A working environment,
loaded with your own material.
For the next two weeks you have the Machine running on Abbott's Similac and Love Without Measure files. The job is to use it on real work and tell us where it fits how you work and where it does not.
The September 2 session opened on the problem of fragmented pilots across the marketing value chain. This trial is the first attempt to string them together in one place. What am I allowed to say, who is this for, what did we learn last time, write me the thing. All of it answered from Abbott's own files.
Your contact
Tom Magnifico is your main contact throughout. tom.magnifico@codeandtheory.com. Reach out with anything, including "I could not get in" and "this is not how we work".
What happens when.
| When | What |
|---|---|
| Now | Access is open. Sign in today so a login problem surfaces early. |
| First three days | Run the 20-minute path once, end to end. |
| Friday 11 September, 11am EST | Open Q&A. Bring what did not work. |
| Week two | Run your own flows and queries, shaped by what comes out of the Q&A. |
Signing in.
- Open abbott-machine.demo.stagwell-machine.com
- Enter your work email. Use the address your Code and Theory contact added for you.
- Look for an email titled "Your sign-in code for The Machine", from The Machine (support@stagwell-machine.com). Enter the code on the sign-in screen.
- If the code does not arrive within a few minutes, check spam and try step 2 once more. If it still does not arrive, email Tom and we will get you a code another way. Do not spend more than five minutes on this.
- If a code has stopped working, go back to step 1 and enter your email again for a fresh one.
What this sandbox is, and what it is not.
Everything here is a choice we made for a two-week sandbox. Reading it first will save you looking for things that are deliberately absent.
The browser is the only surface
There is no Word, Excel or Photoshop add-in in the trial. The add-ins are built and they work. Putting them on Abbott laptops is a Windows deployment that Kevin Kaiser's team runs on its own track, and we kept it out of the trial window so nobody has to wait on it. Everything in this guide runs at the address above.
The material is a fixed set of Abbott's own files
The Machine has been trained on the Similac Love Without Measure material, and that is what every answer is drawn from. It is the pack Abbott sent us. Brand books and Vis ID, creative principles, PDP and PIM product content, prior approved copy, the Sisterhood of Motherhood creative brief, audience segmentation, the Pure Bliss claims prioritisation research, the 2026 Simple launch plans and historical tactical recommendations. You can see the list yourself: open Admin, go to Similac, and the Data and Knowledge tabs show each file by its real name, marked Indexed.
The claims library is a snapshot
It holds Abbott's approved claim records with their statements, footnotes, conditions of use and lifecycle state. It is a point-in-time copy of Abbott's Veeva export, not a live connection.
Claims matching is strict by default
The claims check looks for an exact match. Casing and punctuation count, and there is no fuzzy matching. Copy that sits close to an approved wording without reproducing it exactly will usually come back as a close but not exact match, naming the record it is nearest to. In other cases it will simply report no match found. That strictness is deliberate for this environment. In a real deployment the tolerance is configurable, so if it feels too tight for the way your team writes, that is worth telling us.
The media plan is loaded as data
Abbott's 2026 Similac plan, 26,783 lines carrying spend, impressions and clicks against channel, market, funnel stage and month. Spend, impressions and clicks are where it stops. There are no conversions and no revenue, because the plan does not carry them.
You cannot upload your own material yet
We are holding uploads until everyone is properly provisioned, for security posture reasons. The Admin panel shows a drop zone, and a file dragged into it will fail. Reading what was loaded is the capability here. Adding to it comes later.
Two things shown on September 2 were illustrations
The AEM Sites page assembly and the Adobe Express edit were mockups of where the work goes next. They are not in this environment.
The environment is scoped to one brand and one campaign
Similac, and Love Without Measure under it. Every question is answered against that material, and there is no option to work on something the environment holds nothing for.
If you do one thing, do this.
It runs end to end: what am I cleared to say, through a brief built from Abbott's own material, to what the plan and the history say.
You land on a dashboard that greets you by name, with a message box and a feed of cards under it.
Is the line "No artificial growth hormones" cleared against the approved claims library? Name the claim records you matched.
Not cleared. The answer names the claim records it matched,
states what state each one is in, and tells you the strictest state governs where records disagree. On the line above
it returns two records, CLM-005091 and CLM-005092, and reports both as rejected. The exact records can differ, so
what to check is the shape: a verdict, named records, and a state for each.
It will also hand you a footnote and describe it as required. Ignore that on a line that is not cleared. See the
rough edges above.
This is the beat to judge the whole environment on. Naming the records and the state each one is in is different from a flat "not found", which is what most tools give you whether the language was never approved or was pulled later.
Is "Our closest formula to breast milk" cleared? Name the claim record.
