# Matthew Carlson Consulting > Matthew Carlson is an AI Solutions Engineer and Forward Deployed Engineer in Madison, Wisconsin who designs, builds and runs production AI systems for marketing teams, agencies and professional-services firms across the U.S.: AI agents, voice and chat assistants, CRM automation and reporting pipelines. Before AI he spent 11 years at Duluth Trading running a $40M+ digital marketing budget that drove $200M+ in annual revenue. Key facts: AI Solutions Engineer and Forward Deployed Engineer (he embeds with the client's team and builds inside their systems and data; deliverables are working systems the team owns); Madison, Wisconsin; works nationally by video; solo practice (Matthew Carlson Consulting, founded 2024); currently AI Lead Consultant for a marketing agency (fractional-CTO scope, since March 2026). Client names are confidential. Every project is labeled Live, Pilot, Tested, Delivered or Shipped. --- This file is the full text of the site's main pages, generated from the live HTML. A shorter index is at https://matthewcarlsonconsulting.com/llms.txt. --- # Matthew Carlson | AI Solutions Engineer Source: https://matthewcarlsonconsulting.com/ ## I build the AI systems that run your marketing operations. AI agents, voice and chat assistants, CRM automation and reporting pipelines, designed by someone who ran a $40M marketing budget and shipped with tests, verification gates and a known cost per run. I work [forward deployed](https://matthewcarlsonconsulting.com/faq.html#forward-deployed): inside your systems, with your team, until it runs. - **33**: AI agents in one production client-intelligence platform - **104**: packaged AI skills in a published library - **12–15 hrs**: of manual ad reporting replaced every week - **30 days**: concept to launch, 20+ page enterprise website, solo - **$40M+**: annual marketing budget managed over 11 years ### Six ways AI goes from pilot to production Each one ships with a written specification, a test plan and a cost model. You own everything I build. [All solutions](https://matthewcarlsonconsulting.com/services.html) - [01**AI agents & workflows**: Multi-agent systems, MCP servers and skill libraries that do real work on a schedule or a trigger.→](https://matthewcarlsonconsulting.com/services.html#ai-agents) - [02**Voice & chat AI**: Phone and chat agents that answer, qualify, book and hand off, tested before a customer hears them.→](https://matthewcarlsonconsulting.com/services.html#voice-chat-ai) - [03**CRM automation**: Speed-to-lead, AI-researched onboarding and CRM builds generated from a spec.→](https://matthewcarlsonconsulting.com/services.html#crm-automation) - [04**Reporting & ad-ops automation**: Google Ads scripts, nightly account-health checks and client report packets that build themselves.→](https://matthewcarlsonconsulting.com/services.html#reporting-automation) - [05**AI readiness & roadmaps**: Where AI fits, what to buy versus build, and the order to do it in. Fractional AI lead available.→](https://matthewcarlsonconsulting.com/services.html#ai-strategy) - [06**AI-ready websites**: Sites built to rank on Google and be cited by ChatGPT, Perplexity and Gemini.→](https://matthewcarlsonconsulting.com/services.html#ai-ready-web) ### AI systems already doing the work A few of the systems I've built, each marked live, in pilot, tested or delivered. [Public sources feed one 26-section client profile, which 33 agents and an MCP server read. Status: Live. AI agents #### Multi-agent client intelligence platform One sourced memory of every client, shared by 33 AI agents and reachable from Claude through an MCP server. 39 client profiles · 8-tool MCP server · ~$32–42/month to run](https://matthewcarlsonconsulting.com/projects.html#client-mind) [Status: Tested. AI agents · CRM #### Agents that build inside the CRM Claude and OpenAI Codex construct workflows, pipelines and knowledge bases directly in the platform from written specs. 3.5-hour unattended run · 351 custom values · verified at every step](https://matthewcarlsonconsulting.com/projects.html#agents-that-build) [Status: Tested. Voice AI #### Inbound voice AI agent family Phone agents that answer, qualify, book showroom visits and transfer with a spoken summary. 11 of 11 first-block checks · 50+ case test suite · 6 service branches](https://matthewcarlsonconsulting.com/projects.html#voice-ai) [Status: Live. Reporting #### Account health dashboard & alert bridge Every client CRM account checked nightly; problems become tasks in the tool account managers already use. 18 health flags · read-only · Monday digest email](https://matthewcarlsonconsulting.com/projects.html#account-health) [See all projects →](https://matthewcarlsonconsulting.com/projects.html) ### Engineering habits that make AI safe to depend on 01 #### Spec first A written specification defines every object before anything is built. The spec and the gap report are deliverables. 02 #### Test before demo No client hears a voice agent until a 50+ case test suite and a stop-ship regression pass. 03 #### Read-only by default Reporting systems observe and change nothing. Agents that write run in test environments first. 04 #### Stop and notify When an input looks wrong, the system halts and tells a person how to restart. It doesn't guess. 05 #### Meter every run Token, model and crawl costs are tracked per run, so AI becomes a known unit cost you can price. 06 #### Buy the commodity If a tool already does it well, buy it. Build only where the interpretation is your advantage. ### From conversation to running system #### Discovery A working session on the process, the data and what success is worth in hours or revenue. #### Specification A written spec, test plan and cost model you approve before the build starts. #### Build with verification Built in a test environment, checked against the spec, with a gap report at each phase. #### Launch & hand off Pilot, monitor, document and train your team, or keep me on to run and improve it. ### One question, answered properly Short articles from the work, plus a 20-question checklist that scores whether a process is ready for AI. [Voice AI #### How do you keep an AI voice agent from quoting prices? A written list of what it must never say, a safe action for each, and a test suite before anyone hears it.](https://matthewcarlsonconsulting.com/insights/voice-agent-guardrails.html) [AI agents #### What does an AI client research pull cost, and why meter it? About $4–8 per client, measured. Metering is what makes AI priceable.](https://matthewcarlsonconsulting.com/insights/cost-per-research-pull.html) [Checklist #### Is this process ready for AI? Twenty yes-or-no questions across process, data, systems, people and risk, scored out of 100.](https://matthewcarlsonconsulting.com/ai-readiness-checklist.html) --- # About Matthew Carlson | AI Solutions Engineer, Madison WI Source: https://matthewcarlsonconsulting.com/about.html ## AI Solutions Engineer. Former $40M marketing operator. I build production AI systems for marketing teams, agencies and professional-services firms. Before AI, I spent 11 years at Duluth Trading running a $40M+ digital marketing budget that drove $200M+ in annual revenue, owning analytics, attribution, paid media and data feeds. That's why every system I build is scoped against hours saved, revenue and risk, and why it ships with tests and a cost model instead of a demo. Since 2024 I've been building production systems with AI. Today I lead AI strategy and implementation for a marketing agency, where the work covers a 33-agent client-intelligence platform, voice agents, nightly monitoring