A Small Agency Backup Plan That Survives a Dead Laptop
A small agency backup plan, an encrypted vault and a restore test, plus cold outreach demos, a CRM rebuild and kinetic video ads from one Nairobi Friday.
Friday started with client pages and ended with me building an encrypted vault close to midnight. The real lesson was that I did not have a small agency backup plan, I had a laptop with everything on it. One hardware failure and Dullu Digital would have been a nice memory.
Client pages first, then a bigger question about positioning
I opened the day on a retail client's site, working from two screenshots of his feedback. Added the pages he asked for, checked they matched the brand instead of looking bolted on, and deployed. That part was quick.
Then the harder question landed. He wanted a standalone page for luxury stays, browsable per location, with images, descriptions and price. That is not a page. That is a product. I went back and looked at the whole site again, properly, and realised I had been building a brochure while the client was describing a premium mobility and concierge platform. Most of what is on there now is test data, just enough to create the final impression so he can see the shape of it.
So I stopped adding pages and started phase one of the actual structure. Positioning before polish. A beautiful brochure that cannot scale is a rebuild you pay for twice.
Cold outreach demos and the question of what to charge
Earlier in the week I built a full demo site for a dental clinic prospect and sent it cold. The owner liked the pitch and said he is coming to see me in person. Great. Then I spent twenty minutes on the question I always underprice: what do I ask for?
Building the thing before anyone pays is the easy part. Naming the number is where I lose shillings. I settled on pricing the build quality and the outcome, not my hours, because the hours are getting shorter every month and I refuse to be punished for that.
Another prospect I had been warming up went flat. No real interest. Rather than keep knocking, I pulled a fresh list using signals instead of hope:
- Real estate businesses with strong Google reviews and no website at all
- Tour and excursion operators running their whole business through WhatsApp
- Clinics with a decent reputation and nothing searchable
Reviews are the signal I trust most. A business sitting on hundreds of good reviews with no site is leaving money on the table, and they already know it, which makes the conversation short.
Mid-afternoon I dropped everything to build two visual samples for a boat excursion prospect, working from a zip of their own photos they had sent through. I pulled their public Google reviews into the design because their proof is better than my copy. A coastal hospitality prospect came in asking for detail too, so I spent time studying what sites in that category actually do: how many pages, what composition, what the good ones get right. Unpaid research, but it is the kind that makes the next pitch faster.
The backup plan: encrypted vault, offsite push, restore test
Around lunch I asked a plain question. What on my system is running from my local machine? The answer was more than I was comfortable with.
I do not want to be dependent on one AI vendor, and I do not want to be dependent on this laptop either. Those are the same risk wearing different clothes. Cloudflare is my middle ground, because if this machine dies tonight the business should not notice.
Before deleting a dead project I reviewed the cron tasks first, which saved me from killing something that was still live. Then I did the whole protection job in one sitting:
- Built an encrypted vault with a passphrase, kept off the laptop itself
- Exported the production database (code without data is a shell, I asked why the export was happening and that was the right answer)
- Ran the offsite push so a copy lives somewhere that is not my desk
- Ran an actual restore test, because a backup you have never restored is a rumour
- Noted to copy the vault to a USB stick and a second cloud
One sting. I could not log into GitHub because I lost my authenticator account and I am waiting on recovery. So part of my code protection plan was sitting behind an account I was locked out of. That is exactly the kind of single point of failure this whole exercise was meant to find.
Rebuilding the CRM around costs, not just contacts
I checked whether my admin prospects page was reading from the database or serving me stale cache. That turned into a bigger decision: move admin off a path and onto its own CRM subdomain, rebuild the interface properly in the style of a good open source CRM, and delete the old abandoned one instead of letting it rot.
I also noticed I was manually refreshing to get new prospects. That is a cron job, not a human job.
Then the part that mattered most. I added a financials page: fixed costs, variable costs, customer acquisition cost, targets. Real numbers, including the AI subscription and sales tooling, then a scaling timeline showing what I need to spend on to hit the targets I set. Seeing cost per acquired client next to monthly burn changes which outreach I bother doing.
Late on I looked at integrating the local tax invoicing system and asked how secure that connection actually is. If I am sending invoice data to a government endpoint, I want to know what is in transit. I also made myself define prospects versus leads properly. Sounds basic. If the CRM blurs them, every pipeline number I report to myself is a lie.
Kinetic typography ads instead of putting my face on camera
I checked how DDi is listed on Google. It needs a cover photo, the logo and real photos of the business, and old search results are still describing an earlier version of what I do.
For the campaign I wanted kinetic typography, the text-driven spots the big telcos and ride apps run. The first render was completely wrong. Two of the kinetic cuts were right. From there I built a campaign mind map, then rejected my own version as too rigid, asked for genuinely guerilla thinking, then asked for structure again. Changed my mind three times inside an hour. I also looked at cloning my voice for the commercials, checked a tool, checked alternatives, and dropped the whole idea. I am not going on camera, so animation and stick-figure entertainment do the talking.
What I would tell another founder
Do the restore test. Not the backup, the restore, because the backup is the part that makes you feel safe and the restore is the part that tells you if you are. And when you audit your own setup, follow every dependency to the end: mine led straight to an account I could not log into. Protect the work before you polish it.