Discovery Questions — Empire Teas CRM
Date: 2026-06-30 Purpose: These questions must be answered before CRM data model design can begin. Answering them completes the discovery phase. Total questions: 47 (consolidated and deduplicated from initial-findings.md, sri-lanka-tea-industry.md, and ai-in-workflows.md)
How to use this document
- Questions marked Team can be answered internally if someone has Pipedrive access.
- Questions marked Client must go to Empire Teas directly.
- Work through tiers in order — Tier 1 blocks everything else.
Tier 1 — Blocks Architecture
Must be answered before any CRM design decision. Wrong answers here mean rebuilding from scratch.
| # | Question | Why it must be answered | Who answers |
|---|---|---|---|
| 1 | How many active private label clients vs bulk buyers do you have approximately? | Private label deals have 8+ stages (sample rounds, artwork approval, compliance check). Bulk deals have 3–4 stages and a fast cycle. These are fundamentally different pipelines. If you build one pipeline and later discover you need two, you rebuild from scratch. | Client |
| 2 | What new ERP are they moving to, and does it have an API? | When a deal is won in the CRM, it must hand off to the ERP. If the ERP has an API, the handoff is automated. If it doesn’t, the CRM needs to produce a structured output document instead. The entire integration layer changes based on this answer. | Client |
| 3 | Which teams use the CRM today — sales only, or also export, QA, management, customer service? | Each team needs different views, fields, and permissions. A QA person needs cert and sample fields. An export coordinator needs document checklists. A manager needs pipeline dashboards. If you only design for sales and discover others also need access, you have to add modules mid-build. | Client |
| 4 | What data is still tracked outside the CRM — in Excel, WhatsApp, email, or physical files? | If sample tracking lives in a spreadsheet and quote history lives in someone’s inbox, those are missing from the CRM’s scope. If you don’t discover these external systems upfront, you build a CRM that’s incomplete from day one and people keep using the spreadsheet. | Client |
| 5 | Are leads and deals both used in Pipedrive? If yes, what is the distinction between them? | In Pipedrive, leads are unqualified and deals are active opportunities. If they use both, the new CRM needs a lead stage before the pipeline. If they only use deals, the pipeline starts earlier. This changes the first screen a salesperson sees. | Client / Team |
| 6 | How many total records need migration — leads, contacts, organisations, deals, notes, activities? | 500 records is a weekend migration job. 50,000 records needs a full migration script with deduplication, validation, and a rollback plan. The answer determines how much engineering time the migration alone will take. | Team |
Tier 2 — Shapes the Core Data Model
These determine fields, dropdowns, and forms. Getting these wrong means painful data entry from day one.
| # | Question | Why it must be answered | Who answers |
|---|---|---|---|
| 7 | Which tea grades do you work with most (BOP, BOPF, OP, Dust, etc.)? | Every deal and product brief needs a grade field. If you include all 15 possible grades as dropdown options it’s cluttered. If you include the wrong 5, sales reps type free text which breaks filtering and reporting. The answer gives you the exact dropdown list. | Client |
| 8 | What does a typical private label order look like — what volume and pack format (teabags, loose leaf, tins, sachets)? | Volume is a core field on every deal, but the unit matters. A bulk deal is quoted in MT. A private label deal might be 10,000 boxes of 25 teabags. If the volume field uses the wrong unit, reps convert manually every time or enter inconsistent data. | Client |
| 9 | Which certifications apply to which specific products or SKUs? | This is the data behind one of the most valuable CRM features — automatic cert mismatch detection. When a buyer who requires EU Organic is being quoted, the CRM checks whether the specific product being offered actually holds EU Organic. Without this product-to-cert mapping, the feature cannot be built at all. | Client |
| 10 | How do you currently track which certifications each buyer requires? | This tells you where the cert data lives today. If it’s in a spreadsheet, it needs to be migrated into buyer account records as structured fields. If it exists only in someone’s memory, it’s a knowledge-loss risk the CRM must fix. The answer determines the migration scope for cert data. | Client |
| 11 | What payment terms are standard for your main markets — LC, open account, DA, DP? | If Letter of Credit is common — likely for new buyers in the Middle East and Russia — the quote record needs LC-specific fields: issuing bank, LC number, document submission deadline, LC expiry date. That is a different data model than open account. Build the wrong one and it needs to be redesigned. | Client |
