Five posts a week that answer their own DMs
Finished jobs → written, designed, approved → published, and every comment answered in DM
This business knew it should be posting and mostly wasn't. Someone would remember, put up three things in a week, then go quiet for a month. Hiring for it was hard to justify against everything else the money could do. We built a system that draws on the work the business already does, finished jobs, service list, seasonal patterns, writes and designs the post, puts it in a queue the owner clears from their phone, and publishes on schedule. Then ManyChat handles what the post starts: someone comments, they get a DM, and the useful ones arrive as a name and a number instead of a notification nobody opened. Five posts a week, running at about $115 a month.
What it's worth
What the same output costs to buy
- Posts published per week
- 5
- Posts per month
- ~22
- Social agency retainer at this cadence
- ~$1,800 / mo
- Part-time social manager
- 10 hrs/wk at $30/hr
- Cost of the manager per year
- ~$15,600
- per year, the cheaper of the two
- ~$15,600
What the system costs to run
- Image and video generation
- ~$45 / mo
- Model calls for copy and captions
- ~$25 / mo
- Hosting, scheduling, publishing
- ~$20 / mo
- ManyChat, for the DM automation
- ~$25 / mo
- Running cost per year
- ~$1,380
- per year
- ~$1,380
How this was calculated — every assumption
- This is a cost-of-alternative benchmark, not money taken off an existing bill. The business was not paying an agency before, it was posting sporadically or not at all. So nothing was removed from a budget. The figure says what buying the same output would have cost, which is the only honest way to price a system that replaced nothing.
- Two ways to buy it were priced: an agency retainer at roughly $1,800 a month for this cadence, and a part-time social manager at 10 hours a week at $30 an hour. The cheaper of the two is the one used.
- Running cost is metered, not estimated: generation, model calls, scheduling, and the ManyChat subscription come to about $115 a month at five posts a week. A higher cadence costs more, and ManyChat is priced by contact list size, so a busier DM flow costs more too. Both narrow the gap.
- The comparison is still posting against posting. A social manager or an agency at these rates would not have been answering DMs at all hours either, so the DM automation sits on the benefit side of the ledger and is deliberately not priced into the difference.
- Rounded down from about $14,200.
- Nothing is claimed about reach, followers, or booked work. Conversations started and contact details captured are countable now that the DM flow logs them, but no figure appears here because none has been measured yet. Posting consistently and answering quickly are the deliverables. What they are worth depends on the offer and the local market.
Results at a glance
- Five posts a week, published without anyone remembering to
- About $115 a month to run, against ~$15,600 a year to buy the same output
- Posts built from the business's own jobs and services, not stock filler
- Comments answered in DM within seconds, at any hour
- Qualified conversations arrive as a name, a number, and the thread
- Owner clears a week of approvals from their phone in a few minutes
- Publishing and messaging both go through official APIs, not browser automation
- Nothing posts without a person seeing it first
The problem
Posting was always someone's second job, so it happened in bursts: three posts in a week, then nothing for a month. An empty feed is worse than no feed, because a prospect checking whether a local business is still trading reads the gap as an answer. The obvious fixes both stalled. An agency retainer was hard to justify at the size of the business, and the content an agency produced from a distance did not look like the work. Meanwhile the raw material was already there and going to waste: finished jobs, before-and-afters, the questions customers asked every week. The other half of the problem only showed up on the rare weeks the posting did work. Interest arrived as comments and DMs, at the hours people are on their phones, and got answered whenever someone next opened the app. Usually the following day, sometimes not at all. Somebody who asks how much and hears nothing until tomorrow has already asked someone else.
What was built
Source material
Pulls from what the business already produces: completed jobs and their photos, the service list, common customer questions, and the seasonal pattern of demand. Posts are about this business, not about the industry in general.
Writing and design
Copy and captions written per format, with images generated or built around the job photos. Feed posts, reels, and stories each get their own treatment rather than one asset reposted three ways.
Brand rules
Tone, colours, the services offered, and the claims the business is not allowed to make are all set once and enforced on every generation. Getting this wrong in public is the expensive kind of wrong.
Approval queue
A week of posts lands in one queue the owner clears from their phone. Approve, skip, or send back. Batching it weekly is the difference between a system that gets used and one that gets ignored.
Scheduling and publishing
Approved posts publish through the Instagram Graph API on a set schedule, spread across the week rather than dumped in a batch.
Comment to DM
Posts carry a specific instruction rather than a vague one: comment a keyword and the thing gets sent over. ManyChat watches for it and opens the DM within seconds. A comment is the cheapest action someone can take, which is exactly why it is the one to ask for.
Qualifying in the DM
A short ManyChat flow asks what the job is, roughly where, and roughly when. Three questions, because the fourth is where people stop answering. What comes out is a name, a number, and enough context to price the call.
Handoff
Qualified conversations leave ManyChat by webhook into n8n and land as a real lead with the DM thread attached, rather than sitting in an inbox on someone's phone. Anything the flow cannot answer is passed to a person with the conversation so far.
Logging
Every post, its source material, the comments it drew, and the conversations it started are logged in one place. Source material that has been used is retired so the feed does not circle back to the same job.
What keeps it from breaking
A person approves every post
The system drafts and schedules, the owner approves. Fully automatic publishing saves a few minutes a week and puts the business's name behind text nobody read first. That is the wrong trade at any cadence.
It cannot make claims it can't back
Pricing, guarantees, timescales, and licensing are on a blocklist. A generated post promising something the business does not offer is a complaint, and in some trades a regulatory problem.
Official API, not browser automation
Publishing goes through Meta's Graph API. Automating the app through a browser risks the account itself, and an account ban costs more than every post it would ever have saved.
The same job never posts twice
Used source material is retired. A feed that quietly recycles last quarter's work reads worse than one that posts less often.
The DM flow knows when to stop talking
It answers the question, asks its three, and hands over. It does not try to close anyone. A bot that keeps going after someone has clearly asked for a human is the fastest way to lose a customer you had already won.
Instagram's messaging window is respected
Automated replies stay inside the 24-hour window Meta allows after someone messages first. Reaching back outside it needs a human, not a flow. Working around that rule risks the account, and the account is the whole channel.
How the approach worked
The interesting problem was not generating a post. Anyone can do that in a chat window. It was generating one worth publishing, from this business's own work, at a cadence nobody has to maintain, without handing an unreviewed account to a model. Then it was making the post lead somewhere. Content that nobody can act on is a brochure, and a comment nobody answers until tomorrow is the same missed call the phone system already solved. ManyChat closes that gap: the post asks for a comment, the comment opens a DM, the DM produces a name and a number. Keep the raw material local and specific, keep a person on the approve button and on the end of any real conversation, and let the schedule and the replies run themselves.
The results
The feed posts five times a week now and has stopped going quiet. Content comes from jobs the business actually did, and the owner spends a few minutes a week approving rather than an evening making things. Interest no longer waits for someone to open the app: a comment gets a DM in seconds, at any hour, and the ones worth having arrive as a name, a number, and the thread that produced them. At about $115 a month the whole thing runs for a fraction of what buying the posting alone would cost. How much work it brings in was not measured and is not claimed here.
Three posts in a week, then a month of silence. Stock-looking filler when it did go up, because the good material never made it out of someone's camera roll. Comments and DMs answered the next day, if at all.
Five posts a week on schedule, built from real jobs, approved from a phone in a few minutes. Every comment answered in DM within seconds, qualified, and handed over as a lead.
Tools & technologies
Have a process like this slowing your business down? Tell me your stack and goals and I'll send back a build plan.
Take the stack assessment →