What to gather and the 30–90 minute plan
Set aside a session with one person who knows your HubSpot, the main customer spreadsheet and your accounting system (or the person who runs invoices). Total time: 30 minutes for a quick read, 60–90 for a proper pass.
Run these three real queries against each source: a) "Show the latest invoice and payment status for customer [named customer], including invoice date and amount"; b) "Summarise interactions for contact [named contact] in the last 12 months and list any open deals or next actions"; c) "Which active customers have had no contact in 90 days and have any unpaid invoices?".
For each query, capture the raw answer (copy the HubSpot search results, spreadsheet filter rows, and accounting record or screenshot). Note how long each query took and whether any manual joining or guessing was needed.
Score answers on five practical dimensions
For each query and each source score 0–2 on: Coverage (is the item present?), Freshness (is the data up to date?), Unique identifier (is there a reliable key like customer ID or email?), Provenance & privacy (can you trace where it came from and is it appropriate to use?), Answerability (could an assistant produce a useful answer from this data?).
- 0 = fail or missing; 1 = partial or needs manual steps; 2 = good and straightforward.
Add each query score (max 10). Interpret totals: 0–3 = not ready, 4–7 = limited usefulness (expect many manual fixes), 8–10 = suitable for narrow AI tasks. While you score, note repeated issues (e.g. missing invoice numbers, contacts without emails, stale last-contact dates).
Typical failures, focused fixes and next steps
Common failure patterns: missing or inconsistent IDs (no reliable join key), stale dates (last-contact or invoice dates not recorded), split records across spreadsheet and HubSpot, and privacy flags or consent notes buried in free text. For each pattern pick one small fix: add a persistent customer ID, enforce a last-contact date on each update, move a master copy of invoices to the accounting system only and link the ID in HubSpot, and standardise consent fields instead of free-text notes.
Decide one of four next steps per source: Clean (dedupe, standardise core fields), Connect (add a reliable link between systems), Restrict (limit what an assistant may access and strip PII), or Delay (wait until a critical field is fixed). Record the single highest-impact change you can make in 1–3 hours and one follow-up owner and deadline.
If you want a short result sheet or someone to run the test with your team, Optira can help as a practical helper on the next steps.