partners you can see from inside Salesforce](https://nvg8.ai/field-notes/partner-map): Note. Ask an ISV for its partner plan and you hear two things: get the AEs to bring us in, and recruit the SIs. Neither earns more when the customer stays. The partners who pay an ISV (connected platforms, add-ons, routes to buy such as resellers, public sector contract holders, and marketplaces past the AppExchange) often show up only in finance actuals, not the CRM. The five kinds of partner NVG8 maps, with the SIs and the Salesforce field last. Published October 2026.

- [60 of 65 SIs brought us one deal. The long tail is the partner program.](https://nvg8.ai/field-notes/partner-long-tail): Note. Three years of partner data from a Salesforce ISV: 60 of 65 implementation partners showed up on exactly one deal, and 69 of 92 referring partners sent exactly one referral. Most partner programs are built for the few who come back; the work is making the first deal easy for everyone else with just-in-time enablement (questions to listen for, a customer like theirs, help on the deal itself) and letting the data pick out the repeat partners worth co-marketing investment. The partners most likely to repeat are often not SIs at all: connected platforms, add-ons, and resellers who earn as long as the customer stays. Published October 2026.

- [We gave an agent access to our inbox. It archived 78 emails instead of 12.](https://nvg8.ai/field-notes/agent-inbox-guardrails): Note. A local agent, running on Pi as the harness with Qwen 3.6-35b-a3b served over the LAN via LM Studio, was pointed at a working Gmail mailbox to archive automated clutter. The correct query matched 12 messages. A typo in the sender domain returned zero results, and instead of checking the query the agent broadened it twice, archiving 78 messages in one call plus two live project threads. It never ran the dry run the runbook required in bold. Central finding: an agent has three separable layers, the harness (the loop), the instructions (markdown playbooks and runbooks), and the actions (the tools that touch state), and only the action layer can refuse anything. Both safety rules lived in the instruction layer and did nothing. The fix was a count check inside the archive command that refuses above --max-messages 25. Also covers why the tools were written as plain command line programs rather than an MCP server: pipeable so the message body never enters the model's context at any size, runnable by hand so every documented command was verified before being written down, and portable across harnesses. Related finding: the gmail.compose OAuth scope does not prevent sending, contrary to a common assumption, and no Gmail scope permits drafting while withholding send. Published September 2026.

- [Muse Glimmer vs Qwen 3.8-27b on an Intel Arc Pro B70](https://nvg8.ai/field-notes/muse-vs-qwen-arc-b70): Note. A structured-extraction bake-off between two open-weight models, Muse Glimmer and Qwen 3.8-27b, run entirely on a local Intel Arc Pro B70 (32GB, $949) via LM Studio, because the source document carries confidential customer data that cannot go to a cloud API. Graded on nine mechanical checks including schema conformance, a hard character budget, and metric traceability. Muse Glimmer scored 9/9 in 107 seconds across three byte-identical runs; Qwen 3.8-27b scored 7/9 in 895 seconds and invented two customer facts by misreading vendor product-catalogue lines as the customer's deployed stack. Key findings: the larger model was 8.4x slower, generated 2.4x more tokens, and was miscalibrated in both directions; 88-93% of all generated tokens were reasoning scratchpad; time to first token fell ~70x after the prompt cache warmed. Published August 2026.

- [Eleven partners, 815 opportunities, and a scorecard the board could read](https://nvg8.ai/field-notes/partner-scorecard): Case study. An enterprise sales performance management software company had 10+ active SI partners and hundreds of millions in partner-influenced pipeline, but no way to measure the program or report it to the board. NVG8 built the partner taxonomy (sourced vs. influenced), a scorecard across 11 named partners, a pipeline analysis from the raw Salesforce export (815 opportunities segmented by partner, geography, industry, year), and the enablement layer (learning path, certification tracking, implementation tracker, partner welcome guide). At delivery: $47.5M partner-influenced active pipeline, $28M influenced closed won since FY23. Delivered May 2025 (with an initial 2022 partnership assessment).

- [Answers in Slack, grounded in your own docs](https://nvg8.ai/field-notes/slack-knowledge-bot): Case study. A B2B SaaS company needed staff to get accurate, source-grounded answers in Slack from internal docs and Salesforce Knowledge Articles. NVG8 built and shipped a production multi-tenant RAG platform: LlamaIndex ingestion (txt, PDF, Markdown, DOCX) into a Qdrant vector DB, source-cited answers in Slack, per-client ORG_PREFIX isolation, and a Salesforce Knowledge sync (OAuth 2.0, Bulk API, incremental SOQL change detection, auto re-indexing). Stack: FastAPI on Google Cloud Run, LlamaIndex, Qdrant, Slack API. Delivered February 2025.

- [Two data sources, one clean dataset, and the end of twenty-hour SI prep calls](https://nvg8.ai/field-notes/partner-ops-data): Case study. A Salesforce ISV had years of sales qualification notes and a CRM that recorded what happened but not why. A six-step AI workflow combined structured Salesforce data with unstructured sales notes, normalized industry categories, generated win stories in the Salesforce partner marketing format, and built a queryable intelligence layer reps could use before any partner meeting. SI pitch prep time: 20 hours → 5 minutes.

- [What it actually takes to launch a net-new ISV on the Salesforce AppExchange](https://nvg8.ai/field-notes/zinkt-salesforce-launch): Case study. A revenue operations software company had a working product and no Salesforce ecosystem presence. NVG8 embedded as fractional ecosystem lead and built the go-to-market from scratch: AppExchange listing, Salesforce field activation, a strategic ISV reseller partnership, and an SI partner program. Engagement: starting December 2021, ~28 months.

- [I needed one place for every meeting. So I built a recorder in chat.](https://nvg8.ai/field-notes/meeting-recorder): Case study. Calls happen across Google Meet, Zoom, Teams, phone bridges, and customer links; the useful record should live in one reviewable place. Built from empty folder to working prototype in 90 minutes, polished over ~14 iterations in two weeks. Audio stays local; one cloud call to AWS Bedrock; human in the loop by design. Stack: Swift, ScreenCaptureKit, parakeet-mlx, Claude Sonnet 4.5 on Bedrock, Google Calendar OAuth. Source: https://github.com/NVG8/recorder (MIT).

## Links

- [Home](https://nvg8.ai): Positioning (operate, build, hand off), the five teams NVG8 builds for, why operator-led matters, field notes, and how to start
- [Operate](https://nvg8.ai/operate): Fractional CRO, COO, or Chief Partner Officer. Why fractional, the Establish/Grow/Scale stages, scope and terms.
- [Build](https://nvg8.ai/build): How a build starts (a call, one scoped workflow, then the build), every workflow NVG8 builds grouped by team, who it is for.
- [Field notes](https://nvg8.ai/field-notes): Case studies and notes from inside the work.
- [About](https://nvg8.ai/about): Operator-led practice. Josh Weckesser's twenty years in go-to-market across enterprise hardware and the Salesforce ecosystem (EMC, Salesforce, Kenandy, Xactly, DocuSign), and how NVG8 operates go-to-market and then builds the systems that run it. Delivery team includes a fractional CRO/COO partner and a fractional RevOps partner.
- [Privacy](https://nvg8.ai/privacy): Privacy policy
- [Terms](https://nvg8.ai/terms): Terms of service
- [Full content (markdown)](https://nvg8.ai/llms-full.txt): Complete site content for direct ingestion