| 12 | What custom fields currently exist in Pipedrive? | These fields represent decisions already made about what data matters. They need to be mapped to the new CRM. If a field like “sample feedback status” or “buyer certification requirement” exists and you don’t map it, that data is permanently lost in migration. | Team |
| 13 | Which of those custom fields are business-critical and must survive migration into the new CRM? | Not every custom field is worth migrating — some are abandoned experiments. The client must identify which ones are actively used so the migration script maps them correctly and the new CRM’s data entry forms include them from day one. | Client / Team |
| 14 | What reports and dashboards do owners currently use in Pipedrive? | If management uses a specific pipeline report every Monday, the new CRM must be able to produce an equivalent on launch day. If you don’t know which reports are relied on, you ship a CRM that breaks someone’s weekly workflow. | Client / Team |
| 15 | What automations are currently active in Pipedrive? | Some automations may be business-critical — a follow-up reminder after a quote is submitted, or an alert when a deal goes dormant. These need to be replicated in the new CRM. If you don’t inventory them, people lose automation they depended on without realising it until something falls through the cracks. | Team |
| 16 | What integrations are currently connected to Pipedrive (email, accounting, other tools)? | Any active integration — email sync, an accounting tool, a document tool — must be planned for in the new system. An integration that silently stops working after migration will cause data loss or broken workflows. | Team |
Tier 3 — Shapes Pipelines, Stages, and Automation Rules
These determine stage names, transition rules, and when alerts and reminders fire.
| # | Question | Why it must be answered | Who answers |
|---|---|---|---|
| 17 | What are the current pipeline stages in Pipedrive? | The current stages are the starting point for redesign. Even if they are wrong, knowing what they use tells you what the team is already comfortable with and where the friction is. Designing stages in a vacuum produces something unfamiliar that no one adopts. | Team |
| 18 | How many pipelines exist in Pipedrive today? | Confirms whether they currently distinguish between deal types or force everything into one pipeline. Feeds directly into the one-pipeline-vs-two decision from Q1. | Team |
| 19 | How many rounds of samples does a typical private label deal go through before the buyer approves? | The sample tracking module must match reality. If deals typically go through 3 rounds, the data model needs Sample Round 1, Round 2, Round 3 — each with its own dispatch date, feedback field, and status. A single-round model breaks on real data immediately. | Client |
| 20 | How long does a typical private label deal take from first inquiry to first confirmed order? | Sets the stale deal alert threshold for the private label pipeline. If the average cycle is 5 months, a deal with no activity for 14 days is completely normal. Set the alert to 14 days and salespeople get flooded with false positives and start ignoring all alerts. | Client |
| 21 | How long does a typical bulk deal take from first inquiry to first confirmed order? | Same as Q20 but for the bulk pipeline. Bulk cycles are short — weeks not months — so the stale threshold is different. The two pipelines will have different alert timings. | Client |
| 22 | Are follow-ups currently being missed? | Confirms whether automated follow-up reminders are solving a real problem or adding complexity to something that already works. If the answer is yes, this feature becomes a priority. If no, you need to understand what they’re doing right so you don’t break it. | Client |
| 23 | Can they track sample requests end-to-end today? | Confirms whether the sample tracking module is a new capability or a replacement for something that already exists elsewhere. If they track samples in a shared spreadsheet, that spreadsheet’s structure tells you what fields the module needs. | Client |
| 24 | Can they track quote status and quote aging today? | Confirms whether quote aging alerts and a quote dashboard are solving a gap or duplicating something. Also reveals whether quote data currently lives in the CRM or somewhere else like email or Excel. | Client |
| 25 | Can they see full customer history before a call or meeting? | Confirms whether the AI account briefing feature is solving a real pain. If salespeople currently spend 20 minutes digging through emails before every call, automatic account summaries are high value. If they have a working system, it’s lower priority. | Client |