of every client account, and coding agents that build CRM systems from written specs. I also take on independent engagements, most recently a 30-day website launch and AI roadmap for a healthcare IT consultancy. I work as a **Forward Deployed Engineer**: embedded with your team, in your systems and data, building the solution in your environment and iterating with the people who will run it. The deliverable is a working system your team owns, not a recommendation. [What that means in practice →](https://matthewcarlsonconsulting.com/faq.html#forward-deployed) An outside consultant delivers recommendations and leaves; a forward deployed engineer builds inside the client's systems and hands off a running system. RoleAI Solutions Engineer and Forward Deployed Engineer BuildsAI agents, MCP servers, voice and chat assistants, CRM automation, reporting pipelines Background11 years at Duluth Trading running a $40M+ digital marketing budget ($200M+ revenue) CurrentAI Lead Consultant for a marketing agency (fractional-CTO scope); independent engagements BasedMadison, Wisconsin; clients across the U.S. I work from Madison, Wisconsin, with clients across the U.S. ### Eleven years on the business side of marketing Results from Duluth Trading, 2014–2025. - **$200M+**: annual revenue from a $40M+ digital budget - **25%**: incremental lift proven with a holdout test - **2×**: Facebook revenue growth ($3.5M to $7M+) while cutting ad ratio from 60% to 35% - **$33M**: affiliate program revenue in 2023 - **$8.3M**: in sales from a retargeting channel I launched ### Path Mar 2026 – present #### AI Lead Consultant, marketing agency AI strategy and implementation with fractional-CTO scope: agent platforms, voice AI, CRM automation, reporting and vendor evaluation. Jun 2026 – present #### Website & AI Consultant, healthcare IT consultancy Solo website redesign and launch, post-launch operations and an AI automation roadmap. 2024 – present #### Founder, Matthew Carlson Consulting Custom AI applications, analytics platforms and automation, including SkillEngine and a media mix model platform. 2014 – 2025 #### Senior Marketing & BI Analyst, Duluth Trading $40M+ budget across 15+ channels; analytics, attribution, data feeds, SEO and executive reporting. 2006 – 2009 #### Technical Services, Epic Systems Healthcare software deployments: workflow analysis, requirements and end-user training. ### What I bring - **Agentic systems & MCP:** multi-agent platforms, skill libraries, orchestration with QA gates - **Voice & conversational AI:** dialogue design, knowledge bases, test suites - **CRM & marketing automation:** GoHighLevel, HubSpot-class platforms, APIs and webhooks - **Analytics & reporting automation:** attribution, incrementality testing, Google Ads scripts - **Full-stack development:** Python, TypeScript, React, Supabase, PHP - **AI-ready web & search:** accessible builds, SEO, answer-engine and generative-engine optimization #### Credentials - MCP servers in production - Claude API & skill authoring - GoHighLevel AI Studio & Voice AI - Autonomous coding agents - WCAG AA engineering - Google Ads certified - University of Wisconsin–Madison ### Principles that came from shipping, not slides #### Gate before you scale One graded sample and a consistency check cost hours. Regenerating dozens of outputs from a bad source costs days. #### Confidence is a field Every AI-researched fact is labeled verified, derived or needs input, so people know what they can trust. #### Verify the agent's boundary Autonomous builds run only in named test environments, with location checks and API readback every run. #### Report where the work happens Alerts land as tasks in the tool people already use, not in another dashboard they have to remember. #### Test suites make demos safe A voice agent meets a client only after it passes its test cases and the safety regression. #### Price build and run separately Launch, build, maintenance and usage are separate lines, so nobody is surprised later. --- # AI Solutions: Agents, Voice AI, CRM & Reporting Automation Source: https://matthewcarlsonconsulting.com/services.html ## Six ways I put AI into production Every engagement ships with a written specification, a test plan and a cost model. You own everything I build. [AI agents](https://matthewcarlsonconsulting.com/services.html#ai-agents) [Voice & chat AI](https://matthewcarlsonconsulting.com/services.html#voice-chat-ai) [CRM automation](https://matthewcarlsonconsulting.com/services.html#crm-automation) [Reporting](https://matthewcarlsonconsulting.com/services.html#reporting-automation) [AI readiness & roadmaps](https://matthewcarlsonconsulting.com/services.html#ai-strategy) [AI-ready websites](https://matthewcarlsonconsulting.com/services.html#ai-ready-web) ### AI agents & workflow engineering Multi-agent systems, skill libraries and orchestration pipelines that do real work on a schedule or a trigger, with shared memory, QA gates and a known cost per run. Status: Live A client-intelligence platform with 33 agents, an 8-tool MCP server and 39 client profiles, operating for about $32–42 a month. [Read the case study](https://matthewcarlsonconsulting.com/projects.html#client-mind) Public sources feed one 26-section client profile, which 33 agents and an MCP server read. #### What's included - Multi-agent platforms with a shared, sourced memory of each customer or client - MCP servers that connect Claude to your own data and tools - Skill libraries with versioning, evaluation and a searchable catalog for your team - Orchestration pipelines with validation gates, auto-remediation and cost telemetry - Coding agents (Claude Code, OpenAI Codex) building inside SaaS platforms from a written spec - Webhook-triggered agents that research, write and post results back into your systems ### Voice & conversational AI Phone and chat agents that answer, qualify, book and hand off, built around your routing and tested before a customer ever hears them. Status: Tested An inbound voice agent family for a multi-location retailer: 11 of 11 first-block checks passed, with live calendar booking verified on a call. [Read the case study](https://matthewcarlsonconsulting.com/projects.html#voice-ai) An inbound call reaches the voice agent, which books, transfers, takes a message or sends a follow-up text. #### What's included - Inbound voice agents with booking, message taking and warm transfer with a spoken summary - Website, SMS and social chat assistants writing to the same CRM record - Knowledge-base architecture shared cleanly across brands and bots - Consent capture, safety rules and clear limits on what the agent will say - 50+ case test suites and stop-ship regression checks before launch - Pricing that separates launch, build and maintenance, with usage billed at cost ### CRM & marketing automation GoHighLevel and HubSpot-class systems that respond to leads in under a minute, onboard clients with AI research, and build themselves from a specification. Status: Tested A 10-workflow speed-to-lead engine, built by Claude from a written spec; the first clean test sent a text and an email in under a minute. [Read the case study](https://matthewcarlsonconsulting.com/projects.html#agents-that-build) A written spec is built by Claude or Codex inside a CRM test location, checked by API readback, and gaps loop back to the spec. #### What's included - Speed-to-lead engines with business-hours, compliance and response-time checks - AI-researched client onboarding: the client corrects a profile instead of filling out a questionnaire - Sub-account provisioning, snapshots and custom values at scale - Lifecycle and purchase-driven campaigns from your order or POS data - Webhook, API and MCP integrations between your CRM and AI tools - Automated QA that enrolls test contacts and checks every workflow path ### Reporting, analytics & ad-ops automation The reporting and optimization work your team does by hand every Monday, automated, from someone who ran weekly business reporting for a $40M+ marketing program. Status: Live A 15-script Google Ads automation suite running across an agency's whole portfolio, replacing about 12–15 hours of manual reporting a week. [Read the case study](https://matthewcarlsonconsulting.com/projects.html#google-ads-suite) Every client CRM account is read nightly by a read-only engine with 18 flags, producing CRM tasks, a dashboard and a Monday digest. #### What's included - Google Ads scripts at manager-account level: audits, budget pacing, search-term and wasted-spend analysis - Nightly account-health monitoring with alerts written back into your CRM as tasks - Client report packets generated from your Looker Studio dashboards - Weekly meeting decks built from live data - Attribution architecture: click-ID capture with the CRM as the join point - Pipeline and portfolio analysis that turns raw CRM data into decisions ### AI readiness, roadmaps & fractional AI lead Where AI fits in your business, what to buy versus build, and the order to do it in. Then, if you want, I lead the work. Status: Delivered An AI automation roadmap and systems-integration readout for a healthcare IT consultancy's leadership and finance teams. [See the project](https://matthewcarlsonconsulting.com/projects.html#roadmap) A two-week readiness sprint scores five domains, maps opportunities, decides build versus buy, and ends in a 90-day roadmap. #### What's included - AI readiness audit scoring 5 operational domains out of 100 - Opportunity map ranked by hours saved, revenue impact and risk - Build-vs-buy evaluations with a clear recommendation - Vendor, tool and builder selection, including paid screening tests - Team enablement: skill libraries, playbooks and training - Fractional AI lead or CTO scope on a monthly basis ### AI-ready websites & search visibility Websites built to rank on Google and to be cited by ChatGPT, Perplexity and Gemini, with accessibility engineered in from the first component. Status: Shipped A 20+ page site for a healthcare IT consultancy, concept to live in 30 days, solo; it passed 1,000 organic clicks per 28 days within four weeks of launch. [Read the case study](https://matthewcarlsonconsulting.com/projects.html#website-launch) Design and build, content and QA, and a one-evening launch led to 1,000 organic clicks per 28 days within four weeks. #### What's included - Hand-coded, fast builds with no page-builder bloat - WCAG AA accessibility audited before launch - SEO, answer-engine and generative-engine optimization from day one - LLM visibility audits across ChatGPT, Gemini and Claude, scored 0–100 - llms.txt, structured data and citable, question-led content - Launch runbooks with redirect maps and instant rollback ### Common starting points Systems I've already designed and tested. Each one can be scoped in a single call. #### AI lead research report A new lead triggers an agent that researches the business and briefs your team before the first call. #### Speed-to-lead responder Every form, call and chat answered in under a minute, then qualified and booked. #### Voice receptionist Answers, books, takes messages and transfers with a summary when a person is needed. #### Research-then-verify onboarding AI pre-fills the client profile; the client corrects it in minutes instead of filling out forms. #### Nightly account health Silent forms, stalled deals and waiting leads flagged every morning as tasks. #### Monthly report packet Dashboards in; an internal report, a client report and a meeting agenda out. #### Website asset loader Logo, colors, fonts and photos pulled from a site and filed into your CRM in minutes. #### Missed-call text-back An unanswered call gets a text and an AI qualifier instead of a lost lead. ### Four ways to work together Launch, build and maintenance are priced separately, and AI usage is billed at cost, so you always know what you're paying for. Not sure where you stand? [Take the 20-question readiness checklist](https://matthewcarlsonconsulting.com/ai-readiness-checklist.html). Duration: 2 weeks #### AI Readiness Sprint Scored audit, opportunity map, build-vs-buy calls and a 90-day roadmap. Duration: 2–6 weeks #### Build Sprint One system specified, built, tested and handed off with documentation. Duration: Monthly #### Fractional AI Lead Ongoing builds, operations and team enablement with a set number of days each month. Duration: Monthly #### Managed AI Operations Monitoring, maintenance and iteration of the systems I built for you. Claude API, Claude Code & MCPOpenAI API & CodexGemini GoHighLevelHubSpotSupabase / Postgres PythonTypeScript & ReactGoogle Ads API & Scripts GA4 & Looker StudioFirecrawlWordPress --- # Work: AI Agent, Voice AI & Automation Case Studies Source: https://matthewcarlsonconsulting.com/projects.html ## AI systems already doing the work Agents, voice assistants and automation built for real businesses, with the numbers behind each one. **Live** is in daily use, **Pilot** is with a first client, **Tested** has passed a written test plan, and **Delivered** is a finished analysis or plan. Client names are withheld. - Live / Shipped - Pilot - Tested - Delivered ### Case studies Eight systems, each with how it works, what it took and the measured result. Status: Live. AI agents · Marketing agency · 2026 #### Multi-agent client intelligence platform One structured, sourced memory of every client, shared by 33 AI agents and reachable from Claude through an MCP server. Public sources feed one 26-section client profile, which 33 agents and an MCP server read. - **33**: specialized AI agents - **39**: client profiles in production - **8 tools**: in the MCP server - **~$32–42**: per month to operate ##### The problem Every AI output at the agency started from a blank prompt. Client facts lived in people's heads and drifted, so content came back generic or wrong. ##### Results Commercial alternatives run $49–249 per user per month; this runs the whole agency for about $32–42. Its QA engine has generated 1,504 field-level suggestions for human review. ##### What I built - A 26-section client schema with field-level provenance - 33 agents reading from that shared memory - A prompt library with versioning: 92 prompts, 94 tracked versions - A document layer with embeddings, and a CRM integration that has cataloged 2,590 assets - An MCP server with token auth and audit logging Stack: TypeScript, Supabase / Postgres, Claude API (tiered model routing), MCP Status: Tested. AI agents · CRM · Marketing agency · 2026 #### Agents that build: CRM construction from a spec with Claude & Codex AI coding agents built a marketing agency's CRM systems directly inside the platform from written specifications, verified at every step. A written spec is built by Claude or Codex inside a CRM test location, checked by API readback, and gaps loop back to the spec. - **3.5 hrs**: unattended Codex build run - **351**: custom values built in two days - **10**: workflows built by Claude from spec - **<1 min**: first lead response in testing ##### The problem Each CRM template (pipelines, workflows, fields, emails, knowledge base, pages) took contract builders weeks at about $500 per template, and quality varied. ##### Results A two-day build produced 351 custom values, 20 tags, 21 