# Sigma Logic AI

> Sigma Logic AI designs, builds and maintains production AI systems for growing businesses - support agents, email automation, lead engines and custom models, integrated with the tools you already run.

Sigma Logic AI is an applied AI consultancy based in Dover, Delaware, USA, working
with clients worldwide. Engagements run through four phases, each ending in a
decision point where the client can stop: Diagnose (days 1-2, opportunity map + costed roadmap); Prove (week 1, working prototype + evaluation report); Integrate (weeks 2-3, production deployment + runbook); Operate (ongoing, monitoring, evals, quarterly review).
A single well-scoped workflow typically reaches production in under four weeks.
The first phase is fixed fee and the roadmap is the client's to keep either way.

Contact: contact@sigmalogicai.com | +1 (973) 852-3858

## Start small

- [One workflow in a week](https://www.sigmalogicai.com/services/n8n-make-automation#starter): Fixed scope, fixed price, five working days. Fixed price USD 1500. Pick the one repetitive job that costs you the most time. We scope it on a thirty-minute call, build it, and it is running in your systems within five working days of kickoff. If it is not running by day five, there is no invoice.

## Services

- [AI Customer Support Agents](https://www.sigmalogicai.com/services/ai-support-agents): Most support bots are a filter your customers learn to defeat. We build agents that read your actual documentation, order history and account state, then either finish the job or hand a human the full context - no starting over, no loops. Covers Tier-1 resolution, Inbound voice agents, Outbound voice agents, Agent assist (copilot), Ticket triage and routing, Multilingual support, Conversation summarisation, Automated QA scoring, Escalation design, Knowledge gap analysis.
- [AI Email Automation](https://www.sigmalogicai.com/services/email-automation): Batch sends and five static templates leave most of the value on the table. We build flows that decide what to send, to whom, at what moment - generated per recipient from their actual behaviour, and improved automatically by what the last send taught the system. Covers Lifecycle mapping, Onboarding and activation flows, Abandonment recovery, Winback and reactivation, Renewal and retention, Per-recipient generation, Dynamic segmentation, Send-time and frequency control, Continuous testing, Deliverability engineering, Incrementality measurement.
- [Lead Generation & Management](https://www.sigmalogicai.com/services/lead-generation): Most pipeline is not lost to competitors. It is lost to a slow first response, a lead scored by a rule written two years ago, and a nurture track nobody has read since. We close those gaps with systems that qualify, route and follow up in minutes. Covers Predictive lead scoring, Enrichment and deduplication, Automated qualification, Routing and SLA enforcement, Adaptive nurture, Meeting-prep briefs, Intent signal monitoring, Pipeline hygiene, First-touch drafting.
- [Business Process Automation](https://www.sigmalogicai.com/services/process-automation): The expensive work is rarely inside a system. It is the person moving information between them - rekeying invoices, chasing approvals, reconciling two exports that never quite match. That work is now automatable, including the messy, unstructured parts that defeated older rule engines. Covers Process discovery, Cross-system orchestration, Approval routing, Reconciliation and matching, Exception handling, Data entry elimination, Legacy RPA modernisation, Scheduling and dispatch, Audit trail construction.
- [AI for E-Commerce](https://www.sigmalogicai.com/services/ai-for-ecommerce): Online retail runs on decisions made thousands of times a day: what to show this visitor, what to reorder, what to price, what to flag. Each is small; together they set your margin. We automate them against your live data instead of last quarter’s assumptions. Covers Personalised recommendations, Semantic on-site search, Demand forecasting, Inventory and replenishment, Dynamic pricing, Markdown optimisation, Catalogue enrichment, Visual search and similar items, Size and fit guidance, Fraud and payment risk, Return abuse detection, Churn and lifetime value, Basket and bundle analysis, Review and sentiment mining, Shopping feed optimisation.
- [Custom AI Development](https://www.sigmalogicai.com/services/custom-ai-development): Some problems are specific to you: your terminology, your rules, your data, your edge cases. We build those systems from scratch - properly evaluated, integrated with what you run, and handed over with the source and the documentation so you are never locked in. Covers Retrieval-augmented generation (RAG), Fine-tuning, Open-weight deployment, Multi-step agents, Classical machine learning, Structured extraction, Evaluation harness, Model routing and fallback, Integration middleware.
- [AI Consulting & Strategy](https://www.sigmalogicai.com/services/ai-consulting-strategy): The costly mistake is not picking the wrong model. It is spending two quarters automating something that was never worth automating. We start with the operational reality - where time and money actually go - and produce a sequenced plan with the numbers attached. Covers Readiness assessment, Opportunity mapping, ROI and payback modelling, Build-versus-buy analysis, Vendor evaluation, Data strategy review, Governance and policy design, Sequenced roadmap, Executive briefing.
