33signals ← Back to the site
Show the working

We don't do testimonials. We do evidence.

Anyone can buy a glowing quote. So instead of telling you what our clients say, we'll show you what the research says — every claim on our site, traced to a named, dated, independent source you can check yourself. Where the evidence is strong, we lean on it. Where it's contested, we say so. That second part is the whole point.

Primary sources

McKinsey, MIT, Microsoft, Harvard, Gartner, government data — not our own surveys.

Named & dated

Every figure carries its source and year, so you can verify and so you can see when it ages.

The honest read

Each claim includes its nuance or its limit. We show the counter-evidence too.

Kept current

The research moves fast. So does this page. Stale evidence is noise.

The evidence, section by section — a mirror of the page you came from

01 · The premise

People work for their software. It should be the other way round.

On the site we say: most software makes humans feed it, and the friction degrades the data
~60%
of the knowledge-worker day goes to "work about work" — coordinating, searching, switching apps, chasing status — not the skilled work they were hired for.
Asana, Anatomy of Work Index
~22%
of CRM data decays per year (about 2% a month), rising toward 70% in high-churn sectors, driven by ~11% of professionals changing jobs annually.
Cognism; LinkedIn Economic Graph
~50%
of CRM deployments fail to meet their planned objectives — the analyst range runs from roughly 30% to 70%.
Gartner; Forrester; Johnny Grow, 2025
The 33signals read

The failure is structural, not human. Every system that only stays current when people do extra work will rot, because people optimise for the work, not the system. The fix isn't discipline — it's removing the data entry from the human's path entirely.

The honest read

The CRM-decay and data-cost figures come partly from vendors who sell the cure, so we present them as ranges, not precise truths. The direction is not in dispute; the decimal points are.

Sources
  • Asana, Anatomy of Work Global Index — ~58–60% of time on coordination vs ~26% skilled work, 10,000+ knowledge workers.
  • Gartner (2020) — poor data quality costs the average organization an estimated $12.9M/year (enterprise self-estimate; dated, enterprise-scale).
  • Cognism; LinkedIn Economic Graph — B2B data decay ~2.1%/month; ~10.9% annual job changes.
← Back to this claim on the site
02 · The adoption-value gap

Everyone has the tools. Almost no one has the value.

On the site we say: 88% use AI, ~6% capture value, ~21% have rewired the work — and rewiring is the difference
88%
of organisations now use AI in at least one business function — up from 78% a year earlier.
McKinsey, The State of AI, Nov 2025
~6%
are "high performers" attributing more than 5% of EBIT to AI. Only 39% report any EBIT impact at all.
McKinsey, Nov 2025
#1
Of 25 organisational changes tested, redesigning workflows has the single biggest effect on whether AI shows up in profit. Only ~21% have done it.
McKinsey, March 2025
~95%
of enterprise generative-AI pilots deliver no measurable profit-and-loss impact. The barrier is approach and integration, not the technology.
MIT NANDA, July 2025
The 33signals read

This is the entire case for what we do, proven by the people who'd most like to report otherwise. Adoption is table stakes. The value was never in the tool — it's in untangling the work around it. That is the gap we close, and almost nothing else explains who crosses it.

The honest read

"High performers redesign workflows" is a correlation, not proof of cause — the winners also have better leadership and data. And the MIT figure means "no measurable P&L impact yet," not "failed." We won't oversell either.

Sources
  • McKinsey, The State of AI (Nov 2025) — n=1,993 across 105 nations; 88% adoption, ~6% high performers, 39% any EBIT impact.
  • McKinsey, "How organizations are rewiring to capture value" (March 2025) — workflow redesign the top correlate of EBIT impact; ~21% have redesigned.
  • MIT NANDA, "The GenAI Divide: State of AI in Business" (July 2025) — ~95% of pilots show no measurable P&L impact.
← Back to this claim on the site
03 · How humans actually work

We build for behaviour, not against it.

On the site we say: friction kills updates, the start is the hard part, and the room is your richest data
B=MAP
Behaviour happens only when motivation, ability and a prompt converge. Lowering friction — raising "ability" — is the most reliable lever there is.
Fogg Behavior Model, Stanford
Satisficing
Closed and dropdown fields push people toward low-effort, "good enough" answers — degrading the very data they're meant to capture.
Krosnick, survey methodology
Richest
Decades of research rank face-to-face conversation the highest-bandwidth medium for conveying meaning, nuance and context.
Daft & Lengel, Media Richness Theory
The 33signals read

If a system needs people to do extra work, the behavioural science says they won't — reliably, predictably, forever. So we move the input to where the energy already flows: a reply to an email, a conversation that was happening anyway. Engineer the start to cost almost nothing, and the data arrives on its own.

