Signals

What your customers asked for, counted

You already have the answer to "what do people keep asking us for that we do not sell". It is sitting in a year of WhatsApp conversations that nobody has time to read. Signals reads them.

Two questions, and they are the ones with money in them

What did people want that you could not give them. Not the word "delivery" appearing a lot - the eleven different people who asked for one size you have never stocked, with the conversations behind the eleven so you can go and look.

Which advertising actually worked. A click-to-WhatsApp advert attaches its id to the first message somebody sends. Njiwa keeps it, so the spend and the conversation it bought are the same record.

Worth knowing before you start: an advert id that arrived before you turned Signals on is gone. There is no archive it can be recovered from. The day you switch it on is the first day you can attribute.

The free report has no AI in it

Terms and phrases pulled out of your own messages by Postgres, ranked by how many different people said them rather than how many times they were typed - so one customer sending forty messages about one thing is one person who wanted one thing, not a market.

  • What is rising this week against last.
  • When messages actually arrive, by day and by hour and by number.
  • How often you reply, and how long the first reply takes - counted twice, once with your autoresponder and once without it, because a robot answering in four seconds is not you answering.
  • Who is still waiting, opening onto the conversations themselves.

No model sees anything to produce this, and it costs nothing.

The deep read is the paid one

This one reads whole conversations - what somebody asked for, what you said back, and how it ended - and produces the things counting words cannot reach:

  • What people wanted that you could not serve, by name and by the exact thing they said, so "tyres" becomes "195/65 R15".
  • Where conversations die. Whether people stop replying after you quote a price, after you say how long delivery takes, or because nobody answered them at all.
  • Whether answering faster changes anything on your account, in your trade - not as general advice.
  • The questions you answer over and over, which are the ones worth putting in an auto-reply.

Every figure opens onto the people behind it

A report that says thirteen people were never answered is worth nothing if you cannot find the thirteen. Click the figure and they are there, by name, and clicking one opens that conversation in the Desk with the reply box under it. The list and the figure are the same count, always.

Reading again pays only for what changed

Read a month on Monday and again on Friday, and Friday charges for the conversations that moved. The rest come back out of what Monday already paid for - and the report still covers the whole month, because a report over only the new conversations would rank a month of demand by what four people asked this week.

Groups are left out unless you ask for them

A WhatsApp group on a business number is usually the staff rota or the suppliers' thread, and several people talking among themselves can put a want at the top of a table that no customer ever asked for. They are out by default. Switch them in and the report says how many of every figure came from a group, and you can look at either half on its own.

What it costs

Nothing is included in a package and nothing is charged to your card without you pressing a button. Signals credit is prepaid: you top it up, and a read takes what it takes out of the balance.

  • You are told what a read will cost before it runs, worked out from the actual text that would be sent rather than from an average. Two accounts with the same number of conversations do not cost the same, and a rate card cannot know which one you are.
  • The free report stays free, however often you open it.
  • Every movement is on a statement you can download as a spreadsheet, or have sent to a WhatsApp number as a document.
  • Switching Signals off does not touch the balance. Money you have paid stays paid. Ask us and we will refund what is unused to the card that bought it - the terms say how.

Nobody is opted into this

The deep read sends what your customers wrote to a model at Anthropic. That is a real thing to agree to, so it is agreed to explicitly, by an owner, once, in a dialog that says so in those words - and all of Signals sits behind it, including the free report.

It also collides with something else Njiwa lets you do, and the collision is honest rather than hidden: an account that has told Njiwa not to keep message content cannot opt in at all. There is nothing to read, and switching it on to produce an emptier answer nobody could tell was emptier would be the dishonest version.

What it does not do. It does not tell you what any one customer is like - the readings carry no names by instruction, and the same conversation from twenty people is one reading. It does not predict anything. It does not produce a number that is not in the tables: the last stage of a read drops any finding quoting a figure the counting did not produce, which is the whole reason the rest can be trusted.

Create an account See it beside everything else