At GMT, we build Social Media and SaaS / HR solutions that put your audience and your people on data you own, with AI on top that shows its evidence and waits for a person to approve.
Social media rents you an audience. A platform gives you one.
Social is where people find you. Your own platform is where you keep them. We turn followers into named, consenting members you can reach again, with personalised email, member tiers, merch and events, and sponsors priced on a described audience.
3.2%of followers reached by the average Instagram brand postSocialinsider, Aug 2026
−14%fall in Instagram reach rate in a yearSocialinsider, Aug 2026
23.5%of marketers call organic social impactfulHubSpot, 2026
Your people data sits in six systems: HRIS, payroll, hiring, time tracking, surveys and exit interviews. We join them into one secure database, so you can spot the drift while it's early, see why, and approve the next step with the evidence beside it.
22%of managers are engaged, down from 31% in 2022Gallup, 2022–2025
72%of firms piloting AI agents lack unified dataDeloitte, 2026
36%of SaaS licences go unused on averageZylo 2026, vendor data
Two builds, one from each practice. Client names are withheld while engagements are live; both can be walked through privately on request.
SaaS / HR · Shift-based staffing company
A 400-person event-staffing firm running its people on six systems and a spreadsheet
Security, ushers, concessions and medics working shifts across several states. HR records, payroll, hiring, time tracking, surveys and exit interviews each lived in their own tool. Overtime spikes and pay-rule mistakes surfaced only after payroll had run, and nobody could say in one sentence how each team was doing.
What was built
One people database joining every source, pay rules as configuration with each change tested on real shifts before approval, dashboards for HR, finance and managers, agents that flag attrition risk, overtime and expiring certifications and wait for a named approver, and an employee portal that installs on a phone.
Why it matters commercially
Overtime and turnover become numbers a manager can act on this week, and no payroll change goes live untested.
Concept build with synthetic sample data, prepared as a demonstration. Figures illustrative.
A podcast network with roughly 1.5 million subscribers and no list of its own
Distribution across a major streaming service, a large video channel and a weekly live show, with every one of those platforms keeping the viewer data. The company could not name a single member of its own audience, and went into every renewal and sponsor conversation without evidence of who was actually watching.
What was built
A fan database with a weekly vote as the registration ask, personalised email, member tiers, a merch layer over the existing store, an admin desk and three scheduled reports.
Every project is quoted on its own scope, so there is no price list here. What is consistent is the shape, and it is worth knowing before you get in touch.
Starting point
A short discovery
About a week of scoping: what you have today, what you can see of it, and where a build would pay. Both sides can judge the work before committing.
The core build
Roughly six to eight weeks
Fixed scope and fixed fee, delivered in stages that go live as they are finished rather than in one launch at the end.
After launch
A monthly retainer, optional
We operate, report and keep building while it is useful. Month to month, cancellable, and structured so that leaving is straightforward.
Always
You own the asset
Every account, domain, dataset and line of code is registered to your company from day one. There is no scenario in which ending the relationship costs you the platform.
About Us
GMT builds the platforms underneath social media and HR: one database you own, with AI on top that shows its evidence and asks before it acts. Small, senior, and built to hand over.
Social Media
Social media rents you an audience. A platform gives you one.
Keep posting. Social is where people find you. We build the layer underneath it, where those people become named, consenting members you can reach again on your own terms.
Reel 48 · Monday morning briefingYour Monday morning, handled.GMT’s Voice Agent walks a founder through her audience and gets four approvals in one conversation.Premium solution. The Voice Agent briefing is part of GMT’s premium tier. It is configured and built for each premium client, on that client’s own data and approval rules. Names and figures shown are illustrative.
The reach problem
Most of your followers never see the post.
Platforms decide who sees what you publish. The share of followers reached keeps shrinking, and the platforms keep the data on who those people are.
3.2%
of followers reached by the average Instagram brand post.
Socialinsider, Aug 2026
−14%
fall in Instagram's average reach rate in a year.
Socialinsider, Aug 2026
0.2%
reach rate for a Facebook post that carries a link.
Socialinsider, Jan 2025–Aug 2026
23.5%
of marketers say organic social is impactful, though 40.3% use it.
HubSpot, 2026
Reel 09 · The reach math
100,000 followers. A few thousand see it.
Follow one post from a brand account with 100,000 followers. A few thousand see it, a few hundred engage, and tomorrow it starts again from zero. The bigger the account gets, the smaller the share each post reaches.
