RWS Holdings plc (RWS) — Investment Research Note

Executive summary

RWS is a global language, content and IP-services group that has repositioned itself as an "AI solutions company" delivering data annotation (TrainAI), enterprise localisation, and patent/IP services to 80+ of the world's top 100 brands. The trajectory across the five-year period has been one of gradual decline followed by a sharp reset: revenues fell from £749m in FY22 to £690m in FY25, adjusted PBT halved from £135.7m to £60.4m, an £88m goodwill impairment was booked, the dividend was rebased by ~43%, and both the CEO (Jan 2025) and Chair transitioned. The single most important valuation point today is whether the H1 FY26 return to organic growth (+7% OCC, adj PBT +33%) marks the start of a durable turnaround led by TrainAI and Language Weaver Pro, or a temporary reprieve in a core localisation business being structurally disrupted by generative AI.

Fair value estimate

• Range: 120p – 160p per share → implied market cap £445m – £594m
• Methodology: forward P/E on adjusted EPS, cross-checked with EV/adjusted EBITDA
  • FY26 adj PBT expected £58-65m (H1 £24m, mgmt guides "similar H2 weighting to prior year" where H2 FY25 = £42m) 2026-06 half-year
  • FY26 adj EPS ~12-13p (H1 4.9p; H2 seasonally stronger)
  • Applied multiple: 10-13x forward earnings — a discount to peer LSP/services multiples reflecting AI substitution risk and dividend rebasing, but a premium to distressed multiples reflecting the AI-training optionality and net-cash-adjacent balance sheet
• Comparison: current market cap £357.6m at 111.9p → central fair value ~£520m (140p)
• Upside: c.+25% to midpoint; range spans +7% to +43%

Sector context

Sector classification is Industrial Goods and Services (ICB), but the business economics are closer to IT services / vertical software. Quality/growth/leverage profile is below-typical for the sector: revenue has declined for four consecutive years, gross margin is compressing due to mix shift into lower-margin TrainAI and Renewals (41.6% H1 FY26 vs 43.3% H1 FY25 2026-06 half-year), and the balance sheet — while investment-grade — carries £491m of goodwill against £749m of net assets after last year's impairment. Listed peers: Keywords Studios (private post-EQT LBO — but the closest read-across on AI disruption of language/content services), Learning Technologies Group (related party via chairman Andrew Brode), and Straker Translations (much smaller AIM peer).

Investment thesis (3 bullets)

1. TrainAI is a genuine AI-receiver line delivering real revenue. The Generate segment grew OCC revenue +52% in H1 FY26 to £99.4m (28% of group), driven by an additional programme with an existing tech client, a new global technology client, plus a further logo won for H2 2026-06 half-year. This is high-margin, premium data-annotation work (red-teaming, RLHF, model refinement) sold to hyperscaler-class buyers who need enterprise-grade privacy and validated domain data. Management raised AI-related revenue mix from 26% to 32% of group in one half. 2026-06 half-year
2. Valuation does not require the bull case to work. At 111.9p the stock trades on ~9x FY26 adj EPS on cautious assumptions, with net debt <0.5x EBITDA, £290m of liquidity, and a rebased-but-progressive dividend now covered ~1.7x 2025-12 final results; 2026-06 half-year. The FY25 £88m goodwill impairment cleared the accounting overhang, and the Board has explicitly stated the new dividend level is set to fund the AI roadmap and M&A while remaining prudent.
3. New management + efficiency programme creating operational leverage before a market recovery. CEO Ben Faes (ex-Google/AOL) is executing a technology-first pivot, launched Language Weaver Pro with Cohere in March 2026 (ranked #1 in 31 of 32 languages vs DeepL/Gemini in benchmarks) 2026-04 trading statement, acquired Obviously (£16.5m upfront, £2bn TAM expansion in AI-enabled IP protection) 2026-06 half-year, and the A&M-backed efficiency plan targets a 10% productivity improvement over 18 months from FY26 onwards 2025-12 final results.