| 26 | Can they identify dormant accounts or customers due for reorder? | Confirms whether a dormant account reactivation automation is needed. Tea buyers reorder on cycles — if no one is tracking that and accounts are going cold, this is a direct revenue recovery feature. | Client |
| 27 | Is management pipeline visibility good enough currently? | If management cannot trust the pipeline report, it means data discipline is low — reps are not updating stages consistently. This changes the rollout plan: you need a data discipline process alongside the CRM build, not just better dashboards. | Client |
| 28 | What does Pipedrive not do well for them today — open-ended? | This is the most important qualitative question. Specific answers to other questions tell you what to build. This question surfaces pain you didn’t think to ask about. A client saying “we can never find old quotes” or “we don’t know which samples are waiting on feedback” reveals requirements no predefined question would catch. | Client |
Tier 4 — Shapes AI Features
These block specific AI workflows. Only matter once CRM foundation is decided, but must be answered before AI development starts.
| # | Question | Why it must be answered | Who answers |
|---|---|---|---|
| 29 | Where do inquiry emails arrive — shared mailbox, individual inboxes, website form, WhatsApp? | Determines the AI trigger point. Shared Gmail inbox = Gmail API. Outlook = Microsoft Graph API. WhatsApp = WhatsApp Business API. A mix of channels = multiple triggers. Build the trigger for the wrong channel and the AI never fires on real inquiries. | Client |
| 30 | Are customer communications centralised, or spread across individual sales reps’ inboxes? | If communications are spread across personal inboxes, the AI cannot see the full conversation history when generating account summaries or drafting replies. This changes the RAG design and may require an email forwarding rule or a shared inbox policy before AI features work. | Client |
| 31 | What language are inquiries typically received in — English, Arabic, Turkish, Russian, other? | AI extraction accuracy drops in non-English languages. Arabic and Russian especially need separate prompt testing before the feature goes live. If 30% of inquiries arrive in Arabic and you only test in English, the extraction fails silently on a large portion of real traffic. | Client |
| 32 | Are there data privacy or confidentiality concerns about sending email content to an external AI API? | If the answer is yes, you cannot send buyer email content to Claude’s cloud API. You would need a self-hosted AI model, which is a fundamentally different and significantly more expensive architecture. This is a binary decision that must be made before any AI development begins. | Client |
| 33 | Where are product specs, certifications, and SOPs stored today — Google Drive, SharePoint, physical files? | Required to build the RAG knowledge base and the export documentation checklist. If everything is in Google Drive, you connect to the Drive API. If documents are in physical binders, someone must digitise them before any document AI feature can work. The feature cannot be built until the source is accessible programmatically. | Client |
| 34 | What data can AI safely act on, and what must always go through human approval before action? | Defines the human-in-the-loop boundaries. If pricing, compliance decisions, and export clearance always require human sign-off, the AI must present these as drafts — never auto-apply. Getting this wrong and having AI auto-apply something that needed approval is a trust-breaking incident that kills adoption. | Client |
| 35 | Who reviews and approves AI outputs — one designated person or multiple sales reps? | Determines whether the AI review queue is a single inbox or a team-wide queue with routing logic. If multiple reps each review their own AI outputs, the queue needs ownership assignment. If one person reviews everything, it’s a simpler single-queue design. | Client |
| 36 | Is the team comfortable with AI tools, or is there likely to be resistance to adoption? | If the sales team doesn’t trust AI suggestions, they ignore the review queue and the entire human-in-the-loop model breaks down. Knowing this upfront lets you design AI features as optional hints rather than mandatory steps, and plan for a training session before launch. | Client |
| 37 | Which single workflow would save the most time in the next 60–90 days? | Determines the AI pilot. The three candidates are inquiry capture, lead triage of the 200+ stale Pipedrive leads, and sample/quote tracking. Picking the wrong pilot means delivering something that doesn’t feel valuable to the people using it daily, which hurts adoption of the rest of the system. | Client |
Tier 5 — ERP Integration Design
These determine what the CRM outputs when a deal is won and what comes back from the ERP.