- [Maintenance & Support](https://www.sigmalogicai.com/services/maintenance-support): An AI system does not fail like a server. It keeps returning confident answers that are slowly getting worse - because a vendor changed a model, your product changed, or the questions people ask moved on. Without measurement, you find out from customers. Covers Monitoring and alerting, Scheduled evaluation runs, Drift detection, Prompt and config versioning, Incident response, Provider migration, Cost tracking, Quarterly review.
- [AI Search Visibility (GEO)](https://www.sigmalogicai.com/services/ai-search-visibility): A growing share of buying research never reaches a results page. Someone asks an assistant, gets an answer with three cited sources, and everything else is invisible. Being one of those three is a different discipline from ranking - and unlike most of what gets called AI marketing, it is measurable. Covers Citation baseline audit, Prompt set design, Answer-first content engineering, Structured data implementation, Machine-readable surfaces, Entity consistency, Source authority building, Comparison and alternatives pages, Citation share tracking.
- [AI Training & Enablement](https://www.sigmalogicai.com/services/ai-training-enablement): Most AI training is a vendor demo with a certificate attached. People leave impressed and change nothing on Monday. We run sessions built on your actual tools, your actual documents and your actual workflows, so the output is a changed way of working rather than a completion rate. Covers Leadership briefing, Role-specific workshops, Prompt fundamentals, Tool-specific training, Acceptable-use policy, Champions programme, Change management, Reinforcement session.
- [Fractional AI Leadership](https://www.sigmalogicai.com/services/fractional-ai-lead): Plenty of companies need someone senior owning AI decisions and nowhere near enough work to justify a full-time hire at that level. We take the role on retainer: the roadmap, the vendor calls, the governance, and the awkward job of telling people when something is not worth building. Covers Roadmap ownership, Vendor evaluation and negotiation, Delivery oversight, Governance and policy, Budget ownership, Team structure and hiring, Board reporting, Succession.
- [AI System Rescue](https://www.sigmalogicai.com/services/ai-system-rescue): The person who built it has left. The agency relationship ended. It still runs, mostly, and nobody can tell you whether it still works - only that turning it off feels risky. This is the least glamorous work we do and the most common reason people call us. Covers System archaeology, Quality baseline, Cost and dependency audit, Security and permission review, Documentation reconstruction, Stabilisation, Remediation, Provider migration, Decommissioning.
- [Enterprise Knowledge Assistant](https://www.sigmalogicai.com/services/knowledge-assistant): Your company already knows the answer. It is in a wiki nobody maintains, a PDF on a shared drive, a Slack thread from last March, and one person who is on holiday. We build the layer that finds it, answers in plain language, and shows exactly where it came from. Covers Source connection, Permission-aware retrieval, Cited answers, Conversational interface, Onboarding assistant, Policy and HR self-service, Content gap reporting, Decision and meeting capture.
- [Document Processing (IDP)](https://www.sigmalogicai.com/services/document-processing): Invoices, purchase orders, claims forms, onboarding packs, contracts. Somebody opens each one, reads it, and retypes the numbers into a system that will never know where they came from. This is the most reliably automatable work in most businesses, and the category older automation could never touch. Covers Invoice and AP automation, Purchase order matching, Contract abstraction, Claims processing, Application and form intake, Identity and KYC documents, Poor-quality input handling, Classification and routing, Validation against systems, Human review queue.
- [AI Cost & Performance Audit](https://www.sigmalogicai.com/services/ai-cost-audit): Inference costs tend to grow faster than usage, and the bill rarely says why. Most of the systems we look at are running a frontier model on work a much cheaper one handles identically, sending the same context hundreds of times a day, and paying for three tools that overlap. Covers Spend attribution, Model right-sizing, Context efficiency, Caching strategy, Request routing, Batch versus realtime, Licence and tool overlap, Infrastructure review, Budget alerting.
- [n8n & Make Automation](https://www.sigmalogicai.com/services/n8n-make-automation): Most n8n and Make work is template configuration that breaks the first time an API changes shape. We build workflows the way software gets built: version-controlled, tested, monitored, and with custom nodes written where the off-the-shelf ones do not fit. Covers Workflow design and build, Custom node development, Self-hosted n8n deployment, Zapier migration, Legacy automation rescue, Error handling and alerting, Credential and access management, AI steps inside workflows, Performance and cost tuning, Team enablement.
- [AI Evaluation & QA](https://www.sigmalogicai.com/services/ai-evaluation-qa): "It seems good" is the only quality signal most teams have. It is also the signal that quietly degrades when a provider updates a model, your product changes, or the questions people ask move on. Evaluation replaces the impression with a number you can act on. Covers Evaluation set construction, Rubric design, Automated scoring harness, Human review and annotation, Regression testing in CI, Robustness probing, Per-segment reporting, Confidence calibration, Provider comparison, Acceptance criteria.

