Outside View · SOLVE Chemistry
Prepared for Dr Francys Fernandez and Dr Nathan Barrow · October 2026
Four things I found after Friday.
You told me these. The four findings that follow are what I found and worked out afterwards.
About four minutes. Three short questions, and each answer changes what the next finding says. Who’s answering?
Question 1 of 3
If you could only win one type of customer in 2027, which would it be?
Finding 01
Sanofi has published its own kinetics toolbox.
A Sanofi team has published SKAI (Sanofi Kinetic AI), a machine-learning toolbox that helps scientists determine reaction mechanisms from experimental data. It is set out in a peer-reviewed paper (Industrial & Engineering Chemistry Research, 2025) and was presented at the AIChE Annual Meeting the same year.
The code is public, but free use is limited to academic and non-profit research; anyone else has to contact Sanofi. So it isn’t a free alternative for your prospects. It shows a pharma team with kinetics expertise tackling your problem with its own tools.
Both fit kinetic models to experimental data. I found no published comparison of results, so I can’t tell you whether one gives better answers. I also found nothing saying whether Sanofi uses SKAI in its own development work.
A review of AI in chemical engineering (a preprint, not yet peer-reviewed) by 20 authors, from universities including MIT, Imperial and Manchester and from Linde, Baker Hughes and Schneider Electric, names SKAI as one of several tools that let researchers “more easily adopt ML into their own workflows for kinetic analysis.” It is not the only one: the review also cites Chemfit, developed by Jiscoot and colleagues.
In its summary table, the review lists this as a remaining barrier for this kind of kinetic discovery: “Data quality/quantity requirements for reliable kinetic discovery are often unmet in typical experimental datasets.”
That barrier is the gap your lab fills. It is also what Francys raised on the call: software users need the right support so nobody gives up on poor-quality data. Wherever a tool like SKAI exists, the customer still needs data good enough to feed it.
On the call, Francys said some big corporates already have teams of kineticists, though not many. SKAI is a public example of what such a team can build. For them the question is less “does this work?” than “why use this instead of our own methods?”
That changes what the first conversation needs to prove, and to whom.
You described many API manufacturers running older facilities and only now turning to digitalisation, and SKAI’s licence doesn’t cover their commercial use without Sanofi’s permission. The comparison that matters there is your approach against how they work today.
That makes the case clearer, but it also makes the explaining Francys described more important.
CDMOs are handed chemistry they didn’t develop, and SKAI’s licence doesn’t cover their commercial use without Sanofi’s permission. A toolbox like this also needs data and kinetics expertise; your lab supplies the first, your software the second.
Which suggests the clearest case may sit where in-house alternatives are weakest.
SKAI comes from pharma. The case you described for specialty chemical companies is different: margin on products they have made for decades, rather than speed to market.
They sit outside the 80% you expect from pharma, so it is worth deciding whether they are the first door or the second.
That is a reasonable place to be three months in. SKAI adds one more thing to weigh: which buyers have their own kinetics methods, and which don’t.
Big pharma teams with kineticists may weigh you against their own tools. CDMOs and API manufacturers mostly start from somewhere else. That one difference changes the message, the proof and the order.
Whoever you choose, the next hurdle is the same.
Question 2 of 3
In the AstraZeneca deal, where did most of the eleven months go?
Finding 02
Your customers publish what they’ll check.
Large pharma companies set out their security conditions for suppliers in public documents. Your own customer is one of them.
- “Utilise industry recognised frameworks to ensure controls adequately protect confidentiality, integrity, and availability of data and systems.”
- “You are expected to maintain and implement procedures to monitor and maintain the effectiveness of your cyber security programme.”
- “Before we contract with third parties, we assess them.”
- “Novartis or its nominated party may perform technical and/or other assessments including testing to evaluate security and resilience of Novartis Data and Novartis Environment.”
- “Third Party and its subcontractors shall maintain the security certifications specified in the Agreement and audit reports.”
- Penetration and security tests performed “periodically (at least annually)”, and standards “such as NIST 800 and/or ISO 27001 at minimum.”
These are the questions behind the IT and legal months Nathan described, and the due diligence he expects before anyone uploads confidential data. Your Offerings page already holds strong answers to some of them: sole IP, no training on client chemistry, software that doesn’t need to see the chemistry. I couldn’t find a security certification named there.
Then the first hold-up sits before security: a chemist who wants SOLVE needs a case that the budget holder can approve. But the security assessment is the next gate, and with the software it will be the higher one, because customers will be uploading confidential data.