Cleared, naming CLM-014519 as approved. Conditions of use on that record are marked reviewer-only, so the answer says conditions apply without printing them.
Is "Similac cures infant allergies in one week" cleared? Name the claim record.
No match in the library, said as an answer and not as a clearance. If instead you get a message about not finding "a piece of work", that is the routing behaviour from the rough edges above and not a claims verdict. Re-ask it as a question.
Scroll up to the answer from step 2 and click the tool receipt above it, the line that says what the Machine did.
A one-sentence account of each step with the references behind it.
A claims lookup opens on the claim ids it found. A document search opens on the file names it read and the passages
it used. A step that found nothing says so, and a turn that answered without searching says that too. This is how
you tell an answer drawn from Abbott's material apart from one the model wrote from its own knowledge.
The receipt is collapsed until you click it, and a collapsed one looks empty. Expanded, a document search reads like
"Found 20 passages in 7 files" and lists the file names underneath.
What am I allowed to say about Similac 360 Total Care?
The approved wordings, each with its claim id, its lifecycle state and its conditions of use printed line by line, which is where the mandatory footnotes live. Up to twenty-five come back with the full count stated, so you know whether you are seeing all of it.
Open Briefs in the left sidebar and start a new brief against Love Without Measure.
When you pick the project, a brief name is filled in for you. Click into the field and replace it with your own.
Twelve sections, and a first proposed answer citing around twenty sources from Abbott's own files by name. Expect the Love Without Measure brand book, the Similac brand book, the messaging slides, the 2026 Simple launch plans, the Sisterhood of Motherhood creative brief and the Pure Bliss claims research. Under each proposal you get Accept, Suggest a change, Write my own and Skip this question. Nothing is written into your brief that you did not accept.
On any section, click Suggest a change and ask for something specific.
Cut this to two sentences and drop the competitor names
Two sentences, competitor names gone, substance kept. Scroll down if the change box looks as though it did nothing. It opens below the source list.
What does Love Without Measure stand for, and what tone should we write it in?
An answer specific to Similac and to Love Without Measure, naming the Abbott documents it used. Expect the brand book, the creative principles, the messaging guidelines and the Simple launch plans.
The answer itself reads as prose without file names in it. The files are in the receipt above the answer: expect the Love Without Measure brand book, the Love Without Measure Creative Principles and the Similac brand book.
Generic infant-formula content, a receipt that names no file, or an offer to search the web for what your own brand stands for.
How much did we spend on Retail Search in National in January?
A dollar figure, sourced to the plan data with the sheet it came from named.
How did Love Without Measure perform?
Spend, impressions and clicks broken down by channel. Ask what the figure covers and it will tell you it covers the lines attributed to that campaign, with measurement, credits and retail media outside it.
Open Agents in the left sidebar, find Prior Learnings, and press Open chat. Use the agent for this step and not the ordinary message box. The agent is the thing that reads the recorded learnings and rules on them.
Here is my brief: a spring campaign, February to May, 180 local markets, most of the budget in local social, an influencer line, front-loaded start. What did earlier Similac campaigns teach us, and for each learning tell me whether it applies to this brief?
Each recorded learning quoted as it was written, with the date it was published, a verdict of Applies: Yes, No or Partly, the reasoning, and what to do about it in this campaign. Expect it to disagree with the brief. On the brief above it should say that spreading local social evenly across that many markets buys no depth anywhere, and tell you to concentrate into twenty to thirty markets and hold the rest at national weight.
Six things to ask it.
These are intents, with worked examples. A prompt that only works word for word is not worth much to you, so say it in your own words.
What am I allowed to say
What am I allowed to say about Similac 360 Total Care? What approved claims do we have about immune support? Which of our approved claims carry a mandatory footnote?
Names claim ids, states each one's lifecycle state, and prints the conditions of use in full.
Is this line cleared
Is this line cleared against the approved claims library: [paste your line] Is this headline cleared, and if not, what is the closest approved wording? Check this paragraph against the claims library and rank the problems worst first
Gives a verdict, names the records it matched and the state of each, and where your copy resembles a stored wording without reproducing it, says it resembles and names the record it sits nearest.
Who is this for
Who is Love Without Measure for, based on our own research? Build the audience for a Simple launch push and say what it needs Which of Abbott's audience segments does this brief actually speak to?
Works from the segmentation and prior briefs in the environment and names them. Abbott's own segments, Connected Naturals and Expert Seekers, should appear by name.
What did we learn last time
What did earlier Similac campaigns teach us? Which of those learnings apply to a February to May campaign, and which do not? Show me the numbers behind that recommendation
Separates what the history says from whether it applies here, and says no where the answer is no.
