Showing my working

On 11th September I rode into Antelope Wells, on the Mexican border, and finished the Great Divide Mountain Bike Route: 2,700 miles from Banff, down the spine of the Rockies.

I had the idea four years ago. I set off last summer, and a fall ended that attempt after 42 days. This August I went back for the rest. My bike isn't home yet, but that's another story; you can see progress at robinridesthedivide.com, and in the films.

Proofpoints has taken nearly as long. The idea is two and a half years old, and this is the third time I have built it. Two long pieces of unfinished business have come good in the same month! I'm keen to see how Proofpoints is received.

So: Proofpoints is open for its first users. Any client-side advocacy pro can sign up today with no obligation and use it on real stories.

Last issue I argued that a customer story is no longer trusted just because it exists; it needs the working shown underneath it. I feel the same way about launching Proofpoints, so here is mine.

01  What Proofpoints does

Proofpoints is a platform for the people who create customer stories as part of their advocacy programs, and for the agencies and freelancers who work alongside them.

You put an interview in. Three distinct drafts come out, each scored against a brand standard you define: your terminology and your voice rules. You or your agency edit the best one, with an assistant and coaching on hand, and the integrated workflow tracks reviews and approvals; who is doing what, and what is waiting on whom.

The customer approves once, and that approval is recorded as an event with a date rather than buried in an email thread. The story then goes to a permanent public page in seconds (if you need it), carrying the named advocate, the approval, and structured markup so an AI assistant can find it and cite it accurately. From the same approved story you can produce shorter takes: executive summaries, press releases, translations, as well as slides, socials etc.

It runs on credits. No seats, no annual contract, no implementation project with IT. New, qualifying accounts get 1,000 free credits, enough for four full customer stories. It is hosted securely, and your material is never used to train models.

02  Three things I got wrong

I thought the models were largely interchangeable.
Every story still goes through three different language models, but not for the reason I started with. I assumed more generation meant better stories, and that one model was much like another.

Each turns out to have its own personality. GPT can hardly make a point without first telling you what something isn't. Gemini turns everything up: things are "incredibly" important and why use one word when ten will suffice? Claude is the most concise and the most cautious; it hedges, and it is oddly fond of the word "genuine".

Learning those personalities, and matching them to a brand's voice, has become the most interesting part of the whole build. Scoring the three drafts against your brand standard is how you find out which match worked.

I built it twice before I built it properly.
Versions one and two were Quickbase nailed to WordPress, then twelve WordPress plugins, with Zapier passing work between them. Each taught me something, and I threw both away. This version started from scratch.

The result is enterprise-ready, and as elegantly simple as I believe a powerful multifunctional platform can be.

I underestimated brand context.
I assumed a good transcript and a sensible prompt would get most of the way. What they get you is a story that sounds like everyone else's. A draft starts to sound like your company when the model works from your own material: your terminology, your voice rules, the phrases you have banned. I should have started there.

03  What held up

The interview is still the constraint.
Everything downstream inherits one conversation, and no amount of processing rescues thin material. I suspected this and building the platform has made me sure of it. The same engine will turn a good interview into something excellent and a survey response into something passable.

Coordination is where the money goes.
In my experience across thirty-five programs over twenty-five years, the pattern has barely moved: six to twelve weeks per story, one story in five never published, and very little of that time spent writing. Shortening the approval loop is worth more to a program than any improvement in drafting.

The AI-to-Human blend is not a ratio.
Nobody needs a policy that says a story must be seventy percent human. What helps is a system that treats a social card and a flagship transformation story as different jobs, deserving different amounts of human attention.

04  What Proofpoints does not do

It does not check your customers' numbers.
We confirm that the advocate is real, holds the role, and approved the exact words. Whether their 40% is right is theirs to stand behind, not ours to confirm.

It does not make thin material good.
Give it a weak source and it will produce something competent, quickly and confidently; that is the failure the whole industry faces. Deciding which customers deserve a proper interview is still your call.

It does not replace the editor.
My view on this has not changed, no matter how good the LLMs get. The platform scores drafts, and coaches you from your brand rulebook. It does not notice that the best line in the interview was the throwaway one at the end. That is still a person's job, yours or your agency's.

What every published story guarantees, and what it does not, is at proofpoints.com/standards.

05  Founding Originals

For twenty-five years I have told companies that the best proof comes from customers treated as partners, not sources. Now I have to do it myself, because Proofpoints does not have any customers yet.

So I am looking for the first few. The Founding Originals will be practitioners running real programs, who put real stories through the platform and talk about what happens, good or bad. I would judge the offer the way I have always judged an advocacy program, on whether it gives people something worth having:

1. ·Status: Your stories among the first published on the platform, and your name here when you are ready.
2. ·Access: A direct line to me, and a real say in what I build next.
3. ·Networking: Being part of a small group of fellow Originals, exploring this space
4. ·Education: The opportunity to learn everything I have amassed in 25 years in this space, and from over two years of working with AI, plus your fellow Originals' experience too.

06  What comes next

Proof Positive carries on as before, fortnightly, on the same subject. For six issues I said what I believe to be true. From here I can report what actually happens when people use the thing, including where it contradicts something I wrote previously.

Next issue: the long tail. Most customer stories never get told, and the reason is almost never editorial; it is budget, bandwidth, and a belief that a story about a smaller customer is not worth the effort. AI is redefining what 'effort' is; the old methodology is wrong, and expensively so.

Subscribe if you have not. Forward to one person who has a hundred customers worth writing about and time for only four.

§

Robin Hamilton is the founder of Proofpoints, the customer storytelling platform. Previously, cofounder and CEO of inevidence. Proof Positive is published fortnightly.

Two weeks from now, in your inbox. Fortnightly, no spam, easy unsubscribe.