fields, a 12-stage pipeline, 2 calendars, a 52-entry knowledge base, 11 workflow shells, 28 SMS snippets and 4 pages. The agency put contractor hiring on hold. ##### What I learned The platform's own AI created data objects but not workflows, so the agents worked in the workflow builder instead. The specification and the gap report are the real product. ##### What I built - A specification format that defines every CRM object - A build method where Claude or Codex constructs those objects in the live platform - API readback and a gap report after each phase - Hard boundaries: test locations only, repeated location checks, no live SMS until a compliance-safe number exists - A QA skill that enrolls test contacts across workflows (23 enrollments across 16 workflows) Stack: Claude Code, OpenAI Codex, GoHighLevel API v2, Python Status: Tested. Voice AI · Multi-location retailer · 2026 #### Inbound voice AI agent family Phone agents that answer, qualify, book and hand off, tested against 50+ cases before any client demo. An inbound call reaches the voice agent, which books, transfers, takes a message or sends a follow-up text. - **11 / 11**: first test block passed - **50+**: test cases plus a stop-ship regression - **6**: service branches with matched follow-up texts - **5**: phone agents and a website voice pilot ##### The problem A retailer with several showrooms was missing calls and wanted every caller qualified and routed, without an AI that quotes prices or gives risky advice. ##### Results Showroom booking verified on a live call. Defects found in testing, including an emergency response that could be interrupted, were fixed before the client heard the agent. Pricing proposed as a launch fee, a prepaid build and monthly maintenance, with usage at cost. ##### What I built - An agent that books appointments, takes messages and transfers with a spoken summary - A prequalifying flow: full questions for shoppers, two at most for service calls, immediate transfer on request - Consent-gated follow-up texts sent two minutes after the call, matched to the caller's need - Safety rules: no prices, chemical advice or promised dates; the 24-hour number on safety issues Stack: GoHighLevel Voice AI, Conversation AI, CRM workflows Status: Live. Reporting · Marketing agency · 2026 #### Account health dashboard & alert bridge Every client CRM account checked nightly; problems become tasks in the tool account managers already use. Every client CRM account is read nightly by a read-only engine with 18 flags, producing CRM tasks, a dashboard and a Monday digest. - **18**: health flags per account - **Nightly**: read-only check of every account - **9,600**: open deals analyzed across 27 accounts - **~1 week**: from start to live ##### The problem Account managers checked client accounts by hand. Silent forms, waiting leads and stalled deals surfaced only when a client complained. ##### Results The first portfolio read found 81% of 9,600 open deals idle for over 30 days. The fix was three habits (closing deals out, marking losses, clearing dead deals), not a pipeline redesign. ##### What I built - A Python engine that reads every account through read-only tokens into Supabase - Flags compared against each account's own baseline and against peers - A severity-sorted view by account manager with a one-click weekly client update - An alert bridge that creates CRM tasks and emails for silent forms, dropped lead sources and stalled pipelines - A Monday digest: every open issue per client in one email Stack: Python, Supabase, GoHighLevel API, HTML email Status: Live. Reporting · Ad-ops automation · 2026 #### Google Ads automation suite Portfolio-wide Google Ads reporting and optimization running on a schedule at the manager-account level. Fifteen scripts run across every account in the manager account and produce audits, pacing, search-term reports and the weekly deck. - **15**: production scripts - **~12,666**: lines of code - **12–15 hrs**: of manual reporting saved each week - **3 weeks**: to build ##### The problem Every account needed audits, pacing checks and search-term reviews each week. Done by hand across a whole portfolio, that work crowded out optimization. ##### Results Daily, weekly, monthly and quarterly checks now run on their own, and a companion script builds the 25–29 slide weekly client meeting deck from live data. ##### What I built - Account audits and health checks - Budget pacing with automatic adjustment - Search-term analysis and wasted-spend ranking - Structure optimization and conversion-tracking checks Stack: Google Ads Scripts, Google Ads API & GAQL, Google Sheets Status: Pilot. CRM automation · AI agents · 2026 #### Research-then-verify client onboarding AI researches a new client before kickoff, and the client corrects a pre-filled profile instead of filling out a long questionnaire. A new client triggers a webhook, a Claude agent researches the business, the report returns to the CRM, the client verifies it, and the account builds itself. - **26**: sections in each AI research profile - **<20 min**: unattended research run - **74**: fields covered by an automated save test - **7**: tabs in the client onboarding app ##### The problem Onboarding asked clients for hundreds of fields across long forms. Clients stalled, data arrived incomplete and launches slipped. ##### Results In pilot with the agency team. Three client pilots must pass a written checklist before any new features are added. ##### What I built - A CRM workflow that fires a webhook; a Claude agent researches the business and posts a report back into the CRM - Research that works without a website, using license records, permits and directories - Stop-and-notify when a listed site is broken or belongs to another company - A confidence label on every field (verified, derived or needs input), and a client review and e-sign flow - Automatic sub-account build when onboarding completes Stack: Claude Routines, Firecrawl, GoHighLevel API & webhooks, Supabase Edge Functions Status: Live. AI agents · Enablement · 2026 #### 104-skill AI library & orchestration pipeline An agency's AI work packaged as versioned, searchable skills, plus a one-prompt pipeline that runs nine of them in a gated sequence. One source document is built, validated and frozen, then parallel skills read it to produce 17 deliverables. - **104**: packaged skills across 12 categories - **9**: skills in one gated pipeline - **17**: deliverables per pipeline run - **~$4–8**: measured cost per run ##### The problem Good prompts lived in individual chats. When one early run scaled 43 outputs from an unchecked source, 22 of them had to be regenerated. ##### Results The library is published as a searchable site with team and priority filters, so each team starts from an adopt-first list. ##### What I built - A versioned skill plugin with a manifest for team rollout - A pipeline that builds a source document, validates and freezes it, then runs parallel steps from it - Gates before scale: a validator, a graded voice sample and a capacity check - A quarterly library audit for duplicates and retirement Stack: Claude skills, Claude Code, Firecrawl, Netlify Status: Shipped. Web & search · Healthcare IT consultancy · 2026 #### 30-day website launch for a healthcare IT consultancy Strategy to launch in 30 days as a solo engagement, with accessibility and AI-search groundwork built in. Design and build, content and QA, and a one-evening launch led to 1,000 organic clicks per 28 days within four weeks. - **30 days**: concept to live - **20+**: pages designed and built - **WCAG AA**: audited before launch - **1,000+**: organic clicks per 28 days within 4 weeks of launch ##### The problem The firm needed a site that matched its reputation, on a fixed budget of about 60 hours, without losing the search traffic of the old site. ##### Results Live in one evening via DNS cutover. Organic clicks rose 33% from the launch month to the following month. ##### What I built - A hand-coded WordPress theme with no page builders and sub-second page loads - A redirect map, launch runbook and parallel build with instant rollback - Forms with file uploads, consent gating and invisible anti-spam - llms.txt, FAQ patterns and citable claims for AI search - An AI automation roadmap for the firm's leadership Stack: WordPress (PHP), HTML/CSS/JS, Cloudflare Turnstile, Search Console ### More work Supporting builds, products and analyses. Status: Tested. CRM automation #### Speed-to-lead engine Ten workflows that catch every lead, check business hours and compliance, text first with email fallback and watch response times. First clean test: one text and one email in under a minute Status: Live. Voice & chat AI #### Three-bot chat system with shared knowledge bases Chat assistants for three brands drawing on six modular knowledge bases with clean scope separation. 163 FAQs · all bots validated above 85% accuracy Status: Live. Web & search #### AI visibility audit products Four audits, including an LLM visibility audit that queries ChatGPT, Gemini and Claude with web grounding and scores visibility 0–100. Branded HTML reports from one shared standard Status: Tested. Reporting & analytics #### Automated monthly client report A Looker Studio dashboard goes in; an internal report, a client-facing report and a pre-filled meeting doc come out. Google Ads, Meta, Analytics and Business Profile in one run Status: Tested. CRM automation #### Website asset loader Pulls a client's logo, colors, fonts and photos into an organized folder, then bulk-loads them into the CRM media library. About 350 images loaded in 8 minutes Status: Live. AI agents #### Fact-locked AI content Diagnosed why in-tool AI copy drifted from verified client facts, then built a context block that every research run now produces. Unverified items excluded by rule Status: Delivered. AI agents #### AI automation roadmap A phased automation roadmap and a CRM and HR-system integration readout for a healthcare IT consultancy's leadership and finance teams. Paired with the 30-day website launch Status: Shipped. Products #### SkillEngine An AI productivity platform that saves your best prompts as reusable skills, with Claude, Gemini and OpenAI behind one interface. 277 skills · 48 pages · built in 31 days · [Open the app](https://aicareerskills.netlify.app/) Status: Shipped. Products #### Media mix model platform Bayesian media mix modeling with data connectors, data-quality checks and budget scenario planning. Adstock and saturation curves · scenario comparison ### Lab Smaller builds that show range. - **Client-side security scanner**: 15 security checks, 12 PII types detected, scored A–F - **AI governance platform**: 55 pages built in under a week - **B2B CRM with prospect portals**: Multi-contact pipeline, AI research assistant, ROI calculator --- # FAQ: AI Agents, Voice AI & Automation | Matthew Carlson Source: https://matthewcarlsonconsulting.com/faq.html ## Questions about working with me How I build, test, price and secure AI systems. Don't see your question? [Ask me directly](https://matthewcarlsonconsulting.com/contact.html). ### Working together #### What does an AI Solutions Engineer do? An AI Solutions Engineer designs, builds and runs AI systems that do real work inside a business: agents, voice and chat assistants, automations and reporting pipelines. The job covers the whole path from a business problem to a tested system in production, not only the model or the prompt. In my practice that means writing the specification, building in a test environment, verifying the result, and launching with monitoring and a known cost per run. #### What is a Forward Deployed Engineer? A Forward Deployed Engineer works inside the client's business rather than from the outside. Instead of handing over recommendations or a product and leaving, I embed with your team, work in your actual systems and data, build the solution in your environment, and iterate with the people who will run it. Deliverables are working systems, not slide decks. It's the way I work with my agency clients today: building inside their CRM, reviewing results with the account managers each day, and handing off documentation and tests so the team owns what was built. #### Who do you work with? Mostly marketing agencies, in-house marketing and revenue-operations teams, and professional-services firms. Agencies are a natural fit because I've built AI systems that run across an entire client portfolio. #### How long does a build take? An AI Readiness Sprint takes two weeks. Most single-system builds take two to six weeks from approved spec to handoff. Some move faster: a nightly account-health dashboard went live in about a week, and a 20+ page website launched in 30 days. #### How is pricing structured? Sprints and builds are fixed-scope and fixed-price after discovery. Ongoing work is a monthly retainer. For systems with running costs, such as voice agents, launch, build and maintenance are priced separately and AI usage is billed at cost. I share a quote after our first call. #### Do you work with businesses outside Madison? Yes. I'm based in Madison, Wisconsin and work with clients across the U.S. by video. Madison-area clients can meet in person for kickoffs. #### What's the first step? [Book a 30-minute call](https://matthewcarlsonconsulting.com/contact.html), or [score your process on the readiness checklist](https://matthewcarlsonconsulting.com/ai-readiness-checklist.html) first. Bring the process you'd like AI to handle. We'll talk through the data, the risks and what success is worth, and you'll leave knowing whether it's worth building. #### What size of company do you work with? Mostly small and mid-sized businesses: agencies with a few dozen client accounts, in-house marketing teams of five to fifty, and professional-services firms. The systems I build are sized to that scale and priced for it. If you're larger, the same approach works inside one team or business unit. #### How much of my team's time will this take? Less than you'd expect. A typical build needs a 60-minute discovery session, a short review of the written spec, access to the systems involved, and 20 to 30 minutes a week for check-ins. The people who will run the system join for the pilot and a handoff session at the end. #### Can you work alongside our existing developers or agency? Yes, and I prefer it. I work in your repositories, your CRM and your tools, document everything in writing, and leave your team able to maintain and extend what I built. If you already have an agency, I can take the AI and automation layer while they keep the rest. #### Who owns the system when the engagement ends? You do. Code, prompts, workflows, data and documentation are yours, in your accounts, on your infrastructure. There's no platform fee to keep using what I built. If you want me to keep running and improving it, that's the Managed AI Operations retainer, but it's optional. ### AI systems & safety #### How do you make sure an AI agent won't do something wrong? Several layers: - Agents that change things are built and tested in separate test environments first, with checks that confirm where they're working. - Reporting systems are read-only by default. - Client-facing agents