The honest read

We say conversation is richer, not more honest — and we mean it. Research is clear that for sensitive or embarrassing topics, people actually disclose more in anonymous written formats than face to face. Conversation wins on elaboration and context, not on candour under pressure.

Sources
  • BJ Fogg, Behavior Design Lab, Stanford — Behavior = Motivation × Ability × Prompt; reducing friction is the most dependable lever.
  • Jon Krosnick — closed-format "satisficing" lowers response quality.
  • Daft & Lengel (Management Science, 1986) — Media Richness Theory ranks face-to-face highest.
  • Tourangeau & Yan (2007); Gnambs & Kaspar (2015, meta-analysis, N=125,672) — anonymous written formats elicit more candour on sensitive topics.
← Back to this claim on the site
04 · The meeting drain

Half your week is meetings. Most of it evaporates.

On the site we say: ~11 hours a week in meetings, ~1 in 10 highly productive, and the cost runs into the millions
11+
hours a week the typical knowledge worker spends in meetings — around 28% of the working week. Some studies put it far higher.
Fellow, State of Meetings
~11%
of meetings are rated "highly productive" by the people sitting in them. The rest is where work goes to die.
Atlassian
+252%
increase in weekly time spent in meetings since early 2020 — the overload is recent, and accelerating.
Microsoft Work Trend Index, 2022
The 33signals read

The meeting is the richest, most honest-in-context data your business produces — and you let almost all of it evaporate the moment everyone stands up. We don't give you another summary to drown in. We give you the next person in the room — one that remembers every commitment and keeps the work moving between sessions.

The honest read

The dollar cost of bad meetings is genuinely contested — credible estimates range from about $37B to $399B a year in the US, and per-employee figures are modelled, not measured. We quote the range and let you plug in your own numbers rather than pretend at a precise figure.

Sources
  • Fellow, State of Meetings — ~11.3 hrs/week in meetings (~28% of the workweek).
  • Atlassian — only ~11% of meetings rated highly productive.
  • Microsoft Work Trend Index (2022) — +252% weekly meeting time since Feb 2020.
  • Cost estimates: ~$375B/yr (Fellow's CEO, estimate); $25k/employee/yr (Bloomberg); range $37B–$399B (Atlassian; LSE) — all modelled.
← Back to this claim on the site
05 · The next person in the room

AI does the draft. Humans close the loop.

On the site we say: not a notetaker, but a presence-aware colleague — and human-in-the-loop is the trustworthy model
$1.5B
valuation reached by meeting-AI tool Granola in March 2026 — the category is real, crowded, and maturing fast.
Bloomberg / TechCrunch, 2026
~76%
of executives now treat capable AI more like a coworker to be onboarded and overseen than a tool — human-in-the-loop, by design.
MIT / BCG
Consent
A 2025 class action over recording without all-party consent shows the trust cost of getting the human side wrong.
In re Otter.AI Privacy Litigation, 2025
The 33signals read

The moment a recording bot announces itself, the real conversation stops. Candour, not transcription, is the scarce resource — which is why we design for presence and keep a human closing every loop. For consequential, judgment-heavy work, a person in the loop is what makes the output trustworthy.

The honest read

Human-in-the-loop is not always better. For high-volume, low-stakes, well-defined tasks, full automation wins on speed and consistency. We use the human where the stakes and the judgment are real — not as a reflex.

Sources
  • Bloomberg / TechCrunch (2026) — Granola $125M Series C at a $1.5B valuation; category maturing rapidly.
  • MIT / BCG — ~76% of executives view agentic AI as coworker-like, warranting oversight.
  • In re Otter.AI Privacy Litigation (2025) — consent litigation underscores the trust stakes.
← Back to this claim on the site
06 · The future of work

AI didn't fix the overload. It moved it.