## Free tools

Interactive, no sign-up, and nothing the visitor enters is transmitted or stored.
Each one computes only from figures the visitor supplies, shows its formula, and is
able to return the answer "do not build this".

- [Automation payback calculator](https://www.sigmalogicai.com/tools/automation-payback): Is this task worth automating? Is a task worth automating? Hours returned per week, annual value net of running cost, and payback in months - checked at a conservative rate as well as yours. Method: https://www.sigmalogicai.com/blog/finding-automatable-work
- [Evaluation set size calculator](https://www.sigmalogicai.com/tools/eval-set-size): How many test cases do we actually need? How many test cases you need before a measurement means anything. Margin of error at 95% confidence, sized from the per-category counts that actually matter. Method: https://www.sigmalogicai.com/blog/golden-dataset-size
- [Escalation threshold calculator](https://www.sigmalogicai.com/tools/escalation-threshold): Where should the confidence gate sit? What a confidence gate costs and saves. Compare two settings you have measured: deflection, wrong answers reaching customers, and total weekly cost. Method: https://www.sigmalogicai.com/blog/confidence-thresholds-and-escalation-design
- [llms.txt generator](https://www.sigmalogicai.com/tools/llms-txt-generator): What should our llms.txt contain? Produce an llms.txt in the exact format the proposal specifies: H1, one-line summary, and sections of links with a note each. Flags lines that would break it. Method: https://www.sigmalogicai.com/blog/llms-txt-and-structured-data

