Connect by JBRH Open Connect

Delivery and order status enquiries

Connect answers a status question from the records in your workspace — the Data grid's sheets, an uploaded manifest, Knowledge — and says it does not know when none of those carry the answer. There is no courier-tracking integration in the capability registry, so whatever your operational system knows reaches Connect as data you put there, or it does not reach it at all.

Status
Available What this means
Audience
both
Channels
emailwhatsappphone
In the app
#/data, #/inbox, #/support
Last verified
Product version
6.3.2

Where a grounded answer comes from#

  • The Data grid — thirteen sheets over the workspace's own records: companies, people, leads, prospects, follow-ups, deals, conversations, calls, cases, onboarding, knowledge, files and memory. Queried, not sampled.
  • Files — CSV, TSV, XLSX, PDF, DOCX and more, read by the shared file service. A daily manifest uploaded as a spreadsheet is a legitimate source and answers can cite it.
  • Knowledge — the durable statements: your cut-off times, your service levels as written, your exceptions policy.
  • The relationship record — what this customer was told last time, on whichever channel they used.

The Assistant reaches these through read-only tools — data_query, read_file, ask_file, find_anything, customer_360 — before it reaches any writing one. An answer that cannot be traced to one of them is not an answer this system is willing to give.

Why an answer from a sample is the dangerous failure#

This is not theoretical here. prospect_summary once derived every headline number from the newest 500 rows: driven with 540 contacts whose 40 already-contacted ones were the oldest, it reported 500 organisations when there were 530, and 0 already contacted when there were 40. Nothing errored. The numbers looked exactly as authoritative as correct ones.

Counting is done by the database now, and the conversation list — which used to read 400 rows and then filter, search, count and page in Python over that slice — does its counting in one GROUP BY with the filter and the page in SQL. The reason this matters on a delivery desk is that a status answer built from a slice of a manifest is confidently wrong in exactly the same way, and a customer has no way to tell.

What Connect will refuse#

An ETA it cannot see
If nothing in the workspace carries a delivery date, the honest reply says so and offers the next step. No estimate is manufactured to fill the silence.
A refund, credit or goodwill gesture
safe_sales refuses commercial terms Knowledge does not support and escalates to a person. A published compensation policy is a fact and can be quoted; a discretionary gesture is a decision.
A claim about a third party's system
Connect has no visibility into a courier's network. It can report what your own record says and when it was last updated, which is a different and more defensible statement.
Clearing a contact off the do-not-contact list
A person, recorded against their name. The Assistant cannot do it.

When it becomes a case#

  1. Let the enquiry resolve to a Person and a conversation as normal.

    Result The history is in one place regardless of which channel the customer switched to when they got annoyed.

  2. Open a support case when the answer is 'this has gone wrong' rather than 'here is where it is'.

    Result The case carries context and has a resolution path, and it does not get lost among ordinary status questions.

  3. Create a follow-up with a reason for anything promised — a chase with the depot, a call back tomorrow.

    Result The commitment exists outside the conversation and is visible when overdue. Phone chases drain two per tick and a row more than 24 hours late is closed as missed rather than rung.

  4. Use the task channel for the parts only a person can do.

    Result No drain ever sends a task follow-up; it shows as work for a person on the follow-ups screen.

Where the money goes#

Status enquiries are high-volume and low-value individually, which makes them the best possible fit for written channels and the worst for voice. Email and WhatsApp cost one model call per answer; a phone call costs audio tokens at four times the text input rate with the whole session context re-billed every turn. Outbound SMS notifications, which would be the obvious cheap channel, do not run on the live carrier at all — it carries no SMS — and in India would need DLT registration with the operators regardless.

Questions#

Can Connect look up a tracking number with the courier?

No. There is no courier or tracking integration in the capability registry. What it can do is answer from your own records — a manifest you upload, a data sheet you maintain, what the customer was told last time — and say plainly when none of those carries the answer.

How do I know an answer was not made up?

Answers are grounded in the workspace's own sources and the Assistant's read-only tools are what reach them. The decision log records what was decided, under which rule, and what happened, including refusals. If a reply cannot point at a source, that is visible rather than hidden behind confident wording.

Can it send delivery notifications by text?

Not on the live carrier, which carries no SMS. Inbound SMS, STOP handling and suppression exist, so the channel is real but one-directional here; outbound needs a provider that carries SMS, and in India DLT entity, header and template registration with the telecom operators.