Send the same message to an email list you own and it reaches all 100,000.
Figures on screen: Socialinsider, Jan 2025–Aug 2026. Example account is illustrative.
Social vs owned
They do different jobs. You need both.
One is built for reach. The other is built for knowing, reaching and serving the same people again.
On social media
On your own platform
Who owns the relationship
The platform, under its terms.
You. The record, the consent and the contact details sit with your company.
Who sees what you publish
Whoever the algorithm picks today.
Everyone who opted in.
What you know about them
Anonymous totals: views, likes, broad age bands.
Named, located, consenting records with history and interests.
Reaching them again
Post again and hope, or pay to boost.
Send directly, any time, at no extra cost per person.
Revenue
Payouts and brand deals priced on views.
Membership, merch, tickets and sponsors priced on a described audience.
Risk
Rule changes and reach drops arrive without warning.
A list you own stays reachable whatever happens elsewhere.
Social
The front door
Reach and discovery. People find you where they already spend their time.
The step across
One reason to join
A vote, early access, a list or a ticket. A small action that turns a follower into a record.
Your platform
Where it compounds
Named people you can reach again. Every action writes back to the same record.
Exhibit · Instagram brand accounts
The bigger the following, the smaller the share that sees each post.
Average reach rate by account size, % of followers
1K–5K6.65%
5K–10K5.75%
10K–50K5.50%
50K–100K4.50%
100K–1M3.50%
0%2%4%6%8%
Source: Socialinsider, 872,075 Facebook and Instagram brand posts, Jan 2025–Aug 2026, published 3 Sep 2026. Smaller accounts reach a larger share, but even the best band reaches fewer than 7 in 100 followers.
Video concepts
The ideas, one reel at a time.
Short reels from our series. Tap any one to play it here.
Explore each part
Where we work across social.
Pick an area to see what we build and the numbers behind it.
Build the audience you keep.
Tell us where your audience lives today and what you'd do if you could reach them directly. We'll come back with how we'd approach it.
HRIS, payroll, hiring, time tracking, surveys and exit interviews joined into one secure database. Agents spot the drift while it's early, show you why, and wait for a person to approve the next step.
Reel 47 · Monday morning briefingYour Monday morning, handled.GMT’s Voice Agent walks an executive through his week and gets four approvals in one conversation.Premium solution. The Voice Agent briefing is part of GMT’s premium tier. It is configured and built for each premium client, on that client’s own data and approval rules. Names and figures shown are illustrative.
The problem, in four numbers
Managers carry the team. Most can't see it.
The research points the same way across firms: engagement is decided at manager level, managers are losing ground, and the data they need is split across tools that don't talk.
22%
of managers are engaged, down from 31% in 2022.
Gallup, 2022–2025
72%
of US firms piloting agentic AI lack unified, accessible data.
Deloitte, Apr–Jun 2026
36%
of SaaS licences go unused on average.
Zylo 2026 · vendor data
$4.99M
average cost of a data breach, a record, up 12%.
IBM / Ponemon, 2026
Problem≥70%
of the variance in team engagement comes from the manager.
Only 8% of HR leaders say their managers have the skills to use AI well.
Gallup · Gartner, Jul 2025
Cost$10T
in lost productivity from low engagement last year.
About 9% of global GDP. On the frontline, 76% reported burnout in 2025.
Gallup 2025 · UKG, vendor research
Root cause30%
pull people data from five or more systems.
40% have no unified data warehouse. Only 6% of organisations have a unified view across systems.
HR.com, Aug–Oct 2025 · Accenture 2026
Exhibit
The tools feel productive. The results haven't arrived.
Managers engaged, %
202231%
202522%
Best-practice orgs79%
0255075100
Respondents saying AI…, %
Improved my productivity80%
Had any EBIT impact37%
0255075100
Sources: Gallup, State of the Global Workplace 2026. McKinsey & Company, The State of AI 2026, fieldwork 4 May–8 Jun 2026.
The fair objection
"Using worker data with AI pushes people out."
60% of airline, hospitality and restaurant workers say turnover rose because their employer used worker data with AI and other technology. Deloitte, date not stated.
How we answer it
Built for decisions, not surveillance: insights aggregate by team.
No view shows a group smaller than five people.
Role-based access, and every change is logged.
Agents propose. A named person approves.
The fix, mapped
Every problem gets an answer.
Each finding from the research, paired with what the GMT system does about it.