Key risks (3 bullets)

1. Transform (58% of revenue) is structurally exposed to LLM substitution. Legacy localisation declined -4.6% OCC in H1 FY26, and management is explicitly "pivoting" this segment to "tech-first" — an admission that the traditional model is being disrupted. Slator data cited by mgmt shows LSIs contracting 3% in 2024 2025-06 half-year. If Transform decline accelerates faster than Generate can offset, the mix shift will continue to compress gross margins.
2. TrainAI revenue is concentrated and lumpy. H1 FY26's exceptional growth was driven by "an additional programme with an existing client" — an inherently unpredictable revenue line that resembles project consulting more than recurring SaaS. The Group has flagged in prior periods when a major tech client reduced volumes, and no one client >10% of group revenue disguises concentration in TrainAI specifically 2025-12 final results.
3. Execution risk on a strategy still being defined. New CEO joined Jan 2025, new CFO joined March 2026, Chair transition Dec 2025, dividend rebased Dec 2025, £22m of FY25 exceptional restructuring costs (with £13m more planned for FY26) 2025-12 final results. That is a lot of change in a business already navigating an industry disruption. FY23 results also flagged repeated headwinds and cost actions; a fourth year of guidance cuts would materially reset the equity story.

Operating leverage

RWS is a moderately leveraged operator, not a pure fixed-cost platform. Cost of sales is ~58% of revenue (H1 FY26: £210m COGS vs £360m revenue), meaning gross margin ~42%. Administrative expenses were £157m in H1, giving a fixed/semi-fixed overhead base of ~44% of revenue. Head-count (7,499 FTEs 2026-06 half-year) is the dominant cost — RWS employs c.1,600 in-house linguists plus 5,900 in overheads, sales and technology functions. Incremental TrainAI revenue is human-heavy (annotators/RLHF experts), so contribution margin on TrainAI upside is likely 40-50%, not 70%+. Language Weaver Pro (SaaS licence revenue with Cohere partnership) is where genuine software-style operating leverage sits — but SaaS is still <30% of total licence revenue. Management guides "mid-single digit revenue growth" delivering "moderate margin expansion" — suggestive of ~2-3x operating leverage, not the 5-10x that a pure software business would deliver on an upside revenue surprise. A 10-20% revenue beat above plan would plausibly add 30-50% to operating profit, not multiples of it.

Value-trap signals

• Four-year revenue decline (£749m FY22 → £690m FY25) with FY26 return-to-growth still guided at only mid-single digits
• Goodwill impairment £88m in FY25 (Language Services £65m, Regulated Industries £22m) — an admission the SDL acquisition thesis has partially failed 2025-12 final results
• Dividend rebased 43% in Dec 2025
• Multiple leadership changes in a 15-month period (CEO, CFO, Chair, SID, multiple presidents)
• Related-party transactions — chairman Andrew Brode has significant interest in Learning Technologies Group (small services, arm's-length)
• Legacy translation being cannibalised by LLMs — a structural, not cyclical, headwind

Earnings vs. expectations

• FY22: in line with expectations; adj PBT £135.7m
• FY23: guided lower at H1 (Apr 2023) — full-year adj PBT within revised range £127-136m consensus; delivered £120.1m — miss vs original expectations, in line with revised
• FY24: trading in line with expectations; delivered adj PBT £106.7m in line with FY23 guidance
• FY25: H1 (Apr 2025) — adj PBT guidance cut to £60-70m from prior consensus; delivered £60.4m — at the low end of guidance, meaningful miss vs prior consensus
• H1 FY26: adj PBT £24.0m vs £18.0m prior year — beat vs an admittedly reset base; full-year "in line with expectations"

Pattern: a sequence of guidance cuts across FY23-FY25 followed by a reset base and first genuine beat in H1 FY26. Not a "consistent beat" track record — this is a "reset and beat" pattern that needs another full year to confirm.

Conviction

Conviction: 3 (moderate)

Anchors: (1) clean statutory disclosure with detailed segment reporting; (2) balance sheet is straightforward and net debt is small; (3) multiple valuation approaches (P/E, EV/EBITDA, dividend yield support) all point to a fair value range within £150m of the midpoint.

Limits: (1) the strategic direction is genuinely uncertain — TrainAI could 2x from here, or it could be lumpy and disappoint; (2) core Transform (58% of revenue) is under structural pressure from LLMs and the pivot to a "tech-first" model has not yet demonstrated financial success; (3) leadership is new and the strategy is still being embedded, so the FY26 in-line beat is one data point, not a trend.

Overall score: 425 / 1000

Partial fit for the investor profile. Real AI-training exposure is present but concentrated in one segment (~28% of revenue); valuation is fair rather than compelling; operating leverage is moderate not high; downside protection is genuinely good (strong balance sheet, no covenant risk, big brand roster). The main knock is that Transform (58% of revenue) faces a genuine LLM-substitution risk that keeps this out of the top band.