Categories The Engagement Desk

Two kinds of marketing platform are emerging. The one that closes the loop wins.

The Engagement Desk, Issue 02: The one that closes the loop wins

This is issue 02 of The Engagement Desk, my newsletter for marketing teams in India running email and WhatsApp. It went to subscribers on 17 November 2026.

For most of this year the AI question on our demo calls was “does your tool have AI?” Lately it’s “whose AI?” Teams already pay for Claude or ChatGPT, and they want to know whether their marketing platform will work with it or hand them one more assistant.

That question is splitting marketing platforms in two. One kind builds its own agent into the product. The other opens its data and actions to the assistant you already use, through MCP, an open standard for exactly this. I run most of our own marketing on the second kind now, and this issue is about what that changes.

One take: the daily loop, and why it has to be closed

The biggest change isn’t AI writing copy. It’s that a marketing team can now run the whole cycle every day. Ask your assistant what happened yesterday, and it reads the campaign and automation reports from the live account. Ask it to fix what’s weak, and it drafts the next campaign, the new automation branch or the changed audience right there. A person checks it and presses send. A cycle that used to take a week of exports, meetings and tickets now fits in a morning, and the person who spots the problem is the one who fixes it.

That alone is useful. The real power shows up when the loop is closed. A campaign tool on its own only knows opens and clicks. It doesn’t know which of those leads booked a call, bought, came back or turned out to be junk. That knowledge sits in the CRM, the booking tool, the store or a sales rep’s sheet. When those outcomes flow back to the same customer record, next to the ad they came from and every message they got, your assistant can finally answer the questions that matter. Which source brings customers, not just leads? Which message moves people to pay? Which segment buys again? Finding patterns like these used to be the slow part: exports from five tools, rows matched by hand in a spreadsheet, days before anyone saw an answer. Now Claude or ChatGPT does it in one conversation. So the job has changed. It’s no longer finding the patterns. It’s making sure every system hands its data to your assistant.

Here’s what that looks like for us. We run our outbound with Claude connected to CampaignHQ. It finds the companies, writes the emails and sets up the journey. For a while that was all it could do, and it wasn’t very effective. Claude could send, but it couldn’t see what happened next.

Then we connected everything. My Claude now works across our whole in-house stack: CampaignHQ for email, WhatsApp and Instagram, PostHog for the website, Google Ads and Search Console, Gmail, our call notes from Fathom, the blog on WordPress, Google Drive, and Kumo, where our tasks live. Now, for every lead, it can see how they engaged with each email, which link they clicked, when they came to the website from which campaign and what they looked at there and when they converted. Put together, that stops being a list of metrics and becomes a lead story.

And because the same assistant can act, the story turns into the next step the same day, not next week. None of that is clever AI. It’s a closed loop, with all the data where the assistant can read it, and a person deciding what goes out.

This is why I think the second kind of platform wins. A built-in agent sees only its own product, and the loop always runs across several systems. The platform that matters is the one that takes outcomes in from your other tools and lets your own assistant read across all of them. That’s the bet we’ve made with CampaignHQ, and it’s our commercial interest too, so weigh my view with that in mind.

What changed: both kinds shipped this year

1. Meta put an AI agent inside the WhatsApp Business app for small businesses in India

Since May, eligible small businesses in India can switch on Business AI in the WhatsApp Business app. It answers customers from the business’s own details and catalogue, captures leads and books appointments, and Meta says UPI payments in the chat are coming. The owner can take over any conversation. It lives in the app, not on the Business API.

What to do: whatever answers your customers, a bot like this or your own chatbot on the API, check that each conversation and what came of it lands on the customer’s record. A chat that ends in the app is a lead nobody can learn from.

Source: Meta Newsroom, “Introducing Business AI on WhatsApp for Small Businesses in India”, 14 May 2026, updated 26 May 2026 (checked 11 November 2026)

2. Claude and ChatGPT can now work inside your business tools

Both connect to outside software through MCP. Claude’s custom connectors work on every plan, including free (one connector on Free), and on Team and Enterprise an owner adds a connection once for everyone. ChatGPT’s plugins are built on the same kind of MCP server. In practice the assistant reads and acts in your tools with your login, instead of you pasting exports into a chat window.

What to do: list the tools your loop runs through, from ads to CRM to the place sales get recorded, and ask each vendor whether it has an MCP server and what it can do without a person confirming. Start with read access everywhere. That alone lets the assistant connect the dots.

Source: Anthropic Help Center, “Getting started with custom connectors using remote MCP”; OpenAI developer documentation, plugins and MCP servers (both checked 11 November 2026)

One number: 91%

91% of online adults in India message a business every week.

That’s from a 2025 Kantar study Meta quoted when it launched Business AI here. For Indian brands, chat is where interest turns into a booking or an order, so it’s the richest part of the loop. It’s also the part most often left out of it: conversations sit in a phone or a separate inbox, and the outcome never reaches the record your reports read from.

Source: Kantar, 2025, as cited by Meta Newsroom, 14 May 2026. The case-study figures in that post are self-reported by the businesses quoted.

One playbook: closing the loop in three checks

  1. Do outcomes come back? Pick one real customer from last month and trace them from the ad to the payment. Every gap you hit is a place your assistant is blind. We lost two weeks on demo bookings that showed up in the calendar tool but not where we track leads.
  2. Can your assistant read across the tools? If a vendor’s answer is “use our copilot”, ask how that copilot will see your CRM, your website and your ads.
  3. Does it tell you what it couldn’t see? Ask every week which numbers it couldn’t reach and what it guessed. That list is your map of where the loop is still open.

If you want to try this with CampaignHQ, which we sell: it’s on every plan including free, it works with your permissions, and it can’t launch anything without a person. We’re in ChatGPT’s apps and Claude’s connector directory. Setup takes about five minutes →

Black Friday, 27 November

Ten days out, and the week where a daily loop pays most. Make sure orders come back to the customer record with the campaign they came from. Then each morning of sale week, ask your assistant which offer and which segment actually bought yesterday, and shift the next send before lunch. Get your WhatsApp templates approved this week, so none of it waits on Meta. Then Christmas, and the Meta INR billing deadline on 31 December.

Get the next issue

Is your loop closed yet, and where does it break? I’d like to hear it. The Engagement Desk goes out every other Tuesday. Subscribe here to get each issue in your inbox before it goes up on the blog. If you’d like to talk through any of this for your own team, book a call with me.

Vishal Singhal
Founder, CampaignHQ