What the research showsWhat GMT does about it
Managers drive at least 70% of team engagement, and only 22% of them are engaged.Gallup
→
People analytics, one screen
Headcount, turnover, attrition by team and the hiring funnel in one view a manager can read in a minute.
30% pull people data from five or more systems; 40% have no unified warehouse.HR.com, Aug–Oct 2025
→
Connect and organise
HRIS, payroll, ATS, time, performance, surveys and exits in one secure, role-based database.
Only 34% rate their ability to act on people analytics as good.HR.com, Aug–Oct 2025
→
One clear call
A recommended decision with evidence beside it, approval by the HR lead and COO, and a 90-day outcome check.
88% of HR leaders say AI tools haven't delivered significant business value yet.Gartner, Jul 2025
→
Agents that wait for a yes
Nine scheduled agents scan the data, show their evidence and route by severity: SMS, daily digest or in-app.
49% of hourly managers still fill open shifts by calling or texting one person at a time.Legion Technologies, 2026 · vendor research
→
Workflows that run
Shift coverage, timesheets, pay-period close, onboarding and approvals as tracked steps with owners.
The average breach now costs $4.99M.IBM / Ponemon 2026
→
Privacy by design
Team-level aggregation, a minimum group of five, role-based access and a full audit trail.
The system
Five layers, nine agents, one yes.
Select a layer to see what it does. Screens use an illustrative 400-person, shift-based staffing company.
Built for shift-based teams
The people hardest to reach are the ones it matters most for.
Only 17% of on-site employees whose jobs can't be done remotely are engaged, against 30% of fully remote workers. Overtime, wage-and-hour rules and payroll changes cost real money every week in these teams.
GMT tests every pay-rule change against real shifts before anyone approves it, and flags overtime drift by team.
Gallup, Jan–Dec 2025
Video concepts
Watch the problem, and the fix.
Our SaaS / HR reel series. Every figure on screen is sourced.
Why it pays
Organisations that fix the foundations pull ahead.
4.5 pts
Estimated EBIT margin advantage for data-ready organisations over their peers.
Accenture, 2026
40%
Lower turnover, linked to the strong employee experiences that "Talent Reinventors" are 6.1 times more likely to create.
Accenture · a correlation, not proof of cause
1.6×
Human-centred AI strategies are 1.6 times as likely to beat expected returns as technology-first ones.
Deloitte, 2026 report
Trust
Built for decisions. Not surveillance.
Team-level only
Insights aggregate by team, with a minimum group of five people.
Role-based access
People see what their role needs, and nothing else.
Every change logged
A full audit trail of who saw, changed and approved what.
Yours to keep
The database, accounts and code are registered to your company.
See your whole team in one view.
Tell us which systems hold your people data today and the question you can't answer yet. We'll come back with how we'd approach it.
Four phases, in order, each one earning the next. We fix the foundation before the clever parts, build early and small on live data, and every AI action waits for a person's yes. You keep everything produced at every stage.
What the data says is broken, then each phase in turn: what we do, what you leave with, and the decision that opens the next. It plays as you scroll; pick your practice, or a chapter to jump.
Sources on screen in every exhibit. GMT pages are illustrative.
Exhibit 1 · Engagement modelChoose a practice
Four phases, each one earning the next.
What happens in each phase, what you leave with, and the decision that opens the next one
Phases one and two are where every engagement starts. Nothing in phase three or four is committed to before the foundation is live and producing data worth acting on.Exhibit 1
Exhibit 2 · Gantt workplanIllustrative
A live foundation in eight weeks, the loop on by twelve.
Example workplan, by week
Illustrative Follows GMT's four-phase model. Phase four, grow from evidence, runs month to month after week 12. Not a client plan.Exhibit 2
Exhibit 3 · Decision rights
Everyone knows who decides what.
Who is responsible (R), accountable (A), consulted (C) and informed (I) for the decisions that matter
Decision
Executive sponsor
Your project lead
GMT
Which gap to close first
A
C
R
Data model and what gets captured
I
A
R
Go-live of the foundation
A
R
C
What an AI agent may do on its own
A
R
C
Approving each consequential AI action
I
A
I
Moving to the next phase
A
C
R
Agreed in week one and kept in the project charter. Approval of AI actions always sits with a named person on your side.Exhibit 3
Exhibit 4 · Value driver treeSocial Media
We agree how success is measured before we build.
One outcome, the drivers that move it, and the leading indicators we report every week. Switch practice in Exhibit 1.
