Where this is heading: five dated bets for 2027

Today the WhatsApp agent at El mejor DOC turns 9.4% of the people who ask for an appointment into a booked one. That is the baseline for one of the five bets I am signing here for October 31, 2027. The other four also come with today’s number, so that a year from now anyone can check whether I was right.

Predictions about artificial intelligence are everywhere and almost none of them can be called in. They say “AI will transform business” and never say how much, by when or how it will be measured. This piece is the opposite: five bets about what will happen inside btodigital, the agency I have run in Medellín, Colombia, since 2013, each with its baseline and its target. If I miss, it will show. That is the point.

It is the second piece in a series on the future of work. The first told what the change looks like from inside a small business; this one says where I think it is going.

What are the five bets?

Bet	Baseline (October 2026)	Target for October 2027	How it is measured
1. The monthly report that builds itself	1 client	30 clients	Clients receiving a system-generated monthly report approved by a person
2. Conversation replaces the form	9.4% booked among those who ask	15%	Confirmed appointments over conversations with intent to book
3. Ten people certified on Claude	0 of 10	10 of 10	Active official certifications on the team
4. The team grows in product, not operations	22 operations, 3 sales, 3 administration	Operations at 22 or fewer and 3 new people in product	Payroll by area
5. Every client with its own memory	Pilot	Every client on a monthly contract	Clients with an active workspace consulted at least once a month

Why bet that the monthly report will build itself?

Because it has already happened once. The monthly audience report for a regional news outlet now generates itself: the system pulls analytics, search and site data, strips out automated visits and assembles the document. But it does not go out on its own. It stays a draft until someone on the team reviews it and approves it with one click.

That is the shape I think reporting work will take at every agency: the machine assembles, a person answers for it. I explained how it works in client reports that build themselves. The bet is that within a year this setup covers a good share of our clients, not just one.

Baseline: 1 client. Target: 30 clients.

Why will conversation replace the form?

Because a conversation carries something a form never had: the person’s real question at the moment they have it.

Between September 1 and October 2, 2026, we audited 886 conversations handled by the WhatsApp agent of El mejor DOC, our physician directory, which runs on Atendio. Of those, 736 asked for an appointment or consultation, and about 69 ended in a confirmed appointment: 9.4%. The full audit is in what we learned from 886 conversations of a WhatsApp agent for medical practices.

The finding that taught me most was inside. When someone on the team offered a specific day and time, 34% booked (64 of 189). When only the price was given, with no time, 0.8% booked (2 of 264). No form lets you see that difference, because a form does not talk back.

Baseline: 9.4% booked among those who ask. Target: 15%.

Why ten certified people?

Because the judgment to use these tools well cannot live in a single person, and today it lives in me. As of October 2026, nobody on the team holds an official Claude certification. Zero of ten.

Certification is not the goal. It is a way to force the knowledge to be spread across the team instead of sitting in the founder’s head.

Baseline: 0 of 10. Target: 10 of 10.

Why will the team grow in product and not in operations?

Today we are 28 people in Colombia: 22 in operations, 3 in sales and 3 in administration. There is no product area on our payroll, even though we run our own products such as Atendio, El mejor DOC and the Cerebro IA de btodigital.

The bet is that our next round of growth will not come from adding people who do by hand what a tool can do, but from adding people who build and improve those tools. If by October 2027 operations grew in headcount and product did not, I lost this bet.

Baseline: 22, 3 and 3. Target: operations at 22 people or fewer and 3 new people in product.

Why will every client have its own memory?

Because an AI that does not remember its client is a new intern every morning. The Cerebro IA de btodigital, our client knowledge system, gives each client an isolated memory: their website, documents, audio and brand voice. Before answering, it searches there.

The first use that truly worked was answering a question companies usually try to answer with surveys: what objection do people have when they see a client’s ads and do not buy? The answer came from reading the real comments people left on its posts, grouped and with each one quoted.

Today it is a pilot. The bet is that it stops being one.

Baseline: pilot. Target: every client on a monthly contract, with an active workspace consulted at least once a month.

What we learned

Preparing these bets also left lessons worth signing alongside them.

The baseline was measured before we fixed the agent. The same audit found the system read “good evening” as a goodbye: 22 of 55 automatic closures came right after a greeting. We fixed it on October 3, 2026, and the 9.4% baseline was measured before that fix.

Without a person, nobody booked. Among conversations with an appointment request where nobody from the team ever stepped in, 0 of 127 ended in an appointment. That does not sink bet 2, but it qualifies it: conversation replaces the form, not the person.

Creating workspaces is not the same as clients using them. That is why bet 5 is measured by use, not by workspaces opened.

Certifications are at zero because nobody started. It is not a lack of ability. We never gave it a date or an owner, which is the second condition I demand of any client.

How do you write your own bets?

If you run a company or a department, writing your own five bets will serve you better than reading mine. The recipe:

1. A sentence you can lose. “AI will make us more efficient” cannot be lost. “In a year, 15% of appointments will come through conversation” can.
2. A baseline measured today. If you do not have one, your first task is to measure it. Without a baseline there is no bet, only a wish.
3. A target with a number and a date. The number can be ambitious. The date cannot be “soon.”
4. How it will be measured, written down in advance. Who pulls the data, from which system, with which definition.
5. An owner. Someone with a name who answers for the bet.
6. Publish it, even if only internally. A bet nobody knows about commits you to nothing.

And one rule for choosing them: bet where you have your own data. Bets about what the market will do are opinions; bets about your own operation can be called in.

What do I not think will happen in 2027?

I do not think a small business will run without people. The 127 conversations with no person and not a single appointment are my best argument.

I do not think the form will disappear. Some people will prefer not to chat, and that is fine. What changes is which door is the main one.

I do not think the advantage lies in this season’s model. Models get cheaper and more alike every quarter. What cannot be copied is your own data and the judgment to use it. When an AI project fails, it is almost never because of the technology.

I do not think I will win all of these bets. If I go five for five, I bet too low.

Frequently asked questions

Why publish bets with targets instead of predictions? Because a prediction without a number cannot be checked. A bet with a baseline, target and date can. A year from now I will publish the result of each one, right or wrong.

Where does the 9.4% booking rate come from? From an audit of 886 conversations handled by the El mejor DOC WhatsApp agent between September 1 and October 2, 2026. Of those, 736 asked for an appointment or consultation and about 69 ended in a confirmed appointment.

What happens if a bet fails? It gets reported the same way. A failed bet with its explanation teaches more than one won by luck. That is why each one already has its measurement method written down.

How do I choose what to bet on in my company? Pick processes where you already have your own data and a number for today. If you cannot measure the baseline, your first bet is to start measuring it.

Do these bets only apply to agencies? No. The method works for any company: a sentence you can lose, a baseline, a dated target, a measurement method written in advance and an owner.