pass a written test suite (50+ cases for a voice agent) and a safety regression before anyone outside the team hears them. - When an input looks wrong, the system stops and notifies a person instead of guessing. #### What does it cost to run an AI system after launch? It depends on volume, and I measure it rather than estimate it. Every pipeline I build tracks its own model and data costs per run. For example, a full AI research pull on a new client costs about $4–8. A 33-agent platform serving an entire agency runs for about $32–42 a month. #### How do you handle data privacy and security? - API keys stay on the server, never in the browser. - Agents get the minimum access they need, often read-only. - Sensitive items such as tax IDs and passwords never go through forms or AI prompts. - AI-researched facts are labeled verified, derived or needs input, and unverified details are never presented as fact. #### Do you build voice AI agents? Yes. I design and test inbound phone agents that answer questions, book appointments, take messages and transfer to a person with a spoken summary. They capture consent before follow-up texts and have clear rules about what they won't say, such as prices or technical advice. #### Can AI agents build inside a CRM like GoHighLevel? Yes, with the right method. I write a specification for every object, and Claude or OpenAI Codex builds it directly in the platform. After each phase an API readback and gap report show what was built and what wasn't. One unattended run lasted 3.5 hours, and the two-day build produced 351 custom values, a 12-stage pipeline and a 52-entry knowledge base. #### Should we build or buy? Buy the commodity and own the interpretation. If a tool already measures something well, use it. Build where your data, process or judgment is the advantage. An AI Readiness Sprint ends with a clear build-or-buy call for each opportunity. #### How do you measure whether an AI system is working? Each system has a written success measure before it's built: hours saved per week, response time to a lead, leads caught that were missed before, reports produced without a person. During the pilot those numbers are tracked and reviewed with you, and the system logs its own activity and cost so the numbers hold up after handoff. #### Will AI replace my team? Not in my experience. The systems I build take the repetitive, time-bound work (first replies to leads, nightly checks, report assembly, research before a call) so your people spend their time on judgment and relationships. Every system I ship keeps a person in control of anything that goes to a customer. ### Platforms & integrations #### What platforms do you work with? - **AI:** Claude (API, Claude Code, MCP), OpenAI (API, Codex), Gemini - **CRM & marketing automation:** GoHighLevel, HubSpot - **Data:** Supabase/Postgres, Google Sheets, Looker Studio, GA4 - **Ads:** Google Ads API and Scripts, Meta - **Web:** WordPress, static sites, Netlify If a tool has an API or webhooks, it can usually be connected. #### Can you automate our Google Ads reporting? Yes. I built a 15-script Google Ads suite that runs at the manager-account level across a whole agency portfolio: audits, budget pacing, search-term and wasted-spend analysis, and structure checks. It replaced about 12–15 hours of manual reporting a week. #### What is the SkillEngine Assistant? It's the AI assistant on this site. You can [ask it about my work](https://matthewcarlsonconsulting.com/ai-bot.html) or describe a process and get a first-pass automation plan. I build similar assistants for clients, connected to their own knowledge and tools. #### What is MCP, and why does it matter? The Model Context Protocol (MCP) is an open standard that lets AI assistants such as Claude securely use your data and tools. I've built an MCP server in production use that gives Claude access to an agency's client profiles, with token authentication and audit logging. --- # AI Readiness Checklist: 20 Questions, Scored | Matthew Carlson Source: https://matthewcarlsonconsulting.com/ai-readiness-checklist.html ## Is this process ready for AI? Twenty yes-or-no questions across the five areas I check before any build. Each yes is five points. Your score, and what to do with it, appears as you go. Nothing is stored or sent anywhere. Your readiness score 0/100 Tick what's true today. A rough answer is fine; this is a starting point, not an exam. ### Three ranges, three next steps **0–45** #### Start with the process, not the AI Something upstream isn't ready: the work isn't written down, the data isn't reachable, or nobody will own the result. An AI Readiness Sprint fixes that order of operations in two weeks and ends with a roadmap. **50–70** #### Pick one process and pilot it You're close. Choose the single process with the clearest hours-saved number, run it on a small slice in a test environment, and measure before scaling. **75–100** #### Build it The inputs, outputs and owners exist. A build sprint of two to six weeks takes it from spec to a tested system your team runs. --- # Insights: Building Production AI Systems | Matthew Carlson Source: https://matthewcarlsonconsulting.com/insights/ ## One question, answered properly Short articles from the work: how the systems are built, what they cost to run, and the habits that keep them safe. New ones added as the work produces them. [Voice AI6 min read ### How do you keep an AI voice agent from quoting prices? You write down what it must never say, test every rule before a customer hears it, and give the agent a graceful way out: transfer to a person.](https://matthewcarlsonconsulting.com/insights/voice-agent-guardrails.html) [AI agents5 min read ### What does an AI client research pull cost, and why meter it? A full research pull on a new client costs about $4–8 in model and crawl spend, and I know that because every step is metered. That number is what makes AI priceable.](https://matthewcarlsonconsulting.com/insights/cost-per-research-pull.html) [AI agents6 min read ### Why gate before you fan out Scaling dozens of AI outputs from an unchecked source multiplies every upstream mistake. One graded sample and a consistency check cost hours; regeneration costs days.](https://matthewcarlsonconsulting.com/insights/gate-before-fan-out.html) --- # How do you keep an AI voice agent from quoting prices? | Matthew Carlson Source: https://matthewcarlsonconsulting.com/insights/voice-agent-guardrails.html **50+**: test cases before any client demo **11 / 11**: checks passed in the first test block **6**: service branches, each with its own follow-up **2 min**: after the call, one consented text An inbound call reaches the voice agent, which books, transfers, takes a message or sends a follow-up text. **Short answer:** the agent gets a written list of things it will not do (quote prices, give technical advice, promise dates, confirm stock), a safe behavior for each one (transfer, take a message, give the emergency number), and a test suite that proves the rules hold before anyone outside the team hears it. ### Why price quoting is the first rule A phone agent for a multi-location retailer gets asked prices constantly. A confident, wrong number from an AI is a customer-service problem the day it happens and a trust problem for every call afterward. The same goes for chemical or technical advice, delivery promises and "is it in stock?" Each one sounds helpful and each one creates a commitment the business didn't make. So the agent I built for a multi-location retailer was designed inbound-only and given hard limits. It will retrieve business information, book showroom