Then this is where the months went, and the requirements above are what that review was working through. Every new customer will run some version of it, and the software will face a stricter one than the lab.
Then the months went on terms. Novartis’s requirements, for example, refer to security certifications “specified in the Agreement”, so the legal and security conversations can turn out to be the same conversation, held twice.
Then the cycle started late rather than running slowly. But once the right person is found, these assessments are what they will send next, so the second hold-up is already predictable.
That is useful in itself: nobody has measured it yet. Logging each stage of the next deals, with dates, will show where the months go and give the board a number to watch fall.
Answering these before they are asked is a go-to-market decision, not just an IT one. And the timing matters more than it looks.
Finding 03
For three corporate sign-ups by the end of 2027, you have about 14 weeks.
Using only your own numbers: AstraZeneca took eleven months from first conversation to purchase order, and software, by your account, will take longer. Move the slider to see what shortening the cycle does.
The arithmetic: close month = start month + (11 − months saved), with the close set at December 2027. It uses the lab-service cycle, leaves out the time it takes to find the lead, and assumes software is no slower. So the real dates are earlier, not later.
At today’s pace, every conversation that will become one of those three sign-ups needs to have started within about 14 weeks.
There is a company in a neighbouring field that has already shown how a second lab can follow the first.
Finding 04
Pfizer has a multi-year agreement with a self-driving-lab company.
Telescope Innovations, a chemical technology company serving the pharmaceutical and chemical industries, builds self-driving labs. In the last year it announced:
- December 2025 · Korea’s first self-driving lab for pharma R&D and education, with the Korean Pharmaceutical and Biopharmaceutical Manufacturers Association
- February 2026 · its second self-driving lab at Pfizer, under a multi-year agreement, described as a milestone in the move “from collaborative R&D to potential full-scale deployment”
- June 2026 · a deployment with a third “top-tier global pharmaceutical company”, at its European operations
Two things in that for SOLVE. A company as large as Pfizer has committed to a self-driving-lab company, in writing and for several years. And there a second lab followed the first, a step Telescope describes as a milestone towards “potential full-scale deployment”.
Your AstraZeneca purchase order has just arrived. Francys’ six-month goal is customers who come back.
How a first project becomes the second, and then an agreement, is something you can design rather than wait for.
And one for Nathan
You already publish where kineticists read.
On the call, Nathan asked whether SOLVE needs to publish in Royal Society of Chemistry journals to reach chemists and kineticists. Your own publications list answers it: five of the thirteen papers on your About page are in RSC journals, three in Reaction Chemistry & Engineering and two in Digital Discovery, alongside OPRD, ACS and NeurIPS.
The readers are already there. What isn’t yet clear is what happens after someone reads one.
Question 3 of 3
What should we go deeper on first?
Choose up to two.
Yours to keep
What I found, and what you told me.
- 01Sanofi has published its own kinetics toolbox. Where a prospect has kineticists, you may be weighed against their own methods, not another vendor.
- 02Your customers publish the security conditions they will check, including AstraZeneca.
- 03At an eleven-month cycle, three corporate sign-ups by the end of 2027 means first conversations by the end of January 2027.
- 04Pfizer has a multi-year agreement with a self-driving-lab company, and a second lab there followed the first.
Three things worth checking
- Which of the companies you are speaking to have kinetics or modelling teams of their own?
- What will a customer’s security assessment ask SOLVE for, and is a certification planned?
- What would it take for the AstraZeneca project to become a multi-year agreement?
One ask
A 45-minute call with you both to walk through your full go-to-market map: every part of it, what you can run yourselves, and where outside help would speed things up. Time and cost follow from that conversation.
Or send them to nik@sparkandmatter.com.
Sources
- Sanofi, SKAI repository on GitHub, its non-commercial licence, the paper in Industrial & Engineering Chemistry Research, and AIChE Annual Meeting 2025 abstract
- Baldea M. et al., “Atoms to Processes: The Role of Artificial Intelligence and Machine Learning in Chemical Engineering”, arXiv, 1 October 2026 (section 3.6 and table 1)
- AstraZeneca, Code of Conduct for Third Parties, June 2025
- Novartis, Additional Information Security Requirements
- SOLVE Chemistry, Offerings and FAQ and About and publications, read 20 October 2026
- Telescope Innovations, Korea’s first self-driving lab for pharma, 9 December 2025, second self-driving lab at Pfizer, 23 February 2026, and third deployment with a major global pharmaceutical company, 1 June 2026