## Articles

- [Building an audit trail an AI system can defend](https://www.sigmalogicai.com/blog/ai-audit-trail): What to log at each stage of an AI system so a complaint, an access request or a regulator can be answered from the record, and the 2026 retention rules.
- [What belongs in an AI acceptable use policy](https://www.sigmalogicai.com/blog/ai-acceptable-use-policy): The eight sections an AI acceptable use policy needs, what the vendor training defaults actually are in 2026, and the rules that survive contact with a busy team.
- [Automated decisions and the right to human review](https://www.sigmalogicai.com/blog/automated-decisions-human-review): When GDPR Article 22 applies to an AI system, what meaningful human involvement means after two EU court rulings, and how the UK rewrote the rule in 2026.
- [AI training that changes what people do on Monday](https://www.sigmalogicai.com/blog/ai-training-that-changes-monday): Most AI training produces enthusiasm and no behaviour change. What separates the sessions that stick: your own documents, one workflow per role, a number.
- [Testing an AI system against the human baseline](https://www.sigmalogicai.com/blog/ai-vs-human-baseline): How to measure what your people currently achieve on the same cases, why that number is usually lower than everyone assumes, and how to compare honestly.
- [Rate limits and API quotas across a workflow](https://www.sigmalogicai.com/blog/api-rate-limits-workflows): Why adding workers produces 429s, how to budget concurrency across every API a workflow touches, and the backoff that actually works.
- [Writing content that assistants actually cite](https://www.sigmalogicai.com/blog/content-assistants-actually-cite): The page structures that survive being retrieved and extracted: answer-first openings, self-contained sections, real numbers, and claims a machine can attribute.
- [Context windows and why long documents still fail](https://www.sigmalogicai.com/blog/context-windows-long-documents): A large context window is not a document strategy. Why retrieval accuracy drops in the middle of long inputs, and what to do with a 200-page contract instead.
- [Measuring cost per resolved task](https://www.sigmalogicai.com/blog/cost-per-resolved-task): Total AI spend tells you nothing. The unit metric that does, how to compute it including the human cost of escalation, and why it beats deflection rate.
- [When to write a custom n8n node](https://www.sigmalogicai.com/blog/custom-n8n-node): The threshold where five chained HTTP nodes should become one custom node, what writing one actually involves, and the three cheaper options to exhaust first.
- [Cold email deliverability: what actually decides placement](https://www.sigmalogicai.com/blog/deliverability-for-ai-outreach): The 2026 sender rules from Google, Yahoo and Microsoft, the thresholds that get a domain blocked, safe volume per mailbox, and why generated volume makes it worse.
- [Document extraction: what to verify before you trust it](https://www.sigmalogicai.com/blog/document-extraction-what-to-verify): Field-level accuracy hides which field fails. The validation layer, confidence routing and reconciliation checks that make extracted data safe to post.
- [GEO vs SEO: what actually changes](https://www.sigmalogicai.com/blog/geo-or-seo): What genuinely differs when assistants answer instead of linking, what the click data now shows, how much of GEO is renamed SEO, and where the budget should move.
- [Golden datasets: how big is big enough](https://www.sigmalogicai.com/blog/golden-dataset-size): Why 150-400 cases is usually right, what the confidence interval actually looks like at each size, and why a smaller maintained set beats a larger neglected one.
- [The integration that silently stopped syncing](https://www.sigmalogicai.com/blog/integration-stopped-syncing-silently): A timeline of the failure nobody sees: what each person observed, why none of them connected it, and the three checks that would have caught it on day one.
- [Latency budgets for conversational AI](https://www.sigmalogicai.com/blog/latency-budgets-conversational-ai): Where the seconds actually go in a grounded answer, what users tolerate in chat against voice, and how to buy back time without losing accuracy.
- [Lead scoring that sales actually trusts](https://www.sigmalogicai.com/blog/lead-scoring-sales-trusts): Most scoring models are ignored within a quarter, and the cause is never the algorithm. It is no feedback from closed deals, no explanation, no definition.
- [llms.txt and structured data: what actually matters](https://www.sigmalogicai.com/blog/llms-txt-and-structured-data): What llms.txt is, who reads it and who does not as of 2026, the exact format, what schema markup demonstrably does, and where to spend an hour if you only have one.