appointments to the calendar, update contact details, save the conversation to the contact record, take messages, transfer to a person with a spoken summary, roll to voicemail when nobody answers, and give the 24-hour number on anything that sounds like a safety issue. It declines to quote prices, give chemical advice, promise dates or confirm stock. The decline is polite and short, and it always ends with an action: a transfer, a message, or a text with the right link. ### Design the agent around the caller's time The first version answered questions well in testing, but the discovery call changed the design. The client wanted routing to stay with their own people. So the agent became a prequalifying survey that hands off to the main line: it captures a name, a callback number and permission to text, works through one of six service branches, and transfers. Three rules came out of that redesign and I now use them on every voice agent: - Shoppers get the full set of questions. Service callers get at most two. - "Put me through" is honored on the first ask. No agent should ever argue with a caller who wants a person. - An upset caller skips everything and goes straight to transfer. After the call, a workflow sends one follow-up text matched to the service branch (a buying guide, the online service scheduler, a brochure), two minutes after the call ends, only if the caller agreed to texts. One text per call, no exceptions. A one-line promotions switch lets the client turn seasonal offers on and off without touching the agent. ### Test before anyone hears it None of this reached the client until it passed a written test plan. The first block was eleven checks, and all eleven passed, including booking a showroom visit at the correct time on a live call. Two defects were found and fixed before a demo, one of them an emergency response the caller could interrupt. A later round fixed three more, including a handoff that happened without capturing the caller's details. The full suite is now more than fifty cases, with a regression set of twenty-plus checks that runs on every change and includes every safety behavior. The rule I keep: no client demo until the suite passes. A demo that goes wrong costs more trust than a week of delay. ### Price it so the rules stay enforced Voice agents are not a one-time build. Rules change, branches get added, the regression suite has to keep running. So the pricing separates the parts: a launch fee, a prepaid build allowance for changes, a monthly fee for operation and maintenance, and calls and texts billed at cost. The monthly fee is what keeps the test suite running after launch, which is what keeps the agent from quoting prices six months from now. ### What to ask any voice AI vendor - Show me the written list of things the agent will not do, and what it does instead. - Show me the test cases, and the regression set that runs on every change. - What happens when the caller says "put me through"? - Who owns the knowledge base, the prompts and the call logs after launch? If the answers are specific, the agent is probably safe to put in front of your customers. If they're vague, it isn't yet. ### Quick answers #### Can an AI voice agent give prices? It can, but it shouldn't. A confident wrong number is a customer-service problem and a trust problem. The agent I build declines price questions politely and ends with an action: a transfer, a message, or a text with the right link. #### How do you test a voice agent before launch? With a written test plan. The first block was eleven checks including a live booking; the suite is now more than fifty cases plus a regression set of twenty-plus checks, run on every change, and no client demo happens until it passes. #### What should a voice agent do when the caller asks for a person? Transfer on the first ask, with a spoken summary of the call so far. An agent should never argue with a caller who wants a human. **Have a process AI should be running?** Describe it and I'll tell you whether it's worth building. [Book a call](https://matthewcarlsonconsulting.com/contact.html) or [ask the assistant](https://matthewcarlsonconsulting.com/ai-bot.html). ### More insights - [What does an AI client research pull cost, and why meter it?](https://matthewcarlsonconsulting.com/insights/cost-per-research-pull.html) - [Why gate before you fan out](https://matthewcarlsonconsulting.com/insights/gate-before-fan-out.html) --- # What does an AI client research pull cost, and why meter it? | Matthew Carlson Source: https://matthewcarlsonconsulting.com/insights/cost-per-research-pull.html **$4–8**: measured cost per client research pull **0.8–1.4M**: tokens per pull **17**: deliverables from one prompt **$32–42**: per month to run a 33-agent platform A client research pull crawls and sweeps, builds and freezes one profile, fans out to parallel readers, and reports its cost at the end. **Short answer:** about $4–8 and 0.8–1.4 million tokens per client, measured, not estimated. Metering every run is what turns "AI is expensive" into a unit cost you can put in a proposal. ### What a research pull does When a new client comes into the agency I work with, one prompt runs nine skills in sequence and in parallel: a crawl of the client's site, an external sweep of review platforms, directories and public records, a competitor scan, and then the packaging steps that produce seventeen deliverables, from a structured client profile to a CRM import file, a brand-voice guide and a 50-question FAQ. The whole pipeline runs as a directed graph: a serial build, a validation gate, a checksummed freeze of the source document, four readers running in parallel, and a single join at the end. That's a lot of model calls and a lot of crawling. Without measurement it would be impossible to say what one client costs, which means impossible to price. ### How the metering works - **Every agent run records its tokens and its model.** The platform routes work across three Claude tiers depending on the step, so the cheap model handles extraction and the expensive one handles judgment. The telemetry shows which step used which model and what it cost. - **Crawl credits are metered per step.** The crawler is the other variable cost; each profile declares its budget. The lightest profile, a seed pull that gates the client's onboarding invite, allows at most six external calls and targets five minutes. The full build allows the whole sweep. - **The cost is reported at the end of every run**, next to the artifacts. Nobody has to go looking for it. Measured across real client pulls, the full pipeline lands at roughly $4–8 and 0.8–1.4 million tokens per client. The lighter webhook-triggered research report, which runs when a lead enters the CRM, is budgeted at about eight page scrapes and four searches and finishes in under twenty minutes. ### Why the number matters Three reasons, in the order I discovered them: - **Pricing.** A known unit cost can be marked up, bundled or passed through. An unknown one gets padded, and padding loses deals. - **Design.** Once you can see cost per step, you find the expensive steps and ask whether they need the expensive model. Routing extraction to the small model and judgment to the large one came directly from reading the telemetry. - **Trust.