What did it cost, and what did it do
How much did we spend on Retail Search in National in January? How did Love Without Measure perform by channel? How did BRIDGE Generations compare?
Gives the figure and says where it stops.
Write me the thing
Draft an email subject line and body from this concept Write social copy for this brief and tell me what you could not source Walk me through a creative brief for a Simple launch push
Names the document behind every product claim. Where the material cannot support what a channel wants to say, it says so instead of writing it anyway. Where a line is over its channel's limit, it tells you the real length.
Make it say no.
These three are worth your time, because a tool that agrees with everything is summarising rather than advising.
- Ask for a claim the library does not hold. Is "Similac cures infant allergies in one week" cleared? Watch it report no match, name no source it did not retrieve, and avoid reading like approval.
- Ask for something Abbott's material does not cover. Pick a competitor's internal strategy, or a market the pack says nothing about. It should say it cannot answer and say where that answer would live.
- Ask for a number the data does not carry. What was the conversion rate on Retail Search? The plan holds spend, impressions and clicks. A confident conversion figure would be invented, and that is a finding.
Where to start, by what you do.
If you make the work
Bailey, Meredith, and anyone writing or briefing. Go straight to the 20-minute path and run it once end to end. Then take the prompt menu and use it on a piece of work you actually have in front of you this week. The thing we most want back from you is where the output was not usable and what you had to do to it.
If you review the work
Charlotte, Inna, Liz, and anyone in MLR or compliance. Two things here were built to be inspected, and they are the two to look at closely.
- The claims check. Step 2 of the path. Two behaviours here are worth your time. Where one wording sits in the library in more than one state, it resolves to the strictest one and says so. Where your copy sits near an approved wording without reproducing it, it says it resembles and names the record, so no approval is implied that legal never gave. Check both.
- The tool receipts. Step 3 of the path. Every step the Machine took, with the records behind it, in the conversation itself, one click under the answer. Judge whether that is the depth of trail you would need.
If you evaluate the wiring
Kevin, and anyone assessing this as something Abbott would run. You do not need to log in for this.
- Read the scope section and tell us which of those boundaries matter to Abbott, and which we chose wrongly.
- Read the open questions below and tell us which of them block an Abbott security review from starting.
- The add-in deployment to Windows is the one dependency the trial cannot route around, so anything we can do to make that cheaper for your team is worth a conversation now.
Telling us what you found.
Send it to tom.magnifico@codeandtheory.com. One place means we can see all of it and reply to all of it.
Four fields is enough
- What I tried. A sentence, or the prompt you used.
- What I expected.
- What I got.
- A screenshot.
What happens to it
Everything gets read and triaged. The Friday Q&A is where you see the response, including the honest "we are not going to do that" answers.
Questions from September 2.
Everything asked in the session that was deferred or half-answered, listed so nothing is lost. The Q&A on 11 September is where the open ones get closed.
| Question | Asked by | Where it stands |
|---|---|---|
| Confidence and rigour of sources: how do we know how much to trust an answer | Charlotte | Answered The Machine does not score its own confidence. Every answer cites the records and documents behind it, and the claims library names its snapshot date, so the check is reading the source itself. We are asking whether that is enough for you. |
| Role-based access: who can see whom | Multiple | Answered in part Abbott users here are Members. Code and Theory staff do not appear to you as assignees or reviewers, and the admin user list is not visible to a Member. Roles in a full deployment is open. |
| What an audit report looks like: the override table, what escalates, on what cadence | Inna, Charlotte | Open We are bringing a worked example to Friday so you can mark it up. |
| Image rights and expiry as a workflow gate | Inna | Open Not in this environment. Coming back to you Friday. |
| Model deprecation policy and token monitoring | Kevin | Open Taken offline in the session and still open. Friday. |
| PII handling | Multiple | Open Friday. |
| Tenancy: how Abbott's data is isolated | Multiple | Open The one we want your security team on early. Friday, and then a longer conversation. |
| Geo deployment | Multiple | Answered in session Worth having in writing. Friday. |
After the two weeks.
The Q&A, Friday 11 September, 11am EST
Bring what did not work. The most useful session is the one where we spend most of it on your findings.
Week two, on your own terms
Once the Q&A has settled what the tool does well and where it falls short, week two is yours. Run your own flows and your own queries against the work you actually have on. What you try in that second week tells us most about how this would really get used.
Beyond that
What is in front of you is a two-week sandbox on one brand and one campaign, running in a browser. A full deployment is a different shape: your own material, the add-ins on your laptops, connections to the systems you already use, and a security review before any of it. That conversation follows from these two weeks.
The Machine · Code and Theory · Abbott Trial Guide · September 2026 · Confidential