Softpiper Solutions

Solutions

How we solve business problems with Claude and Gemini, on your existing stack.

We start from the systems you already run (your store, CRM, ERP, inbox and data warehouse) and add AI where it earns its place: agents that answer and act, document workflows that extract and draft with citations, and content pipelines that keep up with your catalogue. Claude where reasoning, tools and traceable documents matter; Gemini where voice, video and generated media do.

Built with Claude · Gemini · Shopify · Wix

Browse all 19 playbooks

Our approach

Production AI without a rip-and-replace

Most of the value is in the systems you already trust. We add a thin, well-governed AI layer on top of them, prove it on one workflow, then widen it.

  • Start from your systems of record

    We connect to what you run today through official APIs and MCP servers: Shopify, HubSpot, Zoho, SAP, TallyPrime, WhatsApp, Google Workspace and Microsoft 365. Your data stays where it lives.

  • Choose the model per task

    Claude and Gemini lead in different places. We pick per step on evaluation results, cost and latency, and often use both in one workflow.

  • Deploy where your data must stay

    Claude runs on Anthropic's API, AWS, Google Cloud and Microsoft Foundry; Gemini on the Gemini API, Google Cloud and Workspace. Feature sets differ by platform, so we map them in discovery.

  • Ship with guardrails, prove it with evals

    Strict tools, human approval for writes and money, audit logs, and an evaluation set of real cases before launch. We agree the KPIs with you up front.

Reference architecture

The pattern behind every playbook

Your systems stay the source of truth. The AI layer reads from them, acts through scoped tools, asks a person before anything consequential, and reports against the numbers you care about.
Your systems (store and website, CRM and ERP, messaging, documents and data warehouse) feed Softpiper's AI layer. It takes requests in through web chat, WhatsApp or voice, routes them with a small fast model, reasons and acts with Claude through scoped tools, creates media with Gemini when needed, grounds answers in your data with citations, and logs and evaluates everything. The outcomes are measured as response time, cycle time, error rate and cost per task.

Your stack

  • Store and websiteShopify, WordPress, Wix
  • CRM and ERPHubSpot, Zoho, Salesforce, SAP, Tally
  • MessagingWhatsApp, email, Slack, Teams
  • DocumentsDrive, SharePoint, PDFs
  • Data warehousePostgres, BigQuery, Snowflake

AI layer

Softpiper AI layer

  • Claude
  • Gemini
  1. Intake, using Web chat · WhatsApp · Gemini Live

    Web chat, WhatsApp or a voice line, with calls transcribed as they happen.

  2. Route, using Claude Haiku or Gemini Flash-Lite

    A small, fast model classifies intent and risk and sends each request down the right path.

  3. Reason and act, using Claude Sonnet / Opus · MCP

    A stronger model works through your systems with strict, schema-checked tools. Writes wait for approval.

  4. Create media, using Nano Banana · Veo

    Product shots, posters and short video when the job needs them, checked against your brand rules.

  5. Ground in your data, using Citations · embeddings

    Retrieval with citations over your documents and warehouse, so answers can be checked.

  6. Govern and measure, using Evals · audit logs

    Audit logs, evaluation sets and cost tracking: Agent Gateway on Google Cloud, OpenTelemetry elsewhere.

Outcomes you measure

  • Response and resolution time
  • Cycle time per order or document
  • Error and rework rate
  • Cost per resolved task

Choosing models

When we choose Claude, and when we choose Gemini

Each leads in different places. We choose per task on evaluation results, cost and where your data has to stay, not on brand loyalty.

We lean on Claude for

ClaudeAnthropic

  • Agents that act across several systems

    Strict, schema-checked tool use over MCP, long multi-step runs, and human approval wired in before writes.

  • Documents where every answer must be traceable

    Native PDF reading with citations to the passage used, structured outputs, and formatted Word, Excel and PowerPoint files.

  • Coding, migrations and dev automation

    Claude Code in the terminal, IDE, GitHub Actions and GitLab CI, with scheduled Routines and the Agent SDK.

  • Microsoft 365 or AWS-first organisations

    Claude for Microsoft 365, Claude in Microsoft Foundry billed through Azure, and Claude in Amazon Bedrock.

We lean on Gemini for

GeminiGoogle

  • Real-time voice and live translation

    The Gemini Live API handles natural, interruptible voice in 70 languages with function calling, plus transcription and text-to-speech.

  • Generative images, video and music

    Nano Banana image models, Veo and Gemini Omni Flash video, and Lyria music. LuxeLens, our own product studio, runs on Gemini image models and Veo.

  • Understanding video

    Agentic video understanding for procedure clips, lectures and ad review.

  • Google Workspace teams and live place data

    Gemini built into Gmail, Docs and Sheets, Workspace Studio flows, and Grounding with Google Search and Maps.

ClaudeGemini

Often, both

A voice line on Gemini can hand a transcript to a Claude agent that updates your CRM. We design the hand-off so each model does the part it is best at.

  • One governed platform on Google Cloud

    Claude and Gemini sit side by side in the Gemini Enterprise Agent Platform (formerly Vertex AI), under one bill and one set of controls.

  • High-volume routing and extraction

    Claude Haiku or Gemini Flash-Lite. We choose on your evaluation results and the cloud you already use.

Capabilities checked against Anthropic and Google documentation in October 2026. Model versions change often; we pick the current one at kickoff.

Playbooks

19 playbooks by industry and problem

Each is a reference architecture: the problem, how we would solve it on your stack, the guardrails, a typical planning range and the KPIs we would agree to track. Starting points for a conversation, not past results.

19 of 19 playbooks

On every build

Guardrails come first

These are written into the architecture and tested in the evaluation set, not added to a prompt at the end.

  • Facts come from your systems

    Prices, stock, dates and ETAs are read from your systems of record. The model explains them; it never invents them.

  • People approve writes and money

    Refunds, orders, postings, quotes and anything regulated wait for a named person to approve.

  • A human is always one step away

    Every assistant says it is an AI and hands over to your team with the full context.

  • Audit logs and evals

    Every tool call is logged, and an evaluation set of real cases runs before launch and after every change.

  • Data minimisation

    Only the fields a workflow needs are collected, with residency and retention set to your rules.

  • Channel and content rules

    WhatsApp opt-in and template rules, AI-content disclosure where required, and rights checked on media inputs.

Pip · Softpiper's AI

Not sure which fits?

Describe the problem. Pip will map it to a playbook.

Pip is Softpiper's AI. Tell it what you're trying to fix and what you run today; it will suggest the closest playbook, explain the model choice and give you a tailored price range. Or book a call with the team.

Book a call