Illustrative Targets are set per client in phase one, from your own baseline.Exhibit 4
Exhibit 5 · Harvey balls
How the GMT method compares.
Three common ways to do this work, scored 0–4 on what clients tell us matters
Big-bang programme
Agency retainer
The GMT method
Illustrative GMT assessment of typical engagement models; not a review of any named firm.Exhibit 5
Operating principles
Five rules we don't break.
They are written into every statement of work, so you can hold us to them.
The foundation comes first
Data joined into one record before dashboards, agents or campaigns. Clever parts on scattered data are demos.
Build early and small
Live behaviour beats a long discovery. Data doesn't need to be perfect to start paying back.
AI proposes, people approve
Anything that spends money, contacts a person or can't be undone waits for a named yes.
Evidence sets the roadmap
What the first months show decides the next build, not a plan written before launch.
You own the asset
Every account, domain, dataset and line of code is yours from day one. Leaving is straightforward.
Start with phase one.
A short, fixed-scope look at where the value is leaking and what closing the gap is worth to you, in money. You keep the findings.
Software that does the work, not just the writing.
A chatbot answers a question. An agent is given an objective, works out the steps, uses the systems it has been granted access to, checks its own output, and comes back with the job done or a clear reason it stopped. The difference matters because one saves you typing and the other takes a process off your desk.
24/7processes that run whether or not anyone is at a desk
Minutesfor work that used to wait for the right person to be free
Every runlogged, reviewable and reversible
Your datastays in your systems, under your access rules
Running right now
Every run, visible as it happens
This is what our agent console looks like on a normal morning. Agents pick up work on schedule, show the evidence for what they found, and stop at the gate when a person needs to say yes.
Done on its ownWaiting for a yesFlagged to a person
Illustrative feed · example data, not a client
Agent consoleLive · Tue 8:14 AM
1,284runs this week
97.6%passed quality checks
3waiting for approval
The market, in four numbers
Most companies are piloting agents they can't yet govern
The gap isn't the model. It's data, controls and managers who know how to work with agents. That's the part we build.
70%
of organisations piloting agentic AI don't feel they can trust and govern their agents.
Deloitte, Apr–Jun 2026
40%
of large organisations are scaling AI agents, up from 27%. Smaller ones are flat at 22%.
McKinsey, May–Jun 2026
4%
of organisations use AI agents in HR, a figure expected to triple within twelve months.
Sapient Insights, 2025–26
8%
of HR leaders believe their managers have the skills to use AI effectively.
Gartner, Jul 2025
The roster
Agents we deploy, by practice
Each one has a job description: what it runs on, how often, the most it may do on its own, and the gate where it stops for a person.
Rung refers to the autonomy ladder further down this page. Every agent starts at Draft and moves up only when its run history supports it.
Built for the rules that are coming
Ready for regulators, not just for demos
Rules on automated decisions about people are tightening. Our controls are designed with them in view, so an audit is a report you print rather than a project you start.
Now
Automated decisions under GDPR
Limits on decisions made solely by automation already apply in the EU. Our agents propose; a person decides.
1 Jan 2027
Colorado employment decision law
A narrower automated decision-making law covering employment decisions takes effect, replacing the 2024 AI Act.
2 Dec 2027
EU AI Act, HR systems
High-risk duties for standalone recruitment and HR systems move to this date under the Digital Omnibus.
Today
Enforcement gaps close
A New York State audit found the city flagged 1 compliance issue across 32 companies where auditors found at least 17.
Sources: Cuatrecasas (EU AI Act timing); Davis Wright Tremaine (Colorado SB26-189, signed 14 May 2026); Office of the New York State Comptroller (audit period July 2023 to June 2025). This is context, not legal advice.
How they actually work
The loop, not the answer
Older automation followed a fixed script and broke the moment reality differed from it. An agent runs a loop instead: understand the goal, decide the next step, take it, look at what came back, and adjust. That loop is what lets it handle the messy middle of a real process rather than only the happy path.
The agent loop
Repeats until the objective is met or a stop condition is hit.
What it is allowed to touch
Capability is granted deliberately, one connection at a time. Nothing is assumed.
The decision that matters
How much rope to give it
Almost every failed agent project skipped a rung. The right level is chosen per task, not per company, and it moves up only once the logs show the level below it was reliable. Most valuable work in a business sits on the middle two rungs and stays there quite happily.
01
AssistPerson does the work
It retrieves, summarises and suggests. A person decides everything. Useful, low risk, and rarely worth building a project around on its own.