** When leadership asks "what is this costing us?", the answer is a number with a date on it. The whole 33-agent platform, serving the entire agency, operates for about $32–42 a month. Commercial alternatives in the same category run $49–249 per user per month. ### Compute is a budget line The same discipline applies when AI coding agents do the building. A multi-hour unattended build inside a CRM, at high compute settings, is real money. I run those in a fresh session per phase and use a small skill that warns when the context window grows larger than the task needs. Treating compute as a line item, rather than a surprise on the invoice, is part of the engineering. ### What to ask before you buy an AI pipeline - What does one run cost, measured on real inputs? - Which steps use which model, and why? - Is crawl or API usage capped per run? - Where does the cost show up after each run? If a vendor can't answer the first question with a number, the pipeline isn't finished. ### Quick answers #### How much does an AI research pipeline cost per run? Measured on real clients, about $4–8 and 0.8–1.4 million tokens per full pull, with crawl credits metered per step. The lighter webhook-triggered report is budgeted at about eight page scrapes and four searches. #### Why route AI work across different model tiers? Because cost telemetry shows which steps need judgment and which only need extraction. Sending extraction to the small model and judgment to the large one came directly from reading per-step costs. #### What does a multi-agent AI platform cost to operate? The 33-agent platform serving an entire agency runs for about $32–42 a month in model and infrastructure spend, against $49–249 per user per month for commercial alternatives in the category. **Have a process AI should be running?** Describe it and I'll tell you whether it's worth building. [Book a call](https://matthewcarlsonconsulting.com/contact.html) or [ask the assistant](https://matthewcarlsonconsulting.com/ai-bot.html). ### More insights - [How do you keep an AI voice agent from quoting prices?](https://matthewcarlsonconsulting.com/insights/voice-agent-guardrails.html) - [Why gate before you fan out](https://matthewcarlsonconsulting.com/insights/gate-before-fan-out.html) --- # Why gate before you fan out | Matthew Carlson Source: https://matthewcarlsonconsulting.com/insights/gate-before-fan-out.html **22 of 43**: outputs regenerated after one ungated run **6**: gates, each ending in a decision **42 × 31**: client profiles × data dimensions audited **2**: auto-remediation passes, then stop A source document passes a validator and a graded voice sample, is frozen with a checksum, and only then fans out to many agents. **Short answer:** when many AI tasks read from one source document, validate the source and grade one sample output before you run the rest. On one engagement, skipping that step meant regenerating 22 of 43 outputs. ### The failure that taught me this The agency I work with runs a 43-skill agent suite: campaign briefs, landing pages, email sequences, FAQ sets, local-SEO audits and more, all reading from a single client profile. Early on, one engagement ran all 43 against a profile nobody had validated. The profile had inconsistencies and the brand voice was slightly off. Every one of the 43 outputs inherited those problems, and 22 had to be regenerated after the client reviewed them. The lesson wasn't "be more careful." It was that the pipeline had no gate between the source and the fan-out, and a pipeline without a gate will eventually do this again. ### Six gates, each a counter-measure to a named failure I rebuilt the process as a five-pass system with a gate layer in front of it. Every gate ends in a decision, not a report: - **Source document validator.** Checks the client profile for internal consistency and returns GO or NO-GO. A NO-GO stops everything. - **Brand voice sample validator.** Runs one representative skill, grades the output A to F against the client's voice anchors (attributes, words to use, words to avoid, signature claims), and only passes on a grade the team would ship. If the voice fails, the account manager fixes the profile instead of scaling 43 wrong-voice outputs. - **Capacity model.** Compares each team member's available hours with what the engagement will ask of them. This exists because an earlier synthesis pass discovered 20–25 hours of work landing on a team that had eight. - **Sprint-zero intake.** A hard gate on twenty critical data points that must be present before any content is produced. - **Engagement goal generator.** Ten ranked candidate goals, scored on market position, competitive angle and trend fit, so the fan-out has a target. - **Skill router.** Routes each of the 43 skills to Critical, High, Medium or Skip for this client, with structural disqualifiers. Not every client needs every output. ### Freeze shared state before parallel reads The second rule follows from the first. Validators change the source document, so the parallel readers must not start until it is frozen. The onboarding pipeline builds the profile serially, runs the validator as a barrier, checksums the frozen document, and only then fans out four parallel readers. Late corrections don't edit the frozen document mid-run; they queue as amendments and are applied once at the join. If one reader fails, it flags the failure and the others finish; one bad branch doesn't kill the run. Auto-remediation is bounded to two passes, so a broken input can't loop forever. ### Keep the library honest Gates protect each run. Something also has to protect the library over time. A quarterly auditor fingerprints every skill, finds near-duplicates, applies retirement criteria to dead skills, and proposes composites, against a target of 200 active skills. And the schema itself gets audited from evidence: a scan of 42 client profiles across 31 data dimensions found social links missing for more than 97% of older profiles and business-listing data captured for one of 42. Twenty recommendations from that scan shipped in the next release. ### Post-mortems should ship as architecture Every gate above is a named counter-measure to a named failure. None of them live in a lessons-learned document; they live in the tooling, where they run whether or not anyone remembers the lesson. That's the real principle. If an AI system failed once in a way that could repeat, the fix is a gate in the pipeline, not a note in a wiki. ### What to ask about any multi-agent system - What validates the source before the agents read it? - Is one output graded before the rest are generated? - What happens to the shared state while agents run in parallel? - What happens when one agent fails? ### Quick answers #### What does it mean to gate before fanning out? Before many AI agents read one shared source, validate the source and grade a single sample output. If either fails, fix the source instead of generating dozens of flawed outputs. #### Why freeze a document before parallel agents read it? Validators change the document. If readers start before it's frozen, they read different versions. A checksummed freeze, with late corrections queued as amendments at the join, keeps every output consistent. #### What happens when one agent in a parallel run fails? It flags the failure and the others finish. Remediation is bounded to two passes so a broken input can't loop forever, and nothing is silently truncated. **Have a process AI should be running?** Describe it and I'll tell you whether it's worth building. [Book a call](https://matthewcarlsonconsulting.com/contact.html) or [ask the assistant](https://matthewcarlsonconsulting.com/ai-bot.html). ### More insights - [How do you keep an AI voice agent from quoting prices?](https://matthewcarlsonconsulting.com/insights/voice-agent-guardrails.html) - [What does an AI client research pull cost, and why meter it?](https://matthewcarlsonconsulting.com/insights/cost-per-research-pull.html)