On the site we say: we swapped email overload for summary and notification overload — more intelligence, more noise
275×
a day the average worker is interrupted — once every two minutes during core hours — by meetings, emails and chats.
Microsoft Work Trend Index, 2025
40%
of employees received AI-generated "workslop" in the last month — plausible-looking output with no substance, pushing the real work downstream to a human.
HBR / Stanford & BetterUp, Sept 2025
15.4%
of the content workers receive is now estimated to be workslop — and half think less of the colleague who sent it.
HBR / Stanford, 2025
The 33signals read

More intelligence has not meant less noise — it has meant more. The answer is not another tool that generates more to keep on top of. It's software that updates itself with humans in the loop, so the work compounds instead of cluttering. Signal over noise, made operational.

The honest read

Overload is the default outcome of AI, not the inevitable one. When workflows are genuinely redesigned, AI does deliver real productivity gains — which is exactly why the redesign, not the tool, is the thing worth paying for.

Sources
  • Microsoft Work Trend Index, "Breaking down the infinite workday" (2025) — interrupted every 2 minutes / 275×/day; 31,000 workers across 31 markets.
  • HBR / BetterUp Labs & Stanford Social Media Lab (Sept 2025) — "workslop": 40% received it last month; ~15.4% of received content.
  • HBR (Feb 2026), Ranganathan & Ye, Berkeley Haas — "AI doesn't reduce work, it intensifies it."
← Back to this claim on the site
07 · Advisory-grade, fast

The partner beats the build.

On the site we say: independent diagnosis is worth paying for, and partnering beats going it alone
Buying or partnering for AI reaches successful deployment about twice as often as building it internally — 67% versus 33%.
MIT NANDA, 2025
50–71%
of businesses not adopting AI cite a lack of expertise or skills as the primary barrier — ahead of cost or regulation. Talent is the bottleneck.
OECD; WEF; national surveys
$8–35k
the typical price of a mid-market AI-readiness assessment. Many "free" vendor assessments are sales discovery in audit clothing.
Multiple consultancy sources, 2026
The 33signals read

The bottleneck was never the tools — it's people who can direct them and a clear read on where to start. That's the case for an independent, glass-box diagnostic. The Signal Check gives you the read for free, in minutes, before you've spent a cent or spoken to us.

The honest read

We have an obvious interest in you valuing advisory. So we benchmark you against your peers using independent data, show our working, and tell you when you're already doing well. A diagnostic that finds a crisis in everyone is a horoscope.

Sources
  • MIT NANDA (2025) — externally bought/partnered AI reaches deployment ~67% vs ~33% for internal builds.
  • OECD; WEF; national adoption surveys — skills/expertise the most-cited barrier to adoption.
  • Multiple 2026 consultancy sources — mid-market AI readiness assessments ~$8k–$35k.
← Back to this claim on the site
08 · Where you are

The gap is local, too.

Our markets show the same pattern — high adoption, shallow strategy — in their own numbers
94% / 14%
South African companies using AI for everyday drafting (94%) versus those with an integrated AI strategy (14%). The adoption-value gap, localised.
World Wide Worx, SA GenAI Roadmap 2025
77%
of UK firms using AI saw no immediate change in revenue; only ~12% reported an increase. Adoption without redesign, again.
UK DSIT
~1 yr
how far small businesses now trail large ones on AI adoption — down from decades in past tech cycles. The catch-up window is open and closing.
US SBA Office of Advocacy, 2025
The 33signals read

Wherever we look — South Africa, the UK, the US — the same shape appears: businesses have the tools and lack the rewiring. The opportunity for the ones who move now is a compounding lead, because the advantage builds on itself quarter after quarter.

The honest read

Survey definitions vary wildly — "using AI" ranges from 16% to 54% in the UK alone depending on who's counting. And globally, smaller firms still lag larger ones meaningfully. We read the trend, not any single headline number.

Sources
  • World Wide Worx / Dell / Intel, SA Generative AI Roadmap (2025) — 94% everyday AI use; 14% integrated strategy; 13% governance frameworks.
  • UK DSIT — ~77% of AI-using firms saw no immediate revenue change; ~12% reported increases.
  • US SBA Office of Advocacy (2025) — small-business adoption now ~1 year behind large firms, vs decades historically.
← Back to this claim on the site
The point of all this

The problem is real. So is the way out.

You don't have to take our word for any of it — that's the idea. When you're ready to see where your own business sits, the read takes three minutes.

Take the Signal Check