02
DraftPerson edits and sends
It produces the finished artefact: the reply, the report, the weekly email, the campaign recap, already filled with the right figures. A person adds judgement and approves. This is where most of the hours are recovered.
03
Act with approvalPerson holds the gate
It carries out the whole process and pauses at the one step that carries consequence, waiting for a yes. Everything either side of that gate is hands-off.
04
Act within limitsPerson sets the rules, reviews the log
Fully autonomous inside a boundary you define: which actions, what value, how often, and when to escalate. Suited to high-volume, low-variance work that is cheap to reverse.
05
Coordinate other agentsPerson owns the outcome
One agent breaks a large objective into parts and hands them to specialised ones. Powerful, and the level where cost and failure modes stop being obvious. We reach for it last, not first.
Levels are chosen task by task. A business commonly runs several at once, and moving a task up a rung is a decision backed by its own run history.
Where it pays
Not everything should be an agent
The honest filter is two questions: how often does this happen, and how much judgement does it need. Frequent work with a clear definition of done is where agents earn their keep. Rare, high-judgement work stays with people, and paying to automate it is how these projects lose money.
The fit map
Illustrative. Each dot is a type of task, placed by how often it recurs and how much judgement it carries.
Where the hours go, before and after
Illustrative shape of a typical week on one process. The work does not vanish; the shape of it changes.
What stops it going wrong
Autonomy is only safe when it is observable
An agent you cannot inspect is a liability regardless of how well it performs on a good day. Every build we put into production carries the same five controls, and they are the part of the work clients notice least and benefit from most.
Control 01
Grounded in your data
Answers are drawn from your own records and documents, with the source attached. When the information is not there, the correct output is to say so rather than to invent something plausible.
Control 02
Scoped access
Each agent gets the narrowest set of permissions that lets it finish its job. It cannot reach systems it was never granted, and credentials are never shared between tasks.
Control 03
A gate on consequence
Anything that spends money, contacts a customer, or cannot be undone waits for a person. Reversible steps run freely, which is what keeps the gate rare enough to respect.
Control 04
A readable log
Every run records what was asked, what it did, what it saw and what it produced. When something looks wrong you can follow the reasoning rather than guess at it.
Control 05
Tested against real cases
Performance is measured on a set of your own past examples with known right answers, before launch and continuously after. Quality is a number you can watch, not a feeling.
And
A stop button
Any agent can be paused instantly without taking down the process around it. The work falls back to the people who did it before, which is why nothing is ever fully staked on the agent.
What the team sees
Agents you can manage like staff
Running agents without reporting is how the savings quietly disappear. The console shows what ran, what it cost, where it stopped and whether quality is holding, so the decision to give a task more autonomy is made on evidence.
Operations view
Illustrative sample. Every figure here is generated for the example.
An agent that nobody is measuring is not an efficiency. It is an unsupervised employee with your credentials.
Where we start
One process, proven, then the next
We do not begin with a platform. We begin with a single process that is costing real hours, put an agent on it at the lowest sensible level of autonomy, and measure it against how the work was done before. If it does not hold up, you have lost weeks rather than a year.
Step one
Find the repetitive hours
A short look at where time goes and which of it is frequent, well defined and reversible. You keep the shortlist whether or not we build from it.
Step two
Build one, at draft level
The first agent produces finished work for a person to approve. Low risk, immediately useful, and it generates the run history that everything after it depends on.
Step three
Earn more autonomy
Once the logs show it is right consistently, the gate moves. The task steps up a rung because the evidence supports it, not because the technology allows it.
This page describes our approach at a high level. Architecture, model selection, evaluation design and commercial terms are covered in a scoped proposal.
About GMT
We build the systems underneath the work.
Social media, SaaS and HR, and the AI agents that run on top. One idea under all of it: the record belongs to you.
Social media rents you an audience. HR software rents you a slice of your people. We build the one thing you keep: a record you own, joined across every system, with AI on top that shows its evidence and asks before it acts.
2
Practices
Social Media for creators, founders and brands. SaaS / HR for people and operations leaders.
46
Explainer reels
Data-led films that teach the problem first, with every source on screen.
15
Languages
Supported by the people platform, so one system serves a team across borders.
What we do
Four ways in. One engine underneath.
Every engagement sits on the same foundation: one database you own, AI that waits for a yes, and reporting your team can run without us.
Social Media
SaaS / HR