| # | Question | Why it must be answered | Who answers |
|---|---|---|---|
| 38 | What should happen operationally when a deal is marked as won in the CRM? | The CRM-to-ERP handoff trigger. Does winning a deal automatically create a sales order in the ERP? Does it notify a specific person? Does it generate a document? Without knowing this, you cannot design the “won” stage or the automation that fires when it’s reached. | Client |
| 39 | What customer and order data must move from CRM to ERP when a deal is won? | The outbound data payload. Every mandatory field in the ERP’s sales order must be present and filled in the CRM deal before it can be marked won. If a required ERP field doesn’t exist in the CRM, the handoff fails or requires manual data entry — which defeats the purpose of integration. | Client |
| 40 | What ERP data should come back into the CRM — order status, shipment status, invoice status, documents? | Determines whether the CRM is a sales-only tool or whether it also shows post-order visibility. If a buyer asks a salesperson “where is my shipment?”, should the salesperson be able to answer from the CRM without switching to the ERP? The answer shapes whether a sync-back is needed at all. | Client |
| 41 | Is there already an API or integration plan defined for the new ERP? | If the ERP has no integration plan or no API, the CRM-ERP handoff must be document-based or manual. This is not a blocker for building the CRM, but it changes what the “won” stage does and must be known before coding the handoff. | Client |
Tier 6 — Migration Cleanup Details
Answer these once you have access to Pipedrive. Most can be resolved by the team without a client meeting.
| # | Question | Why it must be answered | Who answers |
|---|---|---|---|
| 42 | How long have they been using Pipedrive? | Older usage means more accumulated data, more embedded bad habits in the data, and higher risk of inconsistent field usage over time. A 5-year-old Pipedrive instance needs a much more careful migration than a 1-year-old one. | Team |
| 43 | How many active users are on Pipedrive today? | Determines the licence count for the new CRM and the scope of training. 3 users is a one-afternoon training session. 20 users across multiple offices needs a structured rollout plan. | Team |
| 44 | Is email sync enabled in Pipedrive? | If email sync is on, historical customer emails may be stored in Pipedrive and could be migrated or used to seed the AI’s account briefing context. If it’s off, email history lives only in individual inboxes. | Team |
| 45 | Are activities used consistently across the team, or only by some users? | If only 2 of 10 reps log activities, the activity history in Pipedrive is incomplete and unreliable. This changes how much weight you put on migrating activity data — it may not be worth the effort if the data is sparse. | Team / Client |
| 46 | Do they want all records migrated, or only cleaned and qualified records after AI triage? | Migrating everything including spam, duplicates, and dead leads pollutes the new CRM from day one. Migrating only qualified records means the AI triage step happens before migration. This decision shapes the entire migration plan and timeline. | Client |
| 47 | Should original Pipedrive IDs and source history be preserved for audit and reference purposes? | Low effort to preserve during migration, but someone must decide. If they ever need to cross-reference a deal in the new CRM back to the original Pipedrive record, the ID must be stored. If no one needs that, it’s unnecessary clutter. | Client |
What happens after these are answered
- Pipedrive audit — log into Pipedrive directly, export records, document all fields and stages hands-on (not a questions task — requires access)
- Data model design — translate answers into entity schema, field list, pipeline stages, and dropdown values; client reviews and signs off before any code is written
- Tech stack decision — backend language, database, frontend framework
- UI/UX wireframes — key screens: pipeline board, deal view, account page, sample tracking
- Integration contracts — confirm ERP API spec before building the handoff
- Implementation begins