- [Multi-agent systems: when the complexity pays](https://www.sigmalogicai.com/blog/multi-agent-systems-worth-it): The three conditions that justify splitting one agent into several, the costs nobody budgets, and why most multi-agent designs should be a workflow instead.
- [Running n8n in production: queue mode, workers and scaling](https://www.sigmalogicai.com/blog/n8n-queue-mode-scaling): Why the default single-process mode blocks under load, how queue mode with Redis and workers fixes it, and the four limits you hit next.
- [Version controlling n8n workflows](https://www.sigmalogicai.com/blog/n8n-version-control): How to get n8n workflows into git so changes are reviewable and revertable, what the JSON export does not carry, and the promotion path between environments.
- [n8n vs Make: which one, and for what](https://www.sigmalogicai.com/blog/n8n-vs-make): Where n8n and Make genuinely differ in 2026: self-hosting, credits versus executions, the code escape hatch, version control, with the billing arithmetic shown.
- [The prompt that worked until the input changed](https://www.sigmalogicai.com/blog/prompt-broke-when-input-changed): A failure pattern where nothing on your side changed: the inputs drifted, the prompt stayed, and accuracy fell for months before anyone connected the two.
- [Regression testing prompts like code](https://www.sigmalogicai.com/blog/prompt-regression-testing): How to put prompts under version control and CI so a change that improves one case cannot silently break nine others.
- [n8n error handling: retries, alerts and dead letters](https://www.sigmalogicai.com/blog/n8n-error-handling): The default is that a failed workflow is logged and nobody is told. The retry, error-workflow and dead-letter pattern that makes failure visible.
- [RAG permissions: stopping the assistant leaking documents](https://www.sigmalogicai.com/blog/rag-permissions): Why filtering results after retrieval leaks, how to enforce permissions at query time instead, and the four leak paths that survive a correct filter.
- [Scheduled or event-driven: choosing a trigger](https://www.sigmalogicai.com/blog/scheduled-or-event-driven): When polling on a schedule is the right answer despite being unfashionable, when webhooks are worth the delivery complexity, and the hybrid that most systems need.
- [Self-hosted n8n: when data rules force it](https://www.sigmalogicai.com/blog/self-hosted-n8n): When self-hosting n8n is genuinely required rather than preferred, what it actually costs to run, and the four things people forget until the first outage.
- [We shipped without an eval set](https://www.sigmalogicai.com/blog/shipped-without-an-eval-set): What the first year looks like when nobody measured: decisions made on anecdote, changes nobody can justify, and the cost of retrofitting too late.
- [Shopify AI app or custom integration](https://www.sigmalogicai.com/blog/shopify-ai-app-or-custom): When an app from the store does the job, when it hits a ceiling, and the three questions that decide it before you spend anything.
- [Structured output: getting reliable JSON from a model](https://www.sigmalogicai.com/blog/structured-output-json): Schema-constrained decoding, permissive parsing with strict validation, and why a valid object with wrong values is the failure that actually costs you.
- [Tool-calling agents: what to let them touch](https://www.sigmalogicai.com/blog/tool-calling-agent-permissions): How to tier the actions an agent may take, why the dangerous combination is read plus write plus untrusted input, and the controls that survive a bad plan.
- [Webhooks that drop: delivery, duplicates and ordering](https://www.sigmalogicai.com/blog/webhook-idempotency-replay): Why webhook delivery is at-least-once and unordered, what that means for your workflows, and the three defences that make an endpoint reliable.
- [Why automations fail silently](https://www.sigmalogicai.com/blog/why-automations-fail-silently): The automation that reports success while doing nothing is the expensive failure. Why status codes miss it, and the three assertions that catch it.
- [Why ChatGPT recommends your competitor](https://www.sigmalogicai.com/blog/why-chatgpt-recommends-your-competitor): How assistants decide which companies to name, why invisibility is usually a sourcing problem rather than a ranking one, and what you can influence.
- [Zapier, Make or n8n: a decision table](https://www.sigmalogicai.com/blog/zapier-make-or-n8n-decision): Twelve common automation jobs, and which of the three tools is the right answer for each - organised by what you are trying to do rather than by feature.
- [Migrating from Zapier to n8n: what breaks](https://www.sigmalogicai.com/blog/zapier-to-n8n-migration): What does not carry over when you move automations off Zapier, how long it really takes per workflow, and the migration order that keeps the business running.
- [Hallucination rate: what number is acceptable](https://www.sigmalogicai.com/blog/acceptable-hallucination-rate): Why a single hallucination rate is the wrong target, how to measure the fabrications that actually cost you money, and what thresholds are defensible by category.
- [Pilot to production: what the second invoice covers](https://www.sigmalogicai.com/blog/ai-pilot-to-production-cost): Why the gap between a working AI pilot and a production system is usually two to four times the pilot cost, and what that money actually buys.
- [What you own after an AI project: code, prompts and evals](https://www.sigmalogicai.com/blog/ai-project-ownership-code-prompts-evals): The eleven artefacts that decide whether you own an AI system or rent it, the contract language that secures them, and the handover test that proves it.
- [Build or buy AI customer support](https://www.sigmalogicai.com/blog/build-or-buy-ai-customer-support): A decision framework for AI support: five conditions that make buying correct, four that make building correct, and the hybrid most companies should actually run.
- [Building an eval set from real tickets](https://www.sigmalogicai.com/blog/building-an-eval-set-from-real-tickets): How to turn a support queue into a test set that catches regressions: sampling, stratification, labelling, and the cases most teams forget to include.
- [Confidence thresholds and escalation design](https://www.sigmalogicai.com/blog/confidence-thresholds-and-escalation-design): Why model confidence is a poor escalation signal on its own, what to combine it with, and how to tune the gate that decides whether a customer meets a person.
- [Cutting LLM costs without degrading quality](https://www.sigmalogicai.com/blog/cutting-llm-costs-without-degrading-quality): Eight cost reductions ranked by quality risk, from the ones that are free to the ones that are a real trade - and how to prove which is which before shipping.
- [Fixed price or time and materials for AI projects](https://www.sigmalogicai.com/blog/fixed-price-or-time-and-materials-ai): When a fixed price protects you on an AI build, when it quietly costs you more, and the phased structure that avoids the worst of both.
- [Fractional AI lead: when it beats hiring](https://www.sigmalogicai.com/blog/fractional-ai-lead-vs-hiring): What a fractional AI lead actually does, when it is the right shape against hiring or using an agency, and the four conditions under which it fails.
- [How to evaluate an AI agency proposal](https://www.sigmalogicai.com/blog/how-to-evaluate-an-ai-agency-proposal): Fourteen questions that separate an AI proposal that will ship from one that will produce a demo, plus the answers that should end the conversation.
- [How to measure whether an AI system works](https://www.sigmalogicai.com/blog/how-to-measure-whether-an-ai-system-works): A practical method for evaluating a production AI system: choosing the metric, building the test set, setting a baseline, and knowing when a score has moved.
- [Your provider changed the model - what broke](https://www.sigmalogicai.com/blog/provider-changed-the-model-what-broke): A post-mortem pattern: the AI system that degraded without a deploy. Why model updates break working systems, how to detect it, and what to pin.
- [Scoping an AI project that actually ships](https://www.sigmalogicai.com/blog/scoping-an-ai-project-that-ships): Most stalled AI projects were scoped wrong on day one. The predictor is whether the scope was drawn around a workflow or a capability, and four questions settle it.
- [Better decisions, not just faster ones: where AI actually helps](https://www.sigmalogicai.com/blog/where-ai-improves-decisions): Speed is the easy win. The harder question is which recurring decisions are worse than they need to be, and which of three failure modes you are facing.
- [Where the hours actually go: finding automatable work](https://www.sigmalogicai.com/blog/finding-automatable-work): A method for locating the work AI can take off your team without a six-month process-mining exercise, including the four shapes that look automatable and are not.
- [Responsible AI as an engineering decision, not a policy document](https://www.sigmalogicai.com/blog/responsible-ai-engineering-decisions): Disclosure, escalation, data handling and explainability are architecture choices. Cheap in week one, and projects if you retrofit them at month twelve.

- [How we work](https://www.sigmalogicai.com/how-we-work): the four-phase delivery model, principles, and what we deliberately do not do

## Optional

- [About](https://www.sigmalogicai.com/about): who we are, the named author, and what we believe
- [Tools index](https://www.sigmalogicai.com/tools): every free tool on one page
- [Contact](https://www.sigmalogicai.com/contact): start a conversation
- [Privacy](https://www.sigmalogicai.com/